def _run(self):
self.best_network_rep = None
self.best_loss_val_rep = 100000
self._initialize()
x_batch, y_batch = self.dataset.x_train, self.dