class DPCTGAN(CTGANSynthesizer):
"""Differential Private Conditional Table GAN Synthesizer
This code adds Differential Privacy to CTGANSynthesizer from https://github.com/sdv-dev/CTGAN
"""
def __init__(self,
embedding_dim=128,
gen_dim=(256, 256),
dis_dim=(256, 256),
l2scale=1e-6,
batch_size=500,
epochs=300,
pack=1,
log_frequency=True,
disabled_dp=False,
target_delta=None,
sigma=5,
max_per_sample_grad_norm=1.0,
epsilon=1,
verbose=True,
loss='cross_entropy'):
# CTGAN model specific parameters
self.embedding_dim = embedding_dim
self.gen_dim = gen_dim
self.dis_dim = dis_dim
self.l2scale = l2scale
self.batch_size = batch_size
self.epochs = epochs
self.pack = pack
self.log_frequency = log_frequency
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
# opacus parameters
self.sigma = sigma
self.disabled_dp = disabled_dp
self.target_delta = target_delta
self.max_per_sample_grad_norm = max_per_sample_grad_norm
self.epsilon = epsilon
self.epsilon_list = []
self.alpha_list = []
self.loss_d_list = []
self.loss_g_list = []
self.verbose = verbose
self.loss = loss
if self.loss != "cross_entropy":
# Monkeypatches the _create_or_extend_grad_sample function when calling opacus
opacus.supported_layers_grad_samplers._create_or_extend_grad_sample = _custom_create_or_extend_grad_sample
def train(self,
data,
categorical_columns=None,
ordinal_columns=None,
update_epsilon=None):
if update_epsilon:
self.epsilon = update_epsilon
self.transformer = DataTransformer()
self.transformer.fit(data, discrete_columns=categorical_columns)
train_data = self.transformer.transform(data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, self.log_frequency)
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
discriminator = Discriminator(data_dim + self.cond_generator.n_opt,
self.dis_dim, self.loss,
self.pack).to(self.device)
optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
optimizerD = optim.Adam(discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
privacy_engine = opacus.PrivacyEngine(
discriminator,
batch_size=self.batch_size,
sample_size=train_data.shape[0],
alphas=[1 + x / 10.0 for x in range(1, 100)] + list(range(12, 64)),
noise_multiplier=self.sigma,
max_grad_norm=self.max_per_sample_grad_norm,
clip_per_layer=True)
if not self.disabled_dp:
privacy_engine.attach(optimizerD)
one = torch.tensor(1, dtype=torch.float).to(self.device)
mone = one * -1
REAL_LABEL = 1
FAKE_LABEL = 0
criterion = nn.BCELoss()
assert self.batch_size % 2 == 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
steps_per_epoch = len(train_data) // self.batch_size
for i in range(self.epochs):
for id_ in range(steps_per_epoch):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
optimizerD.zero_grad()
if self.loss == 'cross_entropy':
y_fake = discriminator(fake_cat)
# print ('y_fake is {}'.format(y_fake))
label_fake = torch.full(
(int(self.batch_size / self.pack), ),
FAKE_LABEL,
dtype=torch.float,
device=self.device)
# print ('label_fake is {}'.format(label_fake))
errD_fake = criterion(y_fake, label_fake)
errD_fake.backward()
optimizerD.step()
# train with real
label_true = torch.full(
(int(self.batch_size / self.pack), ),
REAL_LABEL,
dtype=torch.float,
device=self.device)
y_real = discriminator(real_cat)
errD_real = criterion(y_real, label_true)
errD_real.backward()
optimizerD.step()
loss_d = errD_real + errD_fake
else:
y_fake = discriminator(fake_cat)
mean_fake = torch.mean(y_fake)
mean_fake.backward(one)
y_real = discriminator(real_cat)
mean_real = torch.mean(y_real)
mean_real.backward(mone)
optimizerD.step()
loss_d = -(mean_real - mean_fake)
max_grad_norm = []
for p in discriminator.parameters():
param_norm = p.grad.data.norm(2).item()
max_grad_norm.append(param_norm)
#pen = calc_gradient_penalty(discriminator, real_cat, fake_cat, self.device)
#pen.backward(retain_graph=True)
#loss_d.backward()
#optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = discriminator(torch.cat([fakeact, c1], dim=1))
else:
y_fake = discriminator(fakeact)
#if condvec is None:
cross_entropy = 0
#else:
# cross_entropy = self._cond_loss(fake, c1, m1)
if self.loss == 'cross_entropy':
label_g = torch.full((int(self.batch_size / self.pack), ),
REAL_LABEL,
dtype=torch.float,
device=self.device)
#label_g = torch.full(int(self.batch_size/self.pack,),1,device=self.device)
loss_g = criterion(y_fake, label_g)
loss_g = loss_g + cross_entropy
else:
loss_g = -torch.mean(y_fake) + cross_entropy
optimizerG.zero_grad()
loss_g.backward()
optimizerG.step()
if not self.disabled_dp:
#if self.loss == 'cross_entropy':
# autograd_grad_sample.clear_backprops(discriminator)
#else:
for p in discriminator.parameters():
if hasattr(p, "grad_sample"):
del p.grad_sample
if self.target_delta is None:
self.target_delta = 1 / train_data.shape[0]
epsilon, best_alpha = optimizerD.privacy_engine.get_privacy_spent(
self.target_delta)
self.epsilon_list.append(epsilon)
self.alpha_list.append(best_alpha)
#if self.verbose:
if not self.disabled_dp:
if self.epsilon < epsilon:
break
self.loss_d_list.append(loss_d)
self.loss_g_list.append(loss_g)
if self.verbose:
print("Epoch %d, Loss G: %.4f, Loss D: %.4f" %
(i + 1, loss_g.detach().cpu(), loss_d.detach().cpu()),
flush=True)
print('epsilon is {e}, alpha is {a}'.format(e=epsilon,
a=best_alpha))
return self.loss_d_list, self.loss_g_list, self.epsilon_list, self.alpha_list
def generate(self, n):
self.generator.eval()
#output_info = self.transformer.output_info
steps = n // self.batch_size + 1
data = []
for i in range(steps):
mean = torch.zeros(self.batch_size, self.embedding_dim)
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
return self.transformer.inverse_transform(data, None)
class PATECTGAN(CTGANSynthesizer):
def __init__(self,
embedding_dim=128,
gen_dim=(256, 256),
dis_dim=(256, 256),
l2scale=1e-6,
epochs=300,
pack=1,
log_frequency=True,
disabled_dp=False,
target_delta=None,
sigma=5,
max_per_sample_grad_norm=1.0,
verbose=True,
loss='cross_entropy',
binary=False,
batch_size=500,
teacher_iters=5,
student_iters=5,
epsilon=1.0,
delta=1e-5):
# CTGAN model specifi3c parameters
self.embedding_dim = embedding_dim
self.gen_dim = gen_dim
self.dis_dim = dis_dim
self.l2scale = l2scale
self.batch_size = batch_size
self.epochs = epochs
self.pack = pack
self.log_frequency = log_frequency
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.verbose = verbose
self.loss = loss
self.binary = binary
self.batch_size = batch_size
self.teacher_iters = teacher_iters
self.student_iters = student_iters
self.epsilon = epsilon
self.delta = delta
self.pd_cols = None
self.pd_index = None
def train(self,
data,
categorical_columns=None,
ordinal_columns=None,
update_epsilon=None):
if update_epsilon:
self.epsilon = update_epsilon
# this is to make sure data has at least 1000 points, may need it become flexible
self.num_teachers = int(len(data) / 5000) + 1
self.transformer = DataTransformer()
self.transformer.fit(data, discrete_columns=categorical_columns)
data = self.transformer.transform(data)
data_partitions = np.array_split(data, self.num_teachers)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
data, self.transformer.output_info, self.log_frequency)
# create conditional generator for each teacher model
cond_generator = [
ConditionalGenerator(d, self.transformer.output_info,
self.log_frequency) for d in data_partitions
]
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
discriminator = Discriminator(data_dim + self.cond_generator.n_opt,
self.dis_dim, self.loss,
self.pack).to(self.device)
student_disc = discriminator
student_disc.apply(weights_init)
teacher_disc = [discriminator for i in range(self.num_teachers)]
for i in range(self.num_teachers):
teacher_disc[i].apply(weights_init)
optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
optimizerS = optim.Adam(student_disc.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
optimizerT = [
optim.Adam(teacher_disc[i].parameters(), lr=2e-4, betas=(0.5, 0.9))
for i in range(self.num_teachers)
]
criterion = nn.BCELoss()
noise_multiplier = 1e-3
alphas = torch.tensor([0.0 for i in range(100)])
l_list = 1 + torch.tensor(range(100))
eps = 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
REAL_LABEL = 1
FAKE_LABEL = 0
criterion = nn.BCELoss()
while eps < self.epsilon:
# train teacher discriminators
for t_2 in range(self.teacher_iters):
for i in range(self.num_teachers):
# print ('i is {}'.format(i))
partition_data = data_partitions[i]
data_sampler = Sampler(partition_data,
self.transformer.output_info)
fakez = torch.normal(mean, std=std)
condvec = cond_generator[i].sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(
self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
optimizerT[i].zero_grad()
if self.loss == 'cross_entropy':
y_fake = teacher_disc[i](fake_cat)
label_fake = torch.full(
(int(self.batch_size / self.pack), ),
FAKE_LABEL,
dtype=torch.float,
device=self.device)
errD_fake = criterion(y_fake, label_fake.float())
errD_fake.backward()
optimizerT[i].step()
# train with real
label_true = torch.full(
(int(self.batch_size / self.pack), ),
REAL_LABEL,
dtype=torch.float,
device=self.device)
y_real = teacher_disc[i](real_cat)
errD_real = criterion(y_real, label_true.float())
errD_real.backward()
optimizerT[i].step()
loss_d = errD_real + errD_fake
# print("Iterator is {i}, Loss D for teacher {n} is :{j}".format(i=t_2 + 1, n=i+1, j=loss_d.detach().cpu()))
# train student discriminator
for t_3 in range(self.student_iters):
data_sampler = Sampler(data, self.transformer.output_info)
fakez = torch.normal(mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
else:
fake_cat = fake
fake_data = fake_cat
predictions, votes = pate(fake_data, teacher_disc,
noise_multiplier)
output = student_disc(fake_data.detach())
# update moments accountant
alphas = alphas + moments_acc(self.num_teachers, votes,
noise_multiplier, l_list)
loss_s = criterion(output, predictions.float().to(self.device))
optimizerS.zero_grad()
loss_s.backward()
optimizerS.step()
# print ('iterator {i}, student discriminator loss is {j}'.format(i=t_3, j=loss_s))
# train generator
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = student_disc(torch.cat([fakeact, c1], dim=1))
else:
y_fake = student_disc(fakeact)
# if condvec is None:
cross_entropy = 0
#else:
# cross_entropy = self._cond_loss(fake, c1, m1)
if self.loss == 'cross_entropy':
label_g = torch.full((int(self.batch_size / self.pack), ),
REAL_LABEL,
dtype=torch.float,
device=self.device)
#label_g = torch.full(int(self.batch_size/self.pack,),1,device=self.device)
loss_g = criterion(y_fake, label_g.float())
loss_g = loss_g + cross_entropy
else:
loss_g = -torch.mean(y_fake) + cross_entropy
optimizerG.zero_grad()
loss_g.backward()
optimizerG.step()
print('generator is {}'.format(loss_g.detach().cpu()))
eps = min((alphas - math.log(self.delta)) / l_list)
def generate(self, n):
self.generator.eval()
steps = n // self.batch_size + 1
data = []
for i in range(steps):
mean = torch.zeros(self.batch_size, self.embedding_dim)
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
generated_data = self.transformer.inverse_transform(data, None)
return generated_data
class CTGAN(object):
"""Conditional Table GAN Synthesizer.
This is the core class of the CTGAN project, where the different components
are orchestrated together.
For more details about the process, please check the [Modeling Tabular data using
Conditional GAN](https://arxiv.org/abs/1907.00503) paper.
Args:
embedding_dim (int):
Size of the random sample passed to the Generator. Defaults to 128.
gen_dim (tuple or list of ints):
Size of the output samples for each one of the Residuals. A Residual Layer
will be created for each one of the values provided. Defaults to (256, 256).
dis_dim (tuple or list of ints):
Size of the output samples for each one of the Discriminator Layers. A Linear Layer
will be created for each one of the values provided. Defaults to (256, 256).
l2scale (float):
Wheight Decay for the Adam Optimizer. Defaults to 1e-6.
batch_size (int):
Number of data samples to process in each step.
"""
def __init__(self,
embedding_dim=128,
gen_dim=(256, 256),
dis_dim=(256, 256),
l2scale=1e-6,
batch_size=500,
epochs=100):
self.embedding_dim = embedding_dim
self.gen_dim = gen_dim
self.dis_dim = dis_dim
self.l2scale = l2scale
self.batch_size = batch_size
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.trained_epoches = 0
self.epochs = epochs
def _apply_activate(self, data):
data_t = []
st = 0
for item in self.transformer.output_info:
if item[1] == 'tanh':
ed = st + item[0]
data_t.append(torch.tanh(data[:, st:ed]))
st = ed
elif item[1] == 'softmax':
ed = st + item[0]
data_t.append(
functional.gumbel_softmax(data[:, st:ed], tau=0.2))
st = ed
else:
assert 0
return torch.cat(data_t, dim=1)
def _cond_loss(self, data, c, m):
loss = []
st = 0
st_c = 0
skip = False
for item in self.transformer.output_info:
if item[1] == 'tanh':
st += item[0]
skip = True
elif item[1] == 'softmax':
if skip:
skip = False
st += item[0]
continue
ed = st + item[0]
ed_c = st_c + item[0]
tmp = functional.cross_entropy(data[:, st:ed],
torch.argmax(c[:, st_c:ed_c],
dim=1),
reduction='none')
loss.append(tmp)
st = ed
st_c = ed_c
else:
assert 0
loss = torch.stack(loss, dim=1)
return (loss * m).sum() / data.size()[0]
def train(self,
train_data,
categorical_columns=tuple(),
epochs=-1,
log_frequency=True,
ordinal_columns=None,
update_epsilon=None,
verbose=False,
mlflow=False):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to init param.
log_frequency (boolean):
Whether to use log frequency of categorical levels in conditional
sampling. Defaults to ``True``.
"""
if (epochs == -1):
epochs = self.epochs
if not hasattr(self, "transformer"):
self.transformer = DataTransformer()
self.transformer.fit(train_data, categorical_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
if not hasattr(self, "cond_generator"):
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, log_frequency)
if not hasattr(self, "generator"):
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt,
self.dis_dim).to(self.device)
if not hasattr(self, "optimizerG"):
self.optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
if not hasattr(self, "optimizerD"):
self.optimizerD = optim.Adam(self.discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
assert self.batch_size % 2 == 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
steps_per_epoch = max(len(train_data) // self.batch_size, 1)
for i in range(epochs):
self.trained_epoches += 1
for id_ in range(steps_per_epoch):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = self.discriminator(fake_cat)
y_real = self.discriminator(real_cat)
pen = self.discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device)
loss_d = -(torch.mean(y_real) - torch.mean(y_fake))
self.optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
self.optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.discriminator(torch.cat([fakeact, c1],
dim=1))
else:
y_fake = self.discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
loss_g = -torch.mean(y_fake) + cross_entropy
self.optimizerG.zero_grad()
loss_g.backward()
self.optimizerG.step()
if (verbose):
if (i % 50 == 0):
print("Epoch %d, Loss G: %.4f, Loss D: %.4f" %
(self.trained_epoches - 1, loss_g.detach().cpu(),
loss_d.detach().cpu()),
flush=True)
if (mlflow):
import mlflow
mlflow.log_metric("loss_d",
float(loss_d.detach().cpu()),
step=self.trained_epoches - 1)
mlflow.log_metric("loss_g",
float(loss_g.detach().cpu()),
step=self.trained_epoches - 1)
def generate(self, n, condition_column=None, condition_value=None):
"""Sample data similar to the training data.
Args:
n (int):
Number of rows to sample.
Returns:
numpy.ndarray or pandas.DataFrame
"""
if condition_column is not None and condition_value is not None:
condition_info = self.transformer.covert_column_name_value_to_id(
condition_column, condition_value)
global_condition_vec = self.cond_generator.generate_cond_from_condition_column_info(
condition_info, self.batch_size)
else:
global_condition_vec = None
steps = n // self.batch_size + 1
data = []
for i in range(steps):
mean = torch.zeros(self.batch_size, self.embedding_dim)
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
if global_condition_vec is not None:
condvec = global_condition_vec.copy()
else:
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
return self.transformer.inverse_transform(data, None)
def save(self, path):
assert hasattr(self, "generator")
assert hasattr(self, "discriminator")
assert hasattr(self, "transformer")
# always save a cpu model.
device_bak = self.device
self.device = torch.device("cpu")
self.generator.to(self.device)
self.discriminator.to(self.device)
torch.save(self, path)
self.device = device_bak
self.generator.to(self.device)
self.discriminator.to(self.device)
@classmethod
def load(cls, path):
model = torch.load(path)
model.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
model.generator.to(model.device)
model.discriminator.to(model.device)
return model
def train(self,
data,
categorical_columns=None,
ordinal_columns=None,
update_epsilon=None):
if update_epsilon:
self.epsilon = update_epsilon
self.transformer = DataTransformer()
self.transformer.fit(data, discrete_columns=categorical_columns)
train_data = self.transformer.transform(data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, self.log_frequency)
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
discriminator = Discriminator(data_dim + self.cond_generator.n_opt,
self.dis_dim, self.loss,
self.pack).to(self.device)
optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
optimizerD = optim.Adam(discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
privacy_engine = opacus.PrivacyEngine(
discriminator,
batch_size=self.batch_size,
sample_size=train_data.shape[0],
alphas=[1 + x / 10.0 for x in range(1, 100)] + list(range(12, 64)),
noise_multiplier=self.sigma,
max_grad_norm=self.max_per_sample_grad_norm,
clip_per_layer=True)
if not self.disabled_dp:
privacy_engine.attach(optimizerD)
one = torch.tensor(1, dtype=torch.float).to(self.device)
mone = one * -1
REAL_LABEL = 1
FAKE_LABEL = 0
criterion = nn.BCELoss()
assert self.batch_size % 2 == 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
steps_per_epoch = len(train_data) // self.batch_size
for i in range(self.epochs):
for id_ in range(steps_per_epoch):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
optimizerD.zero_grad()
if self.loss == 'cross_entropy':
y_fake = discriminator(fake_cat)
# print ('y_fake is {}'.format(y_fake))
label_fake = torch.full(
(int(self.batch_size / self.pack), ),
FAKE_LABEL,
dtype=torch.float,
device=self.device)
# print ('label_fake is {}'.format(label_fake))
errD_fake = criterion(y_fake, label_fake)
errD_fake.backward()
optimizerD.step()
# train with real
label_true = torch.full(
(int(self.batch_size / self.pack), ),
REAL_LABEL,
dtype=torch.float,
device=self.device)
y_real = discriminator(real_cat)
errD_real = criterion(y_real, label_true)
errD_real.backward()
optimizerD.step()
loss_d = errD_real + errD_fake
else:
y_fake = discriminator(fake_cat)
mean_fake = torch.mean(y_fake)
mean_fake.backward(one)
y_real = discriminator(real_cat)
mean_real = torch.mean(y_real)
mean_real.backward(mone)
optimizerD.step()
loss_d = -(mean_real - mean_fake)
max_grad_norm = []
for p in discriminator.parameters():
param_norm = p.grad.data.norm(2).item()
max_grad_norm.append(param_norm)
#pen = calc_gradient_penalty(discriminator, real_cat, fake_cat, self.device)
#pen.backward(retain_graph=True)
#loss_d.backward()
#optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = discriminator(torch.cat([fakeact, c1], dim=1))
else:
y_fake = discriminator(fakeact)
#if condvec is None:
cross_entropy = 0
#else:
# cross_entropy = self._cond_loss(fake, c1, m1)
if self.loss == 'cross_entropy':
label_g = torch.full((int(self.batch_size / self.pack), ),
REAL_LABEL,
dtype=torch.float,
device=self.device)
#label_g = torch.full(int(self.batch_size/self.pack,),1,device=self.device)
loss_g = criterion(y_fake, label_g)
loss_g = loss_g + cross_entropy
else:
loss_g = -torch.mean(y_fake) + cross_entropy
optimizerG.zero_grad()
loss_g.backward()
optimizerG.step()
if not self.disabled_dp:
#if self.loss == 'cross_entropy':
# autograd_grad_sample.clear_backprops(discriminator)
#else:
for p in discriminator.parameters():
if hasattr(p, "grad_sample"):
del p.grad_sample
if self.target_delta is None:
self.target_delta = 1 / train_data.shape[0]
epsilon, best_alpha = optimizerD.privacy_engine.get_privacy_spent(
self.target_delta)
self.epsilon_list.append(epsilon)
self.alpha_list.append(best_alpha)
#if self.verbose:
if not self.disabled_dp:
if self.epsilon < epsilon:
break
self.loss_d_list.append(loss_d)
self.loss_g_list.append(loss_g)
if self.verbose:
print("Epoch %d, Loss G: %.4f, Loss D: %.4f" %
(i + 1, loss_g.detach().cpu(), loss_d.detach().cpu()),
flush=True)
print('epsilon is {e}, alpha is {a}'.format(e=epsilon,
a=best_alpha))
return self.loss_d_list, self.loss_g_list, self.epsilon_list, self.alpha_list
class PATECTGAN(CTGANSynthesizer):
def __init__(
self,
embedding_dim=128,
gen_dim=(256, 256),
dis_dim=(256, 256),
l2scale=1e-6,
epochs=300,
pack=1,
log_frequency=True,
disabled_dp=False,
target_delta=None,
sigma=5,
max_per_sample_grad_norm=1.0,
verbose=False,
loss="cross_entropy", # losses supported: 'cross_entropy', 'wasserstein'
regularization=None, # regularizations supported: 'dragan'
binary=False,
batch_size=500,
teacher_iters=5,
student_iters=5,
sample_per_teacher=1000,
epsilon=8.0,
delta=1e-5,
noise_multiplier=1e-3,
moments_order=100,
):
# CTGAN model specifi3c parameters
self.embedding_dim = embedding_dim
self.gen_dim = gen_dim
self.dis_dim = dis_dim
self.l2scale = l2scale
self.batch_size = batch_size
self.epochs = epochs
self.pack = pack
self.log_frequency = log_frequency
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.verbose = verbose
self.loss = loss
self.regularization = regularization if self.loss != "wasserstein" else "dragan"
self.sample_per_teacher = sample_per_teacher
self.noise_multiplier = noise_multiplier
self.moments_order = moments_order
self.binary = binary
self.batch_size = batch_size
self.teacher_iters = teacher_iters
self.student_iters = student_iters
self.epsilon = epsilon
self.delta = delta
self.pd_cols = None
self.pd_index = None
def train(self,
data,
categorical_columns=None,
ordinal_columns=None,
update_epsilon=None):
if update_epsilon:
self.epsilon = update_epsilon
sample_per_teacher = (self.sample_per_teacher
if self.sample_per_teacher < len(data) else 1000)
self.num_teachers = int(len(data) / sample_per_teacher) + 1
self.transformer = DataTransformer()
self.transformer.fit(data, discrete_columns=categorical_columns)
data = self.transformer.transform(data)
data_partitions = np.array_split(data, self.num_teachers)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
data, self.transformer.output_info, self.log_frequency)
# create conditional generator for each teacher model
cond_generator = [
ConditionalGenerator(d, self.transformer.output_info,
self.log_frequency) for d in data_partitions
]
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
discriminator = Discriminator(data_dim + self.cond_generator.n_opt,
self.dis_dim, self.loss,
self.pack).to(self.device)
student_disc = discriminator
student_disc.apply(weights_init)
teacher_disc = [discriminator for i in range(self.num_teachers)]
for i in range(self.num_teachers):
teacher_disc[i].apply(weights_init)
optimizer_g = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
optimizer_s = optim.Adam(student_disc.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
optimizer_t = [
optim.Adam(teacher_disc[i].parameters(), lr=2e-4, betas=(0.5, 0.9))
for i in range(self.num_teachers)
]
noise_multiplier = self.noise_multiplier
alphas = torch.tensor([0.0 for i in range(self.moments_order)],
device=self.device)
l_list = 1 + torch.tensor(range(self.moments_order),
device=self.device)
eps = 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
real_label = 1
fake_label = 0
criterion = nn.BCELoss() if (self.loss
== "cross_entropy") else self.w_loss
if self.verbose:
print("using loss {} and regularization {}".format(
self.loss, self.regularization))
while eps < self.epsilon:
# train teacher discriminators
for t_2 in range(self.teacher_iters):
for i in range(self.num_teachers):
partition_data = data_partitions[i]
data_sampler = Sampler(partition_data,
self.transformer.output_info)
fakez = torch.normal(mean, std=std).to(self.device)
condvec = cond_generator[i].sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype("float32")).to(
self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
optimizer_t[i].zero_grad()
y_all = torch.cat(
[teacher_disc[i](fake_cat), teacher_disc[i](real_cat)])
label_fake = torch.full(
(int(self.batch_size / self.pack), 1),
fake_label,
dtype=torch.float,
device=self.device,
)
label_true = torch.full(
(int(self.batch_size / self.pack), 1),
real_label,
dtype=torch.float,
device=self.device,
)
labels = torch.cat([label_fake, label_true])
error_d = criterion(y_all, labels)
error_d.backward()
if self.regularization == "dragan":
pen = teacher_disc[i].dragan_penalty(
real_cat, device=self.device)
pen.backward(retain_graph=True)
optimizer_t[i].step()
# train student discriminator
for t_3 in range(self.student_iters):
data_sampler = Sampler(data, self.transformer.output_info)
fakez = torch.normal(mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
else:
fake_cat = fake
fake_data = fake_cat
predictions, votes = pate(fake_data,
teacher_disc,
noise_multiplier,
device=self.device)
output = student_disc(fake_data.detach())
# update moments accountant
alphas = alphas + moments_acc(self.num_teachers,
votes,
noise_multiplier,
l_list,
device=self.device)
loss_s = criterion(output, predictions.float().to(self.device))
optimizer_s.zero_grad()
loss_s.backward()
if self.regularization == "dragan":
vals = torch.cat([predictions, fake_data], axis=1)
ordered = vals[vals[:, 0].sort()[1]]
data_list = torch.split(
ordered,
predictions.shape[0] - int(predictions.sum().item()))
synth_cat = torch.cat(data_list[1:], axis=0)[:, 1:]
pen = student_disc.dragan_penalty(synth_cat,
device=self.device)
pen.backward(retain_graph=True)
optimizer_s.step()
# print ('iterator {i}, student discriminator loss is {j}'.format(i=t_3, j=loss_s))
# train generator
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = student_disc(torch.cat([fakeact, c1], dim=1))
else:
y_fake = student_disc(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
if self.loss == "cross_entropy":
label_g = torch.full(
(int(self.batch_size / self.pack), 1),
real_label,
dtype=torch.float,
device=self.device,
)
loss_g = criterion(y_fake, label_g.float())
loss_g = loss_g + cross_entropy
else:
loss_g = -torch.mean(y_fake) + cross_entropy
optimizer_g.zero_grad()
loss_g.backward()
optimizer_g.step()
eps = min((alphas - math.log(self.delta)) / l_list)
if self.verbose:
print("eps: {:f} \t G: {:f} \t D: {:f}".format(
eps,
loss_g.detach().cpu(),
loss_s.detach().cpu()))
def w_loss(self, output, labels):
vals = torch.cat([labels, output], axis=1)
ordered = vals[vals[:, 0].sort()[1]]
data_list = torch.split(ordered,
labels.shape[0] - int(labels.sum().item()))
fake_score = data_list[0][:, 1]
true_score = torch.cat(data_list[1:], axis=0)[:, 1]
w_loss = -(torch.mean(true_score) - torch.mean(fake_score))
return w_loss
def generate(self, n):
self.generator.eval()
steps = n // self.batch_size + 1
data = []
for i in range(steps):
mean = torch.zeros(self.batch_size, self.embedding_dim)
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
generated_data = self.transformer.inverse_transform(data, None)
return generated_data
def train(self,
train_data,
categorical_columns=tuple(),
epochs=-1,
log_frequency=True,
ordinal_columns=None,
update_epsilon=None,
verbose=False,
mlflow=False):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to init param.
log_frequency (boolean):
Whether to use log frequency of categorical levels in conditional
sampling. Defaults to ``True``.
"""
if (epochs == -1):
epochs = self.epochs
if not hasattr(self, "transformer"):
self.transformer = DataTransformer()
self.transformer.fit(train_data, categorical_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
if not hasattr(self, "cond_generator"):
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, log_frequency)
if not hasattr(self, "generator"):
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt,
self.dis_dim).to(self.device)
if not hasattr(self, "optimizerG"):
self.optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
if not hasattr(self, "optimizerD"):
self.optimizerD = optim.Adam(self.discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
assert self.batch_size % 2 == 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
steps_per_epoch = max(len(train_data) // self.batch_size, 1)
for i in range(epochs):
self.trained_epoches += 1
for id_ in range(steps_per_epoch):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = self.discriminator(fake_cat)
y_real = self.discriminator(real_cat)
pen = self.discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device)
loss_d = -(torch.mean(y_real) - torch.mean(y_fake))
self.optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
self.optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.discriminator(torch.cat([fakeact, c1],
dim=1))
else:
y_fake = self.discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
loss_g = -torch.mean(y_fake) + cross_entropy
self.optimizerG.zero_grad()
loss_g.backward()
self.optimizerG.step()
if (verbose):
if (i % 50 == 0):
print("Epoch %d, Loss G: %.4f, Loss D: %.4f" %
(self.trained_epoches - 1, loss_g.detach().cpu(),
loss_d.detach().cpu()),
flush=True)
if (mlflow):
import mlflow
mlflow.log_metric("loss_d",
float(loss_d.detach().cpu()),
step=self.trained_epoches - 1)
mlflow.log_metric("loss_g",
float(loss_g.detach().cpu()),
step=self.trained_epoches - 1)
def fit(
self,
train_data,
discrete_columns=tuple(),
epochs=300,
verbose=True,
gen_lr=2e-4,
):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to 300.
"""
"""
self.confidence_level = confidence_level
loss_other_name = "loss_bb" if confidence_level != -1 else "loss_d"
history = {"loss_g": [], loss_other_name: []}
"""
# Eli: add Mode-specific Normalization
if not hasattr(self, "transformer"):
self.transformer = DataTransformer()
self.transformer.fit(train_data, discrete_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
if not hasattr(self, "cond_generator"):
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, self.log_frequency)
if not hasattr(self, "generator"):
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
#print(data_dim)
#print(self.cond_generator.n_opt)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt,
self.dis_dim).to(self.device)
"""
#after sample in fit gen_output is 120 not 80(cause gen allmost twice)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
24 + self.cond_generator.n_opt, self.dis_dim
).to(self.device)
"""
if not hasattr(self, "optimizerG"):
self.optimizerG = optim.Adam(
self.generator.parameters(),
lr=gen_lr,
betas=(0.5, 0.9),
weight_decay=self.l2scale,
)
if not hasattr(self, "optimizerD"):
self.optimizerD = optim.Adam(self.discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
assert self.batch_size % 2 == 0
# init mean to zero and std to one
#keep one spot to confidence level, which will be add after normal dis will
mean = torch.zeros(((self.batch_size * self.embedding_dim) - 1),
device=self.device)
std = mean + 1
# steps_per_epoch = max(len(train_data) // self.batch_size, 1)
steps_per_epoch = 10 # magic number decided with Gilad. feel free to change it
loss_other_name = "loss_bb" if self.confidence_levels != [] else "loss_d"
allhist = {"confidence_levels_history": []}
# Eli: start training loop
for current_conf_level in self.confidence_levels:
#need to change, if non confidence give, aka []) no loop will run
#so if no conf, need to jump over conf loop
history = {
"confidence_level": current_conf_level,
"loss_g": [],
loss_other_name: []
}
for i in range(epochs):
self.trained_epoches += 1
for id_ in range(steps_per_epoch):
#we examine confidence lelvel so no need parts which they does not show up
"""
if self.confidence_levels == []:
# discriminator loop
for n in range(self.discriminator_steps):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(
self.batch_size, col[perm], opt[perm]
)
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype("float32")).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = self.discriminator(fake_cat)
y_real = self.discriminator(real_cat)
pen = self.discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device
)
loss_d = -(torch.mean(y_real) - torch.mean(y_fake))
if self.confidence_levels == []: # without bb loss
self.optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
self.optimizerD.step()
"""
# we examine confidence lelvel so no need parts which they does not show up
#as above(no confidence and uses discriminator, no need)
fakez = torch.normal(mean=mean, std=std)
#added here the part of adding conf to sample
#not sure why we need this part in the code but debug shows it this usses this lines
#ask gilad eli
confi = current_conf_level.astype(
np.float32) # generator excpect float
conf = torch.tensor([confi]).to(
self.device
) # change conf to conf input that will sent!!
fakez = torch.cat([fakez, conf], dim=0)
fakez = torch.reshape(
fakez, (self.batch_size, self.embedding_dim))
#end added here the part of adding conf to sample
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.discriminator(
torch.cat([fakeact, c1], dim=1))
else:
y_fake = self.discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
if self.confidence_levels != []:
# generate `batch_size` samples
#samples of fit using yBB in its input
#apply generator twice
gen_out, gen_fakeacts = self.sample(
self.batch_size, current_conf_level)
"""
gen_out=fakeact.detach().cpu().numpy()
gen_out=self.transformer.inverse_transform(gen_out, None)
"""
loss_bb = self._calc_bb_confidence_loss(
gen_out, current_conf_level
) #send specific confidence to loss computation
#return conf bit vector input and bb_y_vec input bit
#send to discriminate to connect gradient again
y_fake = self.discriminator(gen_fakeacts)
#find mean like in original ctgan
#multiple by small number to get realy small value
y_fake_mean = torch.mean(y_fake) * 0.00000000000001
y_fake_mean = y_fake_mean.to(self.device)
#change to float because y_fake is float not double
loss_bb = loss_bb.float()
#add y_fake_mean to lossg like in origin ctgan
#plus loss_bb
loss_g = -y_fake_mean + loss_bb + cross_entropy
#loss_g = -y_fake_mean + cross_entropy
else: # original loss
loss_g = -torch.mean(y_fake) + cross_entropy
self.optimizerG.zero_grad()
loss_g.backward()
"""
print("gradients\n")
for p in self.generator.parameters():
print(p.grad)
"""
"""
for name, param in self.generator.named_parameters():
#if param.requires_grad:
print(name)
print(param.grad)
"""
self.optimizerG.step()
loss_g_val = loss_g.detach().cpu()
loss_other_val = locals()[loss_other_name].detach().cpu()
history["loss_g"].append(loss_g.item())
history[loss_other_name].append(loss_other_val.item())
if verbose:
print(
f"Epoch {self.trained_epoches}, Loss G: {loss_g_val}, {loss_other_name}: {loss_other_val}",
flush=True,
)
allhist["confidence_levels_history"].append(history)
return allhist
def fit(self,
train_data,
discrete_columns=tuple(),
epochs=300,
log_frequency=True):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to 300.
log_frequency (boolean):
Whether to use log frequency of categorical levels in conditional
sampling. Defaults to ``True``.
"""
self.discrete_columns = discrete_columns
# Fit OH encoder
self.oe = OneHotEncoder(sparse=False)
if discrete_columns:
if type(train_data) is np.ndarray:
self.oe.fit(train_data[:, discrete_columns])
self.cat_column_idxes = discrete_columns
else:
self.oe.fit(train_data[discrete_columns])
features = train_data.columns
self.cat_column_idxes = [
features.index(c) for c in discrete_columns
]
else:
self.cat_column_idxes = []
self.transformer = DataTransformer()
self.transformer.fit(train_data, discrete_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, log_frequency)
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
discriminator = Discriminator(data_dim + self.cond_generator.n_opt,
self.dis_dim).to(self.device)
optimizerG = optim.Adam(self.generator.parameters(), lr=self.gen_lr)
optimizerD = optim.Adam(
discriminator.parameters(),
lr=self.dis_lr,
)
assert self.batch_size % 2 == 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
steps_per_epoch = max(len(train_data) // self.batch_size, 1)
for i in tqdm(range(epochs)):
for id_ in range(steps_per_epoch):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = discriminator(fake_cat)
y_real = discriminator(real_cat)
pen = discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device)
loss_d = -(nanmean(y_real) - nanmean(y_fake))
optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = discriminator(torch.cat([fakeact, c1], dim=1))
else:
y_fake = discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
loss_g = -nanmean(y_fake) + cross_entropy
optimizerG.zero_grad()
loss_g.backward()
optimizerG.step()
if self.verbose:
print("Epoch %d, Loss G: %.4f, Loss D: %.4f" %
(i + 1, loss_g.detach().cpu(), loss_d.detach().cpu()),
flush=True)
self.discriminator = discriminator
class PATECTGAN(CTGANSynthesizer):
def __init__(
self,
embedding_dim=128,
gen_dim=(256, 256),
dis_dim=(256, 256),
l2scale=1e-6,
epochs=300,
pack=1,
log_frequency=True,
disabled_dp=False,
target_delta=None,
sigma=5,
max_per_sample_grad_norm=1.0,
loss='cross_entropy',
regularization=None, #regularizations supported: 'dragan'
binary=False,
batch_size=500,
teacher_iters=5,
student_iters=5,
sample_per_teacher=1000,
epsilon=8.0,
delta=1e-5,
noise_multiplier=1e-3,
moments_order=100):
# CTGAN model specifi3c parameters
self.embedding_dim = embedding_dim
self.gen_dim = gen_dim
self.dis_dim = dis_dim
self.l2scale = l2scale
self.batch_size = batch_size
self.epochs = epochs
self.pack = pack
self.log_frequency = log_frequency
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.loss = loss
self.regularization = regularization
self.sample_per_teacher = sample_per_teacher
self.noise_multiplier = noise_multiplier
self.moments_order = moments_order
self.binary = binary
self.batch_size = batch_size
self.teacher_iters = teacher_iters
self.student_iters = student_iters
self.epsilon = epsilon
self.delta = delta
self.pd_cols = None
self.pd_index = None
def train(self,
data,
categorical_columns=None,
ordinal_columns=None,
update_epsilon=None,
verbose=False,
mlflow=False):
if update_epsilon:
self.epsilon = update_epsilon
if not hasattr(self, "transformer"):
self.transformer = DataTransformer()
self.transformer.fit(data, discrete_columns=categorical_columns)
data = self.transformer.transform(data)
data_dim = self.transformer.output_dimensions
if not hasattr(self, "cond_generator"):
self.cond_generator = ConditionalGenerator(
data, self.transformer.output_info, self.log_frequency)
if not hasattr(self, "generator"):
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt, self.dis_dim, self.loss,
self.pack).to(self.device)
if not hasattr(self, "student_disc"):
self.student_disc = self.discriminator
self.student_disc.apply(weights_init)
if not hasattr(self, "teacher_disc"):
# this is to make sure each teacher has some samples
sample_per_teacher = self.sample_per_teacher if self.sample_per_teacher < len(
data) else 1000
self.num_teachers = int(len(data) / sample_per_teacher) + 1
self.data_partitions = np.array_split(data, self.num_teachers)
# create conditional generator for each teacher model
self.cond_generator_t = [
ConditionalGenerator(d, self.transformer.output_info,
self.log_frequency)
for d in self.data_partitions
]
self.teacher_disc = [
self.discriminator for i in range(self.num_teachers)
]
for i in range(self.num_teachers):
self.teacher_disc[i].apply(weights_init)
if not hasattr(self, "optimizerG"):
self.optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
if not hasattr(self, "optimizerS"):
self.optimizerS = optim.Adam(self.student_disc.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
if not hasattr(self, "optimizerT"):
self.optimizerT = [
optim.Adam(self.teacher_disc[i].parameters(),
lr=2e-4,
betas=(0.5, 0.9)) for i in range(self.num_teachers)
]
if not hasattr(self, "train_eps"):
self.alphas = torch.tensor(
[0.0 for i in range(self.moments_order)], device=self.device)
self.l_list = 1 + torch.tensor(range(self.moments_order),
device=self.device)
self.train_eps = 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
one = torch.tensor(1, dtype=torch.float).to(self.device)
mone = one * -1
REAL_LABEL = 1
FAKE_LABEL = 0
criterion = nn.BCELoss() if (self.loss
== "cross_entropy") else self.WLoss
if (verbose):
print("using loss {} and regularization {}".format(
self.loss, self.regularization))
epoch = 0
while self.train_eps < self.epsilon:
# train teacher discriminators
for t_2 in range(self.teacher_iters):
for i in range(self.num_teachers):
partition_data = self.data_partitions[i]
data_sampler = Sampler(partition_data,
self.transformer.output_info)
fakez = torch.normal(mean, std=std).to(self.device)
condvec = self.cond_generator_t[i].sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(
self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
self.optimizerT[i].zero_grad()
y_all = torch.cat([
self.teacher_disc[i](fake_cat),
self.teacher_disc[i](real_cat)
])
label_fake = torch.full(
(int(self.batch_size / self.pack), 1),
FAKE_LABEL,
dtype=torch.float,
device=self.device)
label_true = torch.full(
(int(self.batch_size / self.pack), 1),
REAL_LABEL,
dtype=torch.float,
device=self.device)
labels = torch.cat([label_fake, label_true])
errD = criterion(y_all, labels)
errD.backward()
if (self.regularization == 'dragan'):
pen = self.teacher_disc[i].dragan_penalty(
real_cat, device=self.device)
pen.backward(retain_graph=True)
self.optimizerT[i].step()
# train student discriminator
for t_3 in range(self.student_iters):
data_sampler = Sampler(data, self.transformer.output_info)
fakez = torch.normal(mean, std=std).to(self.device)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
else:
fake_cat = fake
fake_data = fake_cat
predictions, votes = pate(fake_data, self.teacher_disc,
self.noise_multiplier, self.device)
output = self.student_disc(fake_data.detach())
# update moments accountant
self.alphas = self.alphas + moments_acc(
self.num_teachers, votes, self.noise_multiplier,
self.l_list, self.device)
loss_s = criterion(output, predictions.float().to(self.device))
self.optimizerS.zero_grad()
loss_s.backward()
if (self.regularization == 'dragan'):
vals = torch.cat(
[predictions.float().to(self.device), fake_data],
axis=1)
ordered = vals[vals[:, 0].sort()[1]]
data_list = torch.split(
ordered,
predictions.shape[0] - int(predictions.sum().item()))
synth_cat = torch.cat(data_list[1:], axis=0)[:, 1:]
pen = self.student_disc.dragan_penalty(synth_cat,
device=self.device)
pen.backward(retain_graph=True)
self.optimizerS.step()
self.train_eps = min(
(self.alphas - math.log(self.delta)) / self.l_list)
#train generator
fakez = torch.normal(mean=mean, std=std).to(self.device)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.student_disc(torch.cat([fakeact, c1], dim=1))
else:
y_fake = self.student_disc(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
if self.loss == 'cross_entropy':
label_g = torch.full((int(self.batch_size / self.pack), 1),
REAL_LABEL,
dtype=torch.float,
device=self.device)
loss_g = criterion(y_fake, label_g.float())
loss_g = loss_g + cross_entropy
else:
loss_g = -torch.mean(y_fake) + cross_entropy
self.optimizerG.zero_grad()
loss_g.backward()
self.optimizerG.step()
if (verbose):
print('eps: {:f} \t G: {:f} \t D: {:f}'.format(
self.train_eps,
loss_g.detach().cpu(),
loss_s.detach().cpu()))
if (mlflow):
import mlflow
mlflow.log_metric("loss_g",
float(loss_g.detach().cpu()),
step=epoch)
mlflow.log_metric("loss_d",
float(loss_s.detach().cpu()),
step=epoch)
mlflow.log_metric("epsilon", float(self.train_eps), step=epoch)
epoch += 1
def WLoss(self, output, labels):
vals = torch.cat([labels, output], axis=1)
ordered = vals[vals[:, 0].sort()[1]]
data_list = torch.split(ordered,
labels.shape[0] - int(labels.sum().item()))
fake_score = data_list[0][:, 1]
true_score = torch.cat(data_list[1:], axis=0)[:, 1]
w_loss = -(torch.mean(true_score) - torch.mean(fake_score))
return w_loss
def generate(self, n):
self.generator.eval()
steps = n // self.batch_size + 1
data = []
for i in range(steps):
mean = torch.zeros(self.batch_size, self.embedding_dim)
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
generated_data = self.transformer.inverse_transform(data, None)
return generated_data
def save(self, path):
assert hasattr(self, "generator")
assert hasattr(self, "teacher_disc")
assert hasattr(self, "student_disc")
assert hasattr(self, "cond_generator")
assert hasattr(self, "cond_generator_t")
# always save a cpu model.
device_bak = self.device
self.device = torch.device("cpu")
self.generator.to(self.device)
for i in range(self.num_teachers):
self.teacher_disc[i].to(self.device)
self.student_disc.to(self.device)
torch.save(self, path)
self.device = device_bak
self.generator.to(self.device)
for i in range(self.num_teachers):
self.teacher_disc[i].to(self.device)
self.student_disc.to(self.device)
@classmethod
def load(cls, path):
model = torch.load(path)
model.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
model.generator.to(model.device)
model.student_disc.to(model.device)
for i in range(model.num_teachers):
model.teacher_disc[i].to(model.device)
return model
class CTGANSynthesizer(object):
"""Conditional Table GAN Synthesizer.
This is the core class of the CTGAN project, where the different components
are orchestrated together.
For more details about the process, please check the [Modeling Tabular data using
Conditional GAN](https://arxiv.org/abs/1907.00503) paper.
Args:
embedding_dim (int):
Size of the random sample passed to the Generator. Defaults to 128.
gen_dim (tuple or list of ints):
Size of the output samples for each one of the Residuals. A Residual Layer
will be created for each one of the values provided. Defaults to (256, 256).
dis_dim (tuple or list of ints):
Size of the output samples for each one of the Discriminator Layers. A Linear Layer
will be created for each one of the values provided. Defaults to (256, 256).
l2scale (float):
Weight Decay for the Adam Optimizer. Defaults to 1e-6.
batch_size (int):
Number of data samples to process in each step.
discriminator_steps (int):
Number of discriminator updates to do for each generator update.
From the WGAN paper: https://arxiv.org/abs/1701.07875. WGAN paper
default is 5. Default used is 1 to match original CTGAN implementation.
log_frequency (boolean):
Whether to use log frequency of categorical levels in conditional
sampling. Defaults to ``True``.
blackbox_model:
Model that implements fit, predict, predict_proba
"""
def __init__(
self,
embedding_dim=128,
gen_dim=(256, 256),
dis_dim=(256, 256),
l2scale=1e-6,
batch_size=500,
discriminator_steps=1,
log_frequency=True,
blackbox_model=None,
preprocessing_pipeline=None,
bb_loss="logloss",
confidence_levels=[], #default value if no confidence given
):
self.embedding_dim = embedding_dim
self.gen_dim = gen_dim
self.dis_dim = dis_dim
self.l2scale = l2scale
self.batch_size = batch_size
self.log_frequency = log_frequency
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.trained_epoches = 0
self.discriminator_steps = discriminator_steps
self.blackbox_model = blackbox_model
self.preprocessing_pipeline = preprocessing_pipeline
self.confidence_levels = confidence_levels #set here not in fit
self.bb_loss = bb_loss
#print("self.confidence_levels")
@staticmethod
def _gumbel_softmax(logits, tau=1, hard=False, eps=1e-10, dim=-1):
"""Deals with the instability of the gumbel_softmax for older versions of torch.
For more details about the issue:
https://drive.google.com/file/d/1AA5wPfZ1kquaRtVruCd6BiYZGcDeNxyP/view?usp=sharing
Args:
logits:
[…, num_features] unnormalized log probabilities
tau:
non-negative scalar temperature
hard:
if True, the returned samples will be discretized as one-hot vectors,
but will be differentiated as if it is the soft sample in autograd
dim (int):
a dimension along which softmax will be computed. Default: -1.
Returns:
Sampled tensor of same shape as logits from the Gumbel-Softmax distribution.
"""
if version.parse(torch.__version__) < version.parse("1.2.0"):
for i in range(10):
transformed = functional.gumbel_softmax(logits,
tau=tau,
hard=hard,
eps=eps,
dim=dim)
if not torch.isnan(transformed).any():
return transformed
raise ValueError("gumbel_softmax returning NaN.")
return functional.gumbel_softmax(logits,
tau=tau,
hard=hard,
eps=eps,
dim=dim)
def _apply_activate(self, data):
data_t = []
st = 0
for item in self.transformer.output_info:
if item[1] == "tanh":
ed = st + item[0]
data_t.append(torch.tanh(data[:, st:ed]))
st = ed
elif item[1] == "softmax":
ed = st + item[0]
transformed = self._gumbel_softmax(data[:, st:ed], tau=0.2)
data_t.append(transformed)
st = ed
else:
assert 0
return torch.cat(data_t, dim=1)
def _cond_loss(self, data, c, m):
loss = []
st = 0
st_c = 0
skip = False
for item in self.transformer.output_info:
if item[1] == "tanh":
st += item[0]
skip = True
elif item[1] == "softmax":
if skip:
skip = False
st += item[0]
continue
ed = st + item[0]
ed_c = st_c + item[0]
tmp = functional.cross_entropy(
data[:, st:ed],
torch.argmax(c[:, st_c:ed_c], dim=1),
reduction="none",
)
loss.append(tmp)
st = ed
st_c = ed_c
else:
assert 0
loss = torch.stack(loss, dim=1)
return (loss * m).sum() / data.size()[0]
def fit(
self,
train_data,
discrete_columns=tuple(),
epochs=300,
verbose=True,
gen_lr=2e-4,
):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to 300.
"""
"""
self.confidence_level = confidence_level
loss_other_name = "loss_bb" if confidence_level != -1 else "loss_d"
history = {"loss_g": [], loss_other_name: []}
"""
# Eli: add Mode-specific Normalization
if not hasattr(self, "transformer"):
self.transformer = DataTransformer()
self.transformer.fit(train_data, discrete_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
if not hasattr(self, "cond_generator"):
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, self.log_frequency)
if not hasattr(self, "generator"):
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
#print(data_dim)
#print(self.cond_generator.n_opt)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt,
self.dis_dim).to(self.device)
"""
#after sample in fit gen_output is 120 not 80(cause gen allmost twice)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
24 + self.cond_generator.n_opt, self.dis_dim
).to(self.device)
"""
if not hasattr(self, "optimizerG"):
self.optimizerG = optim.Adam(
self.generator.parameters(),
lr=gen_lr,
betas=(0.5, 0.9),
weight_decay=self.l2scale,
)
if not hasattr(self, "optimizerD"):
self.optimizerD = optim.Adam(self.discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
assert self.batch_size % 2 == 0
# init mean to zero and std to one
#keep one spot to confidence level, which will be add after normal dis will
mean = torch.zeros(((self.batch_size * self.embedding_dim) - 1),
device=self.device)
std = mean + 1
# steps_per_epoch = max(len(train_data) // self.batch_size, 1)
steps_per_epoch = 10 # magic number decided with Gilad. feel free to change it
loss_other_name = "loss_bb" if self.confidence_levels != [] else "loss_d"
allhist = {"confidence_levels_history": []}
# Eli: start training loop
for current_conf_level in self.confidence_levels:
#need to change, if non confidence give, aka []) no loop will run
#so if no conf, need to jump over conf loop
history = {
"confidence_level": current_conf_level,
"loss_g": [],
loss_other_name: []
}
for i in range(epochs):
self.trained_epoches += 1
for id_ in range(steps_per_epoch):
#we examine confidence lelvel so no need parts which they does not show up
"""
if self.confidence_levels == []:
# discriminator loop
for n in range(self.discriminator_steps):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(
self.batch_size, col[perm], opt[perm]
)
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype("float32")).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = self.discriminator(fake_cat)
y_real = self.discriminator(real_cat)
pen = self.discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device
)
loss_d = -(torch.mean(y_real) - torch.mean(y_fake))
if self.confidence_levels == []: # without bb loss
self.optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
self.optimizerD.step()
"""
# we examine confidence lelvel so no need parts which they does not show up
#as above(no confidence and uses discriminator, no need)
fakez = torch.normal(mean=mean, std=std)
#added here the part of adding conf to sample
#not sure why we need this part in the code but debug shows it this usses this lines
#ask gilad eli
confi = current_conf_level.astype(
np.float32) # generator excpect float
conf = torch.tensor([confi]).to(
self.device
) # change conf to conf input that will sent!!
fakez = torch.cat([fakez, conf], dim=0)
fakez = torch.reshape(
fakez, (self.batch_size, self.embedding_dim))
#end added here the part of adding conf to sample
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.discriminator(
torch.cat([fakeact, c1], dim=1))
else:
y_fake = self.discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
if self.confidence_levels != []:
# generate `batch_size` samples
#samples of fit using yBB in its input
#apply generator twice
gen_out, gen_fakeacts = self.sample(
self.batch_size, current_conf_level)
"""
gen_out=fakeact.detach().cpu().numpy()
gen_out=self.transformer.inverse_transform(gen_out, None)
"""
loss_bb = self._calc_bb_confidence_loss(
gen_out, current_conf_level
) #send specific confidence to loss computation
#return conf bit vector input and bb_y_vec input bit
#send to discriminate to connect gradient again
y_fake = self.discriminator(gen_fakeacts)
#find mean like in original ctgan
#multiple by small number to get realy small value
y_fake_mean = torch.mean(y_fake) * 0.00000000000001
y_fake_mean = y_fake_mean.to(self.device)
#change to float because y_fake is float not double
loss_bb = loss_bb.float()
#add y_fake_mean to lossg like in origin ctgan
#plus loss_bb
loss_g = -y_fake_mean + loss_bb + cross_entropy
#loss_g = -y_fake_mean + cross_entropy
else: # original loss
loss_g = -torch.mean(y_fake) + cross_entropy
self.optimizerG.zero_grad()
loss_g.backward()
"""
print("gradients\n")
for p in self.generator.parameters():
print(p.grad)
"""
"""
for name, param in self.generator.named_parameters():
#if param.requires_grad:
print(name)
print(param.grad)
"""
self.optimizerG.step()
loss_g_val = loss_g.detach().cpu()
loss_other_val = locals()[loss_other_name].detach().cpu()
history["loss_g"].append(loss_g.item())
history[loss_other_name].append(loss_other_val.item())
if verbose:
print(
f"Epoch {self.trained_epoches}, Loss G: {loss_g_val}, {loss_other_name}: {loss_other_val}",
flush=True,
)
allhist["confidence_levels_history"].append(history)
return allhist
#special sample for fit/train part
#extra bit added to input as y,apply gen twice
#in second time with ybb after sent gen_out if first apply generator
#first time with y zero bit second time with bb conf bit
#not relevant function for now
def fit_sample(self,
n,
confidence_level,
condition_column=None,
condition_value=None):
"""Sample data similar to the training data.
Choosing a condition_column and condition_value will increase the probability of the
discrete condition_value happening in the condition_column.
Args:
n (int):
Number of rows to sample.
condition_column (string):
Name of a discrete column.
condition_value (string):
Name of the category in the condition_column which we wish to increase the
probability of happening.
Returns:
numpy.ndarray or pandas.DataFrame
"""
if condition_column is not None and condition_value is not None:
condition_info = self.transformer.covert_column_name_value_to_id(
condition_column, condition_value)
global_condition_vec = (
self.cond_generator.generate_cond_from_condition_column_info(
condition_info, self.batch_size))
else:
global_condition_vec = None
steps = n // self.batch_size + 1
data = []
gen_input_layer = self.generator.seq[0].fc #get linearinpputlayer
#now first sample part need to be
#with out not gradients updates
#so for all first part we do:
with torch.no_grad():
for i in range(steps):
#check if it is okay decrease noise by one for adding conf
#add conf ass input to sample function, as number vector
#give one vector place to adding y to gen noise
mean = torch.zeros(self.batch_size, (self.embedding_dim - 2))
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
#fakez = torch.reshape(fakez,(-1,))
confidence_level = confidence_level.astype(
np.float32) #generator excpect float
#create C column vector of confidence level C
conf = torch.zeros(self.batch_size, 1) + confidence_level
conf = conf.to(
self.device) #change conf to conf input that will sent!!
#first sample will geet zero as y column vector
yzero = torch.zeros(self.batch_size, 1).to(self.device)
#adding y bit to fakez gen noise,generator input
fakez = torch.cat([fakez, conf, yzero], dim=1)
fakez = torch.reshape(fakez,
(self.batch_size, self.embedding_dim))
if global_condition_vec is not None:
condvec = global_condition_vec.copy()
else:
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
#conftens = torch.tensor([0.5]);
#fakez = torch.cat([fakez, conftens], dim=1)#check to delelte!!
#reset to zero weights of y_zero and his bias
gen_input_layer.weight[-1] = 0
#-1 is last line, where I considered yzero weights to be
#which contains all weights of last bit of input
gen_input_layer.bias[-1] = 0
#-1 is last bias,where I considered yzero weights to be
#in biases vector which contain the bias of last bit
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
# instead of return data
# we will use it here to give it to BB and get new conf y
gen_out = self.transformer.inverse_transform(data, None)
y_prob_for_y_zero = self.blackbox_model.predict_proba(gen_out)
y_conf_gen_for_y_zero = y_prob_for_y_zero[:, 0].astype(np.float32)
#now we use this y conf and put it as y bit instead 0 and sample normally
#######
#second part of samples with BB y conf bit
#######
if condition_column is not None and condition_value is not None:
condition_info = self.transformer.covert_column_name_value_to_id(
condition_column, condition_value)
global_condition_vec = (
self.cond_generator.generate_cond_from_condition_column_info(
condition_info, self.batch_size))
else:
global_condition_vec = None
steps = n // self.batch_size + 1
data = []
for i in range(steps):
#check if it is okay decrease noise by one for adding conf
#add conf ass input to sample function, as vector cloumn
#decrease by one more for y vec (-2 for y bit a conf level in input)
mean = torch.zeros(self.batch_size, (self.embedding_dim - 2))
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
#fakez = torch.reshape(fakez,(-1,))
confidence_level = confidence_level.astype(
np.float32) #generator excpect float
#conf = torch.tensor([confidence_level],requires_grad=True).to(self.device)#change conf to conf input that will sent!!
conf = torch.zeros(self.batch_size, 1) + confidence_level
#conf is target in BBCE loss function
#target ccant have grdient descent True for some reason
#conf.requires_grad = True
conf.to(self.device)
#fix y bb bit
#y_conf_gen_for_y_zero = y_conf_gen_for_y_zero.astype(np.float32)#generator
#y_BB_bit=torch.tensor([y_conf_gen_for_y_zero], requires_grad=True).to(self.device)
#y_BB_column = torch.zeros(self.batch_size,1) + y_conf_gen_for_y_zero
y_BB_column = torch.tensor([y_conf_gen_for_y_zero],
requires_grad=True).to(self.device)
#y_BB_column.requires_grad = True
#take Transpose because shape is 1*50 - need 50*1
y_BB_column = y_BB_column.T
y_BB_column.to(self.device)
#put y_bb bit in fakez gen noise ,generator input
fakez = torch.cat([fakez, conf, y_BB_column], dim=1)
fakez = torch.reshape(fakez, (self.batch_size, self.embedding_dim))
#fakez = fakez.astype(np.float32)
if global_condition_vec is not None:
condvec = global_condition_vec.copy()
else:
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
#conftens = torch.tensor([0.5]);
#fakez = torch.cat([fakez, conftens], dim=1)#check to delelte!!
#turn on input gradient
#fakez.requires_grad = True
#turn on inputs gradient
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
# return data
#return self.transformer.inverse_transform(data, None)
gen_out = self.transformer.inverse_transform(data, None)
#ret conf level vector and bb_y_vector
#check if return the below written or above written
return conf, y_BB_column
#normal sample for creating in the expirements(with no train)
def sample(self,
n,
confidence_level,
condition_column=None,
condition_value=None):
"""Sample data similar to the training data.
Choosing a condition_column and condition_value will increase the probability of the
discrete condition_value happening in the condition_column.
Args:
n (int):
Number of rows to sample.
condition_column (string):
Name of a discrete column.
condition_value (string):
Name of the category in the condition_column which we wish to increase the
probability of happening.
Returns:
numpy.ndarray or pandas.DataFrame
"""
if condition_column is not None and condition_value is not None:
condition_info = self.transformer.covert_column_name_value_to_id(
condition_column, condition_value)
global_condition_vec = (
self.cond_generator.generate_cond_from_condition_column_info(
condition_info, self.batch_size))
else:
global_condition_vec = None
steps = n // self.batch_size + 1
data = []
our_fackeacts = []
for i in range(steps):
#check if it is okay decrease noise by one for adding conf
#add conf ass input to sample function, as number vector column
mean = torch.zeros(self.batch_size, (self.embedding_dim - 1))
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
#fakez = torch.reshape(fakez,(-1,))
confidence_level = confidence_level.astype(
np.float32) #generator excpect float
#create C column vector of confidence level C
conf = torch.zeros(self.batch_size, 1) + confidence_level
conf = conf.to(
self.device) #change conf to conf input that will sent!!
fakez = torch.cat([fakez, conf], dim=1)
fakez = torch.reshape(fakez, (self.batch_size, self.embedding_dim))
if global_condition_vec is not None:
condvec = global_condition_vec.copy()
else:
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
our_fackeacts.append(fakeact)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
gen_fakeacts = torch.cat(our_fackeacts, 0)
return self.transformer.inverse_transform(data, None), gen_fakeacts
def save(self, path):
assert hasattr(self, "generator")
assert hasattr(self, "discriminator")
assert hasattr(self, "transformer")
# always save a cpu model.
device_bak = self.device
self.device = torch.device("cpu")
self.generator.to(self.device)
self.discriminator.to(self.device)
torch.save(self, path)
self.device = device_bak
self.generator.to(self.device)
self.discriminator.to(self.device)
@classmethod
def load(cls, path):
model = torch.load(path)
model.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
model.generator.to(model.device)
model.discriminator.to(model.device)
return model
def _calc_bb_confidence_loss(self, gen_out, conf_level):
#added specific conf level as input arguement
y_prob = self.blackbox_model.predict_proba(gen_out)
y_conf_gen = y_prob[:, 0] # confidence scores
# create vector with the same size of y_confidence filled with `confidence_level` values
if isinstance(conf_level, list):
conf = np.random.choice(conf_level)
else:
conf = conf_level
y_conf_wanted = np.full(len(y_conf_gen), conf)
# to tensor
y_conf_gen = torch.tensor(y_conf_gen,
requires_grad=True).to(self.device)
y_conf_wanted = torch.tensor(y_conf_wanted).to(self.device)
# loss
bb_loss = self._get_loss_by_name(self.bb_loss)
bb_loss_val = bb_loss(y_conf_gen, y_conf_wanted)
return bb_loss_val
@staticmethod
def _get_loss_by_name(loss_name):
if loss_name == "log":
return torch.nn.BCELoss()
elif loss_name == "l1":
return torch.nn.L1Loss()
elif loss_name == "l2":
return torch.nn.L1Loss()
elif loss_name == "focal":
return WeightedFocalLoss()
else:
raise ValueError(f"Unknown loss name '{loss_name}'")
def train(self,
data,
categorical_columns=None,
ordinal_columns=None,
update_epsilon=None,
verbose=False,
mlflow=False):
if update_epsilon:
self.epsilon = update_epsilon
if not hasattr(self, "transformer"):
self.transformer = DataTransformer()
self.transformer.fit(data, discrete_columns=categorical_columns)
data = self.transformer.transform(data)
data_dim = self.transformer.output_dimensions
if not hasattr(self, "cond_generator"):
self.cond_generator = ConditionalGenerator(
data, self.transformer.output_info, self.log_frequency)
if not hasattr(self, "generator"):
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt, self.dis_dim, self.loss,
self.pack).to(self.device)
if not hasattr(self, "student_disc"):
self.student_disc = self.discriminator
self.student_disc.apply(weights_init)
if not hasattr(self, "teacher_disc"):
# this is to make sure each teacher has some samples
sample_per_teacher = self.sample_per_teacher if self.sample_per_teacher < len(
data) else 1000
self.num_teachers = int(len(data) / sample_per_teacher) + 1
self.data_partitions = np.array_split(data, self.num_teachers)
# create conditional generator for each teacher model
self.cond_generator_t = [
ConditionalGenerator(d, self.transformer.output_info,
self.log_frequency)
for d in self.data_partitions
]
self.teacher_disc = [
self.discriminator for i in range(self.num_teachers)
]
for i in range(self.num_teachers):
self.teacher_disc[i].apply(weights_init)
if not hasattr(self, "optimizerG"):
self.optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
if not hasattr(self, "optimizerS"):
self.optimizerS = optim.Adam(self.student_disc.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
if not hasattr(self, "optimizerT"):
self.optimizerT = [
optim.Adam(self.teacher_disc[i].parameters(),
lr=2e-4,
betas=(0.5, 0.9)) for i in range(self.num_teachers)
]
if not hasattr(self, "train_eps"):
self.alphas = torch.tensor(
[0.0 for i in range(self.moments_order)], device=self.device)
self.l_list = 1 + torch.tensor(range(self.moments_order),
device=self.device)
self.train_eps = 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
one = torch.tensor(1, dtype=torch.float).to(self.device)
mone = one * -1
REAL_LABEL = 1
FAKE_LABEL = 0
criterion = nn.BCELoss() if (self.loss
== "cross_entropy") else self.WLoss
if (verbose):
print("using loss {} and regularization {}".format(
self.loss, self.regularization))
epoch = 0
while self.train_eps < self.epsilon:
# train teacher discriminators
for t_2 in range(self.teacher_iters):
for i in range(self.num_teachers):
partition_data = self.data_partitions[i]
data_sampler = Sampler(partition_data,
self.transformer.output_info)
fakez = torch.normal(mean, std=std).to(self.device)
condvec = self.cond_generator_t[i].sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(
self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
self.optimizerT[i].zero_grad()
y_all = torch.cat([
self.teacher_disc[i](fake_cat),
self.teacher_disc[i](real_cat)
])
label_fake = torch.full(
(int(self.batch_size / self.pack), 1),
FAKE_LABEL,
dtype=torch.float,
device=self.device)
label_true = torch.full(
(int(self.batch_size / self.pack), 1),
REAL_LABEL,
dtype=torch.float,
device=self.device)
labels = torch.cat([label_fake, label_true])
errD = criterion(y_all, labels)
errD.backward()
if (self.regularization == 'dragan'):
pen = self.teacher_disc[i].dragan_penalty(
real_cat, device=self.device)
pen.backward(retain_graph=True)
self.optimizerT[i].step()
# train student discriminator
for t_3 in range(self.student_iters):
data_sampler = Sampler(data, self.transformer.output_info)
fakez = torch.normal(mean, std=std).to(self.device)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
else:
fake_cat = fake
fake_data = fake_cat
predictions, votes = pate(fake_data, self.teacher_disc,
self.noise_multiplier, self.device)
output = self.student_disc(fake_data.detach())
# update moments accountant
self.alphas = self.alphas + moments_acc(
self.num_teachers, votes, self.noise_multiplier,
self.l_list, self.device)
loss_s = criterion(output, predictions.float().to(self.device))
self.optimizerS.zero_grad()
loss_s.backward()
if (self.regularization == 'dragan'):
vals = torch.cat(
[predictions.float().to(self.device), fake_data],
axis=1)
ordered = vals[vals[:, 0].sort()[1]]
data_list = torch.split(
ordered,
predictions.shape[0] - int(predictions.sum().item()))
synth_cat = torch.cat(data_list[1:], axis=0)[:, 1:]
pen = self.student_disc.dragan_penalty(synth_cat,
device=self.device)
pen.backward(retain_graph=True)
self.optimizerS.step()
self.train_eps = min(
(self.alphas - math.log(self.delta)) / self.l_list)
#train generator
fakez = torch.normal(mean=mean, std=std).to(self.device)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.student_disc(torch.cat([fakeact, c1], dim=1))
else:
y_fake = self.student_disc(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
if self.loss == 'cross_entropy':
label_g = torch.full((int(self.batch_size / self.pack), 1),
REAL_LABEL,
dtype=torch.float,
device=self.device)
loss_g = criterion(y_fake, label_g.float())
loss_g = loss_g + cross_entropy
else:
loss_g = -torch.mean(y_fake) + cross_entropy
self.optimizerG.zero_grad()
loss_g.backward()
self.optimizerG.step()
if (verbose):
print('eps: {:f} \t G: {:f} \t D: {:f}'.format(
self.train_eps,
loss_g.detach().cpu(),
loss_s.detach().cpu()))
if (mlflow):
import mlflow
mlflow.log_metric("loss_g",
float(loss_g.detach().cpu()),
step=epoch)
mlflow.log_metric("loss_d",
float(loss_s.detach().cpu()),
step=epoch)
mlflow.log_metric("epsilon", float(self.train_eps), step=epoch)
epoch += 1
def fit(
self,
train_data,
discrete_columns=tuple(),
epochs=300,
confidence_level=-1,
verbose=True,
gen_lr=2e-4,
):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to 300.
"""
self.confidence_level = confidence_level
loss_other_name = "loss_bb" if confidence_level != -1 else "loss_d"
history = {"loss_g": [], loss_other_name: []}
# Eli: add Mode-specific Normalization
if not hasattr(self, "transformer"):
self.transformer = DataTransformer()
self.transformer.fit(train_data, discrete_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
if not hasattr(self, "cond_generator"):
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, self.log_frequency)
if not hasattr(self, "generator"):
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt,
self.dis_dim).to(self.device)
if not hasattr(self, "optimizerG"):
self.optimizerG = optim.Adam(
self.generator.parameters(),
lr=gen_lr,
betas=(0.5, 0.9),
weight_decay=self.l2scale,
)
if not hasattr(self, "optimizerD"):
self.optimizerD = optim.Adam(self.discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
assert self.batch_size % 2 == 0
# init mean to zero and std to one
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
# steps_per_epoch = max(len(train_data) // self.batch_size, 1)
steps_per_epoch = 10 # magic number decided with Gilad. feel free to change it
# Eli: start training loop
for i in range(epochs):
self.trained_epoches += 1
for id_ in range(steps_per_epoch):
if self.confidence_level == -1:
# discriminator loop
for n in range(self.discriminator_steps):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col,
opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size,
col[perm], opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype("float32")).to(
self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = self.discriminator(fake_cat)
y_real = self.discriminator(real_cat)
pen = self.discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device)
loss_d = -(torch.mean(y_real) - torch.mean(y_fake))
if self.confidence_level == -1: # without bb loss
self.optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
self.optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.discriminator(torch.cat([fakeact, c1],
dim=1))
else:
y_fake = self.discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
if self.confidence_level != -1:
# generate `batch_size` samples
gen_out = self.sample(self.batch_size)
loss_bb = self._calc_bb_confidence_loss(gen_out)
loss_g = loss_bb + cross_entropy
else: # original loss
loss_g = -torch.mean(y_fake) + cross_entropy
self.optimizerG.zero_grad()
loss_g.backward()
self.optimizerG.step()
loss_g_val = loss_g.detach().cpu()
loss_other_val = locals()[loss_other_name].detach().cpu()
history["loss_g"].append(loss_g.item())
history[loss_other_name].append(loss_other_val.item())
if verbose:
print(
f"Epoch {self.trained_epoches}, Loss G: {loss_g_val}, {loss_other_name}: {loss_other_val}",
flush=True,
)
return history
class CTGANSynthesizer(object):
"""Conditional Table GAN Synthesizer.
This is the core class of the CTGAN project, where the different components
are orchestrated together.
For more details about the process, please check the [Modeling Tabular data using
Conditional GAN](https://arxiv.org/abs/1907.00503) paper.
Args:
embedding_dim (int):
Size of the random sample passed to the Generator. Defaults to 128.
gen_dim (tuple or list of ints):
Size of the output samples for each one of the Residuals. A Residual Layer
will be created for each one of the values provided. Defaults to (256, 256).
dis_dim (tuple or list of ints):
Size of the output samples for each one of the Discriminator Layers. A Linear Layer
will be created for each one of the values provided. Defaults to (256, 256).
l2scale (float):
Weight Decay for the Adam Optimizer. Defaults to 1e-6.
batch_size (int):
Number of data samples to process in each step.
discriminator_steps (int):
Number of discriminator updates to do for each generator update.
From the WGAN paper: https://arxiv.org/abs/1701.07875. WGAN paper
default is 5. Default used is 1 to match original CTGAN implementation.
log_frequency (boolean):
Whether to use log frequency of categorical levels in conditional
sampling. Defaults to ``True``.
blackbox_model:
Model that implements fit, predict, predict_proba
"""
def __init__(
self,
embedding_dim=128,
gen_dim=(256, 256),
dis_dim=(256, 256),
l2scale=1e-6,
batch_size=500,
discriminator_steps=1,
log_frequency=True,
blackbox_model=None,
preprocessing_pipeline=None,
bb_loss="logloss",
):
self.embedding_dim = embedding_dim
self.gen_dim = gen_dim
self.dis_dim = dis_dim
self.l2scale = l2scale
self.batch_size = batch_size
self.log_frequency = log_frequency
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.trained_epoches = 0
self.discriminator_steps = discriminator_steps
self.blackbox_model = blackbox_model
self.preprocessing_pipeline = preprocessing_pipeline
self.confidence_level = -1 # will set in fit
self.bb_loss = bb_loss
@staticmethod
def _gumbel_softmax(logits, tau=1, hard=False, eps=1e-10, dim=-1):
"""Deals with the instability of the gumbel_softmax for older versions of torch.
For more details about the issue:
https://drive.google.com/file/d/1AA5wPfZ1kquaRtVruCd6BiYZGcDeNxyP/view?usp=sharing
Args:
logits:
[…, num_features] unnormalized log probabilities
tau:
non-negative scalar temperature
hard:
if True, the returned samples will be discretized as one-hot vectors,
but will be differentiated as if it is the soft sample in autograd
dim (int):
a dimension along which softmax will be computed. Default: -1.
Returns:
Sampled tensor of same shape as logits from the Gumbel-Softmax distribution.
"""
if version.parse(torch.__version__) < version.parse("1.2.0"):
for i in range(10):
transformed = functional.gumbel_softmax(logits,
tau=tau,
hard=hard,
eps=eps,
dim=dim)
if not torch.isnan(transformed).any():
return transformed
raise ValueError("gumbel_softmax returning NaN.")
return functional.gumbel_softmax(logits,
tau=tau,
hard=hard,
eps=eps,
dim=dim)
def _apply_activate(self, data):
data_t = []
st = 0
for item in self.transformer.output_info:
if item[1] == "tanh":
ed = st + item[0]
data_t.append(torch.tanh(data[:, st:ed]))
st = ed
elif item[1] == "softmax":
ed = st + item[0]
transformed = self._gumbel_softmax(data[:, st:ed], tau=0.2)
data_t.append(transformed)
st = ed
else:
assert 0
return torch.cat(data_t, dim=1)
def _cond_loss(self, data, c, m):
loss = []
st = 0
st_c = 0
skip = False
for item in self.transformer.output_info:
if item[1] == "tanh":
st += item[0]
skip = True
elif item[1] == "softmax":
if skip:
skip = False
st += item[0]
continue
ed = st + item[0]
ed_c = st_c + item[0]
tmp = functional.cross_entropy(
data[:, st:ed],
torch.argmax(c[:, st_c:ed_c], dim=1),
reduction="none",
)
loss.append(tmp)
st = ed
st_c = ed_c
else:
assert 0
loss = torch.stack(loss, dim=1)
return (loss * m).sum() / data.size()[0]
def fit(
self,
train_data,
discrete_columns=tuple(),
epochs=300,
confidence_level=-1,
verbose=True,
gen_lr=2e-4,
):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to 300.
"""
self.confidence_level = confidence_level
loss_other_name = "loss_bb" if confidence_level != -1 else "loss_d"
history = {"loss_g": [], loss_other_name: []}
# Eli: add Mode-specific Normalization
if not hasattr(self, "transformer"):
self.transformer = DataTransformer()
self.transformer.fit(train_data, discrete_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
if not hasattr(self, "cond_generator"):
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, self.log_frequency)
if not hasattr(self, "generator"):
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
if not hasattr(self, "discriminator"):
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt,
self.dis_dim).to(self.device)
if not hasattr(self, "optimizerG"):
self.optimizerG = optim.Adam(
self.generator.parameters(),
lr=gen_lr,
betas=(0.5, 0.9),
weight_decay=self.l2scale,
)
if not hasattr(self, "optimizerD"):
self.optimizerD = optim.Adam(self.discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
assert self.batch_size % 2 == 0
# init mean to zero and std to one
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
# steps_per_epoch = max(len(train_data) // self.batch_size, 1)
steps_per_epoch = 10 # magic number decided with Gilad. feel free to change it
# Eli: start training loop
for i in range(epochs):
self.trained_epoches += 1
for id_ in range(steps_per_epoch):
if self.confidence_level == -1:
# discriminator loop
for n in range(self.discriminator_steps):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col,
opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size,
col[perm], opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype("float32")).to(
self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = self.discriminator(fake_cat)
y_real = self.discriminator(real_cat)
pen = self.discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device)
loss_d = -(torch.mean(y_real) - torch.mean(y_fake))
if self.confidence_level == -1: # without bb loss
self.optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
self.optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.discriminator(torch.cat([fakeact, c1],
dim=1))
else:
y_fake = self.discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
if self.confidence_level != -1:
# generate `batch_size` samples
gen_out = self.sample(self.batch_size)
loss_bb = self._calc_bb_confidence_loss(gen_out)
loss_g = loss_bb + cross_entropy
else: # original loss
loss_g = -torch.mean(y_fake) + cross_entropy
self.optimizerG.zero_grad()
loss_g.backward()
self.optimizerG.step()
loss_g_val = loss_g.detach().cpu()
loss_other_val = locals()[loss_other_name].detach().cpu()
history["loss_g"].append(loss_g.item())
history[loss_other_name].append(loss_other_val.item())
if verbose:
print(
f"Epoch {self.trained_epoches}, Loss G: {loss_g_val}, {loss_other_name}: {loss_other_val}",
flush=True,
)
return history
def sample(self, n, condition_column=None, condition_value=None):
"""Sample data similar to the training data.
Choosing a condition_column and condition_value will increase the probability of the
discrete condition_value happening in the condition_column.
Args:
n (int):
Number of rows to sample.
condition_column (string):
Name of a discrete column.
condition_value (string):
Name of the category in the condition_column which we wish to increase the
probability of happening.
Returns:
numpy.ndarray or pandas.DataFrame
"""
if condition_column is not None and condition_value is not None:
condition_info = self.transformer.covert_column_name_value_to_id(
condition_column, condition_value)
global_condition_vec = (
self.cond_generator.generate_cond_from_condition_column_info(
condition_info, self.batch_size))
else:
global_condition_vec = None
steps = n // self.batch_size + 1
data = []
for i in range(steps):
mean = torch.zeros(self.batch_size, self.embedding_dim)
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
if global_condition_vec is not None:
condvec = global_condition_vec.copy()
else:
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
# return data
return self.transformer.inverse_transform(data, None)
def save(self, path):
assert hasattr(self, "generator")
assert hasattr(self, "discriminator")
assert hasattr(self, "transformer")
# always save a cpu model.
device_bak = self.device
self.device = torch.device("cpu")
self.generator.to(self.device)
self.discriminator.to(self.device)
torch.save(self, path)
self.device = device_bak
self.generator.to(self.device)
self.discriminator.to(self.device)
@classmethod
def load(cls, path):
model = torch.load(path)
model.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
model.generator.to(model.device)
model.discriminator.to(model.device)
return model
def _calc_bb_confidence_loss(self, gen_out):
y_prob = self.blackbox_model.predict_proba(gen_out)
y_conf_gen = y_prob[:, 0] # confidence scores
# create vector with the same size of y_confidence filled with `confidence_level` values
if isinstance(self.confidence_level, list):
conf = np.random.choice(self.confidence_level)
else:
conf = self.confidence_level
y_conf_wanted = np.full(len(y_conf_gen), conf)
# to tensor
y_conf_gen = torch.tensor(y_conf_gen,
requires_grad=True).to(self.device)
y_conf_wanted = torch.tensor(y_conf_wanted).to(self.device)
# loss
bb_loss = self._get_loss_by_name(self.bb_loss)
bb_loss_val = bb_loss(y_conf_gen, y_conf_wanted)
return bb_loss_val
@staticmethod
def _get_loss_by_name(loss_name):
if loss_name == "log":
return torch.nn.BCELoss()
elif loss_name == "l1":
return torch.nn.L1Loss()
elif loss_name == "l2":
return torch.nn.L1Loss()
elif loss_name == "focal":
return WeightedFocalLoss()
else:
raise ValueError(f"Unknown loss name '{loss_name}'")
def fit(self,
train_data,
eval_interval,
eval,
discrete_columns=tuple(),
epochs=300,
log_frequency=True):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to 300.
log_frequency (boolean):
Whether to use log frequency of categorical levels in conditional
sampling. Defaults to ``True``.
"""
train = train_data
self.transformer = DataTransformer()
self.transformer.fit(train_data, discrete_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, log_frequency)
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt, self.dis_dim).to(self.device)
optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
optimizerD = optim.Adam(self.discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
assert self.batch_size % 2 == 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
# figure for plotting gradients; ax1 for generator and ax2 for discriminator
#fig, (ax1, ax2) = plt.subplots(1, 2)
steps_per_epoch = max(len(train_data) // self.batch_size, 1)
for i in range(epochs):
for id_ in range(steps_per_epoch):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = self.discriminator(fake_cat)
y_real = self.discriminator(real_cat)
pen = self.discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device)
loss_d = -(torch.mean(y_real) - torch.mean(y_fake))
optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
#plot gradients
#self.plot_grad_flow(self.discriminator.named_parameters(), ax2, 'Gradient flow for D')
optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.discriminator(torch.cat([fakeact, c1],
dim=1))
else:
y_fake = self.discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
loss_g = -torch.mean(y_fake) + cross_entropy
optimizerG.zero_grad()
loss_g.backward()
# plot gradients
#self.plot_grad_flow(self.generator.named_parameters(), ax1, 'Gradient flow for G')
optimizerG.step()
print("Epoch %d, Loss G: %.4f, Loss D: %.4f" %
(i + 1, loss_g.detach().cpu(), loss_d.detach().cpu()),
flush=True)
#check model results every x epochs
#fig2, ax3 = plt.subplots(1)
if eval:
stats_real = train[self.demand_column].describe()
stats_real_week = train.groupby('Weekday')[
self.demand_column].describe()
stats_real_month = train.groupby('Month')[
self.demand_column].describe()
if ((i + 1) % eval_interval == 0):
eval_sample = self.sample(1000)
sample = pd.DataFrame(eval_sample,
columns=eval_sample.columns)
#sample.loc[:, self.demand_column].hist(bins=50, alpha=0.4, label='fake')
#ax3.hist(pd.DataFrame(train, columns=train.columns).loc[:, self.demand_column], bins=50, alpha=0.4, label='real')
#fig2.legend()
#fig2.show()
print(
(sample[self.demand_column].describe() - stats_real) /
stats_real)
print(' ')
print(((sample.groupby('Weekday')[
self.demand_column].describe() - stats_real_week) /
stats_real_week).T)
print(' ')
print(((sample.groupby('Month')[
self.demand_column].describe() - stats_real_month) /
stats_real_month).T)
plt.show()
def fit(self,
train_data,
discrete_columns=(),
epochs=300,
log_frequency=True):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to 300.
log_frequency (boolean):
Whether to use log frequency of categorical levels in conditional
sampling. Defaults to ``True``.
"""
self.transformer = DataTransformer()
self.transformer.fit(train_data, discrete_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, log_frequency)
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
discriminator = Discriminator(data_dim + self.cond_generator.n_opt,
self.dis_dim).to(self.device)
optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
optimizerD = optim.Adam(discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
assert self.batch_size % 2 == 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
train_losses = []
early_stopping = EarlyStopping(patience=self.patience, verbose=False)
steps_per_epoch = max(len(train_data) // self.batch_size, 1)
for i in range(epochs):
for id_ in range(steps_per_epoch):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = discriminator(fake_cat)
y_real = discriminator(real_cat)
pen = discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device)
loss_d = -(torch.mean(y_real) - torch.mean(y_fake))
train_losses.append(loss_d.item())
optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = discriminator(torch.cat([fakeact, c1], dim=1))
else:
y_fake = discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
loss_g = -torch.mean(y_fake) + cross_entropy
train_losses.append(loss_g.item())
optimizerG.zero_grad()
loss_g.backward()
optimizerG.step()
early_stopping(np.average(train_losses))
if early_stopping.early_stop:
print("GAN: Early stopping after epochs {}".format(i))
break
train_losses = []
class CTGANSynthesizer(object):
"""Conditional Table GAN Synthesizer.
This is the core class of the CTGAN project, where the different components
are orchestrated together.
For more details about the process, please check the [Modeling Tabular data using
Conditional GAN](https://arxiv.org/abs/1907.00503) paper.
Args:
embedding_dim (int):
Size of the random sample passed to the Generator. Defaults to 128.
gen_dim (tuple or list of ints):
Size of the output samples for each one of the Residuals. A Resiudal Layer
will be created for each one of the values provided. Defaults to (256, 256).
dis_dim (tuple or list of ints):
Size of the output samples for each one of the Discriminator Layers. A Linear Layer
will be created for each one of the values provided. Defaults to (256, 256).
l2scale (float):
Wheight Decay for the Adam Optimizer. Defaults to 1e-6.
batch_size (int):
Number of data samples to process in each step.
"""
def __init__(self,
demand_col,
embedding_dim=128,
gen_dim=(256, 256),
dis_dim=(256, 256),
l2scale=1e-6,
batch_size=500):
self.embedding_dim = embedding_dim
self.gen_dim = gen_dim
self.dis_dim = dis_dim
self.l2scale = l2scale
self.batch_size = batch_size
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.demand_column = demand_col
def _apply_activate(self, data):
data_t = []
st = 0
for item in self.transformer.output_info:
if item[1] == 'tanh':
ed = st + item[0]
data_t.append(torch.tanh(data[:, st:ed]))
st = ed
elif item[1] == 'softmax':
ed = st + item[0]
data_t.append(
functional.gumbel_softmax(data[:, st:ed], tau=0.2))
st = ed
else:
assert 0
return torch.cat(data_t, dim=1)
def _cond_loss(self, data, c, m):
loss = []
st = 0
st_c = 0
skip = False
for item in self.transformer.output_info:
if item[1] == 'tanh':
st += item[0]
skip = True
elif item[1] == 'softmax':
if skip:
skip = False
st += item[0]
continue
ed = st + item[0]
ed_c = st_c + item[0]
tmp = functional.cross_entropy(data[:, st:ed],
torch.argmax(c[:, st_c:ed_c],
dim=1),
reduction='none')
loss.append(tmp)
st = ed
st_c = ed_c
else:
assert 0
loss = torch.stack(loss, dim=1)
return (loss * m).sum() / data.size()[0]
def plot_grad_flow(self, named_parameters, axis, plot_title):
'''Plots the gradients flowing through different layers in the net during training.
Can be used for checking for possible gradient vanishing / exploding problems.
Usage: Plug this function in Trainer class after loss.backwards() as
"plot_grad_flow(self.model.named_parameters())" to visualize the gradient flow'''
ave_grads = []
max_grads = []
layers = []
for n, p in named_parameters:
if (p.requires_grad) and ("bias" not in n):
layers.append(n)
ave_grads.append(p.grad.abs().mean())
max_grads.append(p.grad.abs().max())
axis.bar(np.arange(len(max_grads)),
max_grads,
alpha=0.1,
lw=1,
color="c")
axis.bar(np.arange(len(max_grads)),
ave_grads,
alpha=0.1,
lw=1,
color="b")
axis.hlines(0, 0, len(ave_grads) + 1, lw=2, color="k")
axis.set_xticklabels(layers, rotation="vertical")
axis.set_xlim(left=0, right=len(ave_grads))
axis.set_ylim(bottom=-0.001,
top=1.) # zoom in on the lower gradient regions
axis.set_xlabel("Layers")
axis.set_ylabel("average gradient")
axis.set_title(plot_title)
plt.grid(True)
axis.legend([
Line2D([0], [0], color="c", lw=4),
Line2D([0], [0], color="b", lw=4),
Line2D([0], [0], color="k", lw=4)
], ['max-gradient', 'mean-gradient', 'zero-gradient'],
loc='upper left')
def fit(self,
train_data,
eval_interval,
eval,
discrete_columns=tuple(),
epochs=300,
log_frequency=True):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to 300.
log_frequency (boolean):
Whether to use log frequency of categorical levels in conditional
sampling. Defaults to ``True``.
"""
train = train_data
self.transformer = DataTransformer()
self.transformer.fit(train_data, discrete_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, log_frequency)
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
self.discriminator = Discriminator(
data_dim + self.cond_generator.n_opt, self.dis_dim).to(self.device)
optimizerG = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
optimizerD = optim.Adam(self.discriminator.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
assert self.batch_size % 2 == 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
# figure for plotting gradients; ax1 for generator and ax2 for discriminator
#fig, (ax1, ax2) = plt.subplots(1, 2)
steps_per_epoch = max(len(train_data) // self.batch_size, 1)
for i in range(epochs):
for id_ in range(steps_per_epoch):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = self.discriminator(fake_cat)
y_real = self.discriminator(real_cat)
pen = self.discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device)
loss_d = -(torch.mean(y_real) - torch.mean(y_fake))
optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
#plot gradients
#self.plot_grad_flow(self.discriminator.named_parameters(), ax2, 'Gradient flow for D')
optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = self.discriminator(torch.cat([fakeact, c1],
dim=1))
else:
y_fake = self.discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
loss_g = -torch.mean(y_fake) + cross_entropy
optimizerG.zero_grad()
loss_g.backward()
# plot gradients
#self.plot_grad_flow(self.generator.named_parameters(), ax1, 'Gradient flow for G')
optimizerG.step()
print("Epoch %d, Loss G: %.4f, Loss D: %.4f" %
(i + 1, loss_g.detach().cpu(), loss_d.detach().cpu()),
flush=True)
#check model results every x epochs
#fig2, ax3 = plt.subplots(1)
if eval:
stats_real = train[self.demand_column].describe()
stats_real_week = train.groupby('Weekday')[
self.demand_column].describe()
stats_real_month = train.groupby('Month')[
self.demand_column].describe()
if ((i + 1) % eval_interval == 0):
eval_sample = self.sample(1000)
sample = pd.DataFrame(eval_sample,
columns=eval_sample.columns)
#sample.loc[:, self.demand_column].hist(bins=50, alpha=0.4, label='fake')
#ax3.hist(pd.DataFrame(train, columns=train.columns).loc[:, self.demand_column], bins=50, alpha=0.4, label='real')
#fig2.legend()
#fig2.show()
print(
(sample[self.demand_column].describe() - stats_real) /
stats_real)
print(' ')
print(((sample.groupby('Weekday')[
self.demand_column].describe() - stats_real_week) /
stats_real_week).T)
print(' ')
print(((sample.groupby('Month')[
self.demand_column].describe() - stats_real_month) /
stats_real_month).T)
plt.show()
def sample(self, n, condition=None, seed=0):
"""Sample data similar to the training data.
Args:
n (int):
Number of rows to sample.
condvec (Tuple):
Number of column and value of condition to sample
seed (int):
Seed to create result reproducibility
Returns:
numpy.ndarray or pandas.DataFrame
"""
self.generator.eval()
steps = n // self.batch_size + 1
data = []
np.random.seed(seed)
torch.manual_seed(seed)
for i in range(steps):
mean = torch.zeros(self.batch_size, self.embedding_dim)
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
if condition is None:
#select random condition
condvec = self.cond_generator.sample_zero(self.batch_size)
else:
condvec = self.cond_generator.sample_condition(
self.batch_size, condition, self.transformer)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
# sets model back to training mode
self.generator.train()
return self.transformer.inverse_transform(data, None)
def train(self, data, categorical_columns=None, ordinal_columns=None, update_epsilon=None):
if update_epsilon:
self.epsilon = update_epsilon
# this is to make sure data has at least 1000 points, may need it become flexible
self.num_teachers = int(len(data) / 5000) + 1
self.transformer = DataTransformer()
self.transformer.fit(data, discrete_columns=categorical_columns)
data = self.transformer.transform(data)
data_partitions = np.array_split(data, self.num_teachers)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(data, self.transformer.output_info, self.log_frequency)
# create conditional generator for each teacher model
cond_generator = [ConditionalGenerator(d, self.transformer.output_info, self.log_frequency) for d in data_partitions]
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt,
self.gen_dim,
data_dim).to(self.device)
discriminator = Discriminator(
data_dim + self.cond_generator.n_opt,
self.dis_dim,
self.loss,
self.pack).to(self.device)
student_disc = discriminator
student_disc.apply(weights_init)
teacher_disc = [discriminator for i in range(self.num_teachers)]
for i in range (self.num_teachers):
teacher_disc[i].apply(weights_init)
optimizerG = optim.Adam(
self.generator.parameters(), lr=2e-4, betas=(0.5, 0.9), weight_decay=self.l2scale)
optimizerS = optim.Adam(student_disc.parameters(), lr=2e-4, betas=(0.5, 0.9))
optimizerT = [optim.Adam(teacher_disc[i].parameters(), lr=2e-4, betas=(0.5, 0.9)) for i in range(self.num_teachers)]
criterion = nn.BCELoss()
noise_multiplier = 1e-3
alphas = torch.tensor([0.0 for i in range(100)])
l_list = 1 + torch.tensor(range(100))
eps = 0
mean = torch.zeros(self.batch_size, self.embedding_dim, device=self.device)
std = mean + 1
REAL_LABEL = 1
FAKE_LABEL = 0
criterion = nn.BCELoss()
while eps < self.epsilon:
# train teacher discriminators
for t_2 in range(self.teacher_iters):
for i in range(self.num_teachers):
# print ('i is {}'.format(i))
partition_data = data_partitions[i]
data_sampler = Sampler(partition_data, self.transformer.output_info)
fakez = torch.normal(mean, std=std)
condvec = cond_generator[i].sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm], opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
optimizerT[i].zero_grad()
if self.loss == 'cross_entropy':
y_fake =teacher_disc[i](fake_cat)
label_fake = torch.full((int(self.batch_size/self.pack),), FAKE_LABEL, dtype=torch.float, device=self.device)
errD_fake = criterion(y_fake, label_fake.float())
errD_fake.backward()
optimizerT[i].step()
# train with real
label_true = torch.full((int(self.batch_size/self.pack),), REAL_LABEL, dtype=torch.float, device=self.device)
y_real = teacher_disc[i](real_cat)
errD_real = criterion(y_real, label_true.float())
errD_real.backward()
optimizerT[i].step()
loss_d = errD_real + errD_fake
# print("Iterator is {i}, Loss D for teacher {n} is :{j}".format(i=t_2 + 1, n=i+1, j=loss_d.detach().cpu()))
# train student discriminator
for t_3 in range(self.student_iters):
data_sampler = Sampler(data, self.transformer.output_info)
fakez = torch.normal(mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm], opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
else:
fake_cat = fake
fake_data = fake_cat
predictions, votes = pate(fake_data, teacher_disc, noise_multiplier)
output = student_disc(fake_data.detach())
# update moments accountant
alphas = alphas + moments_acc(self.num_teachers, votes, noise_multiplier, l_list)
loss_s = criterion(output, predictions.float().to(self.device))
loss_s.backward()
optimizerS.step()
# print ('iterator {i}, student discriminator loss is {j}'.format(i=t_3, j=loss_s))
# train generator
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = student_disc(torch.cat([fakeact, c1], dim=1))
else:
y_fake = student_disc(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
if self.loss=='cross_entropy':
label_g = torch.full((int(self.batch_size/self.pack),), REAL_LABEL, dtype=torch.float, device=self.device)
#label_g = torch.full(int(self.batch_size/self.pack,),1,device=self.device)
loss_g = criterion(y_fake, label_g.float())
loss_g = loss_g + cross_entropy
else:
loss_g = -torch.mean(y_fake) + cross_entropy
optimizerG.zero_grad()
loss_g.backward()
optimizerG.step()
print ('generator is {}'.format(loss_g.detach().cpu()))
eps = min((alphas - math.log(self.delta)) / l_list)
class CTGANSynthesizer(object):
"""Conditional Table GAN Synthesizer.
This is the core class of the CTGAN project, where the different components
are orchestrated together.
For more details about the process, please check the [Modeling Tabular data using
Conditional GAN](https://arxiv.org/abs/1907.00503) paper.
Args:
embedding_dim (int):
Size of the random sample passed to the Generator. Defaults to 128.
gen_dim (tuple or list of ints):
Size of the output samples for each one of the Residuals. A Resiudal Layer
will be created for each one of the values provided. Defaults to (256, 256).
dis_dim (tuple or list of ints):
Size of the output samples for each one of the Discriminator Layers. A Linear Layer
will be created for each one of the values provided. Defaults to (256, 256).
l2scale (float):
Wheight Decay for the Adam Optimizer. Defaults to 1e-6.
batch_size (int):
Number of data samples to process in each step.
"""
def __init__(self,
embedding_dim=128,
gen_dim=(256, 256),
dis_dim=(256, 256),
l2scale=1e-6,
batch_size=500,
gen_lr=2e-4,
dis_lr=2e-4,
verbose=False):
self.embedding_dim = embedding_dim
self.gen_dim = gen_dim
self.dis_dim = dis_dim
self.l2scale = l2scale
self.batch_size = batch_size
self.gen_lr = gen_lr
self.dis_lr = dis_lr
self.device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.verbose = verbose
def _apply_activate(self, data):
data_t = []
st = 0
for item in self.transformer.output_info:
if item[1] == 'tanh':
ed = st + item[0]
data_t.append(torch.tanh(data[:, st:ed]))
st = ed
elif item[1] == 'softmax':
ed = st + item[0]
weights = functional.gumbel_softmax(data[:, st:ed], tau=0.2)
while np.isnan(torch.sum(weights).cpu().detach().numpy()):
weights = functional.gumbel_softmax(data[:, st:ed],
tau=0.2)
data_t.append(weights)
st = ed
else:
assert 0
return torch.cat(data_t, dim=1)
def _cond_loss(self, data, c, m):
loss = []
st = 0
st_c = 0
skip = False
for item in self.transformer.output_info:
if item[1] == 'tanh':
st += item[0]
skip = True
elif item[1] == 'softmax':
if skip:
skip = False
st += item[0]
continue
ed = st + item[0]
ed_c = st_c + item[0]
tmp = functional.cross_entropy(data[:, st:ed],
torch.argmax(c[:, st_c:ed_c],
dim=1),
reduction='none')
loss.append(tmp)
st = ed
st_c = ed_c
else:
assert 0
loss = torch.stack(loss, dim=1)
return (loss * m).sum() / data.size()[0]
def fit(self,
train_data,
discrete_columns=tuple(),
epochs=300,
log_frequency=True):
"""Fit the CTGAN Synthesizer models to the training data.
Args:
train_data (numpy.ndarray or pandas.DataFrame):
Training Data. It must be a 2-dimensional numpy array or a
pandas.DataFrame.
discrete_columns (list-like):
List of discrete columns to be used to generate the Conditional
Vector. If ``train_data`` is a Numpy array, this list should
contain the integer indices of the columns. Otherwise, if it is
a ``pandas.DataFrame``, this list should contain the column names.
epochs (int):
Number of training epochs. Defaults to 300.
log_frequency (boolean):
Whether to use log frequency of categorical levels in conditional
sampling. Defaults to ``True``.
"""
self.discrete_columns = discrete_columns
# Fit OH encoder
self.oe = OneHotEncoder(sparse=False)
if discrete_columns:
if type(train_data) is np.ndarray:
self.oe.fit(train_data[:, discrete_columns])
self.cat_column_idxes = discrete_columns
else:
self.oe.fit(train_data[discrete_columns])
features = train_data.columns
self.cat_column_idxes = [
features.index(c) for c in discrete_columns
]
else:
self.cat_column_idxes = []
self.transformer = DataTransformer()
self.transformer.fit(train_data, discrete_columns)
train_data = self.transformer.transform(train_data)
data_sampler = Sampler(train_data, self.transformer.output_info)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
train_data, self.transformer.output_info, log_frequency)
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
discriminator = Discriminator(data_dim + self.cond_generator.n_opt,
self.dis_dim).to(self.device)
optimizerG = optim.Adam(self.generator.parameters(), lr=self.gen_lr)
optimizerD = optim.Adam(
discriminator.parameters(),
lr=self.dis_lr,
)
assert self.batch_size % 2 == 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
steps_per_epoch = max(len(train_data) // self.batch_size, 1)
for i in tqdm(range(epochs)):
for id_ in range(steps_per_epoch):
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype('float32')).to(self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
y_fake = discriminator(fake_cat)
y_real = discriminator(real_cat)
pen = discriminator.calc_gradient_penalty(
real_cat, fake_cat, self.device)
loss_d = -(nanmean(y_real) - nanmean(y_fake))
optimizerD.zero_grad()
pen.backward(retain_graph=True)
loss_d.backward()
optimizerD.step()
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = discriminator(torch.cat([fakeact, c1], dim=1))
else:
y_fake = discriminator(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
loss_g = -nanmean(y_fake) + cross_entropy
optimizerG.zero_grad()
loss_g.backward()
optimizerG.step()
if self.verbose:
print("Epoch %d, Loss G: %.4f, Loss D: %.4f" %
(i + 1, loss_g.detach().cpu(), loss_d.detach().cpu()),
flush=True)
self.discriminator = discriminator
def sample(self, n, test_instance=None):
"""Sample data similar to the training data.
Args:
n (int):
Number of rows to sample.
test_instance (Optional[np.ndarray]):
Test sample to condition sampler with
Returns:
numpy.ndarray or pandas.DataFrame
"""
steps = n // self.batch_size + 1
data = []
for i in range(steps):
mean = torch.zeros(self.batch_size, self.embedding_dim)
std = mean + 1
fakez = torch.normal(mean=mean, std=std).to(self.device)
if test_instance is not None and len(self.cat_column_idxes) > 0:
if type(test_instance) is pd.DataFrame:
test_instance = test_instance.values
test_cat_onehot = self.oe.transform(
test_instance[:, self.cat_column_idxes]).ravel()
nonzeroes = np.nonzero(test_cat_onehot)[0]
samples = np.random.choice(
nonzeroes, size=self.batch_size).astype(np.int8)
condvec = np.zeros(
(self.batch_size, self.cond_generator.n_opt))
condvec[list(range(self.batch_size)), samples] = 1
else:
condvec = self.cond_generator.sample_zero(self.batch_size)
if condvec is None:
pass
else:
c1 = condvec
c1 = torch.from_numpy(c1).type(torch.FloatTensor).to(
self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
data.append(fakeact.detach().cpu().numpy())
data = np.concatenate(data, axis=0)
data = data[:n]
return self.transformer.inverse_transform(data, None)
def train(self,
data,
categorical_columns=None,
ordinal_columns=None,
update_epsilon=None):
if update_epsilon:
self.epsilon = update_epsilon
sample_per_teacher = (self.sample_per_teacher
if self.sample_per_teacher < len(data) else 1000)
self.num_teachers = int(len(data) / sample_per_teacher) + 1
self.transformer = DataTransformer()
self.transformer.fit(data, discrete_columns=categorical_columns)
data = self.transformer.transform(data)
data_partitions = np.array_split(data, self.num_teachers)
data_dim = self.transformer.output_dimensions
self.cond_generator = ConditionalGenerator(
data, self.transformer.output_info, self.log_frequency)
# create conditional generator for each teacher model
cond_generator = [
ConditionalGenerator(d, self.transformer.output_info,
self.log_frequency) for d in data_partitions
]
self.generator = Generator(
self.embedding_dim + self.cond_generator.n_opt, self.gen_dim,
data_dim).to(self.device)
discriminator = Discriminator(data_dim + self.cond_generator.n_opt,
self.dis_dim, self.loss,
self.pack).to(self.device)
student_disc = discriminator
student_disc.apply(weights_init)
teacher_disc = [discriminator for i in range(self.num_teachers)]
for i in range(self.num_teachers):
teacher_disc[i].apply(weights_init)
optimizer_g = optim.Adam(self.generator.parameters(),
lr=2e-4,
betas=(0.5, 0.9),
weight_decay=self.l2scale)
optimizer_s = optim.Adam(student_disc.parameters(),
lr=2e-4,
betas=(0.5, 0.9))
optimizer_t = [
optim.Adam(teacher_disc[i].parameters(), lr=2e-4, betas=(0.5, 0.9))
for i in range(self.num_teachers)
]
noise_multiplier = self.noise_multiplier
alphas = torch.tensor([0.0 for i in range(self.moments_order)],
device=self.device)
l_list = 1 + torch.tensor(range(self.moments_order),
device=self.device)
eps = 0
mean = torch.zeros(self.batch_size,
self.embedding_dim,
device=self.device)
std = mean + 1
real_label = 1
fake_label = 0
criterion = nn.BCELoss() if (self.loss
== "cross_entropy") else self.w_loss
if self.verbose:
print("using loss {} and regularization {}".format(
self.loss, self.regularization))
while eps < self.epsilon:
# train teacher discriminators
for t_2 in range(self.teacher_iters):
for i in range(self.num_teachers):
partition_data = data_partitions[i]
data_sampler = Sampler(partition_data,
self.transformer.output_info)
fakez = torch.normal(mean, std=std).to(self.device)
condvec = cond_generator[i].sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
real = torch.from_numpy(real.astype("float32")).to(
self.device)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
real_cat = torch.cat([real, c2], dim=1)
else:
real_cat = real
fake_cat = fake
optimizer_t[i].zero_grad()
y_all = torch.cat(
[teacher_disc[i](fake_cat), teacher_disc[i](real_cat)])
label_fake = torch.full(
(int(self.batch_size / self.pack), 1),
fake_label,
dtype=torch.float,
device=self.device,
)
label_true = torch.full(
(int(self.batch_size / self.pack), 1),
real_label,
dtype=torch.float,
device=self.device,
)
labels = torch.cat([label_fake, label_true])
error_d = criterion(y_all, labels)
error_d.backward()
if self.regularization == "dragan":
pen = teacher_disc[i].dragan_penalty(
real_cat, device=self.device)
pen.backward(retain_graph=True)
optimizer_t[i].step()
# train student discriminator
for t_3 in range(self.student_iters):
data_sampler = Sampler(data, self.transformer.output_info)
fakez = torch.normal(mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
real = data_sampler.sample(self.batch_size, col, opt)
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
perm = np.arange(self.batch_size)
np.random.shuffle(perm)
real = data_sampler.sample(self.batch_size, col[perm],
opt[perm])
c2 = c1[perm]
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
fake_cat = torch.cat([fakeact, c1], dim=1)
else:
fake_cat = fake
fake_data = fake_cat
predictions, votes = pate(fake_data,
teacher_disc,
noise_multiplier,
device=self.device)
output = student_disc(fake_data.detach())
# update moments accountant
alphas = alphas + moments_acc(self.num_teachers,
votes,
noise_multiplier,
l_list,
device=self.device)
loss_s = criterion(output, predictions.float().to(self.device))
optimizer_s.zero_grad()
loss_s.backward()
if self.regularization == "dragan":
vals = torch.cat([predictions, fake_data], axis=1)
ordered = vals[vals[:, 0].sort()[1]]
data_list = torch.split(
ordered,
predictions.shape[0] - int(predictions.sum().item()))
synth_cat = torch.cat(data_list[1:], axis=0)[:, 1:]
pen = student_disc.dragan_penalty(synth_cat,
device=self.device)
pen.backward(retain_graph=True)
optimizer_s.step()
# print ('iterator {i}, student discriminator loss is {j}'.format(i=t_3, j=loss_s))
# train generator
fakez = torch.normal(mean=mean, std=std)
condvec = self.cond_generator.sample(self.batch_size)
if condvec is None:
c1, m1, col, opt = None, None, None, None
else:
c1, m1, col, opt = condvec
c1 = torch.from_numpy(c1).to(self.device)
m1 = torch.from_numpy(m1).to(self.device)
fakez = torch.cat([fakez, c1], dim=1)
fake = self.generator(fakez)
fakeact = self._apply_activate(fake)
if c1 is not None:
y_fake = student_disc(torch.cat([fakeact, c1], dim=1))
else:
y_fake = student_disc(fakeact)
if condvec is None:
cross_entropy = 0
else:
cross_entropy = self._cond_loss(fake, c1, m1)
if self.loss == "cross_entropy":
label_g = torch.full(
(int(self.batch_size / self.pack), 1),
real_label,
dtype=torch.float,
device=self.device,
)
loss_g = criterion(y_fake, label_g.float())
loss_g = loss_g + cross_entropy
else:
loss_g = -torch.mean(y_fake) + cross_entropy
optimizer_g.zero_grad()
loss_g.backward()
optimizer_g.step()
eps = min((alphas - math.log(self.delta)) / l_list)
if self.verbose:
print("eps: {:f} \t G: {:f} \t D: {:f}".format(
eps,
loss_g.detach().cpu(),
loss_s.detach().cpu()))
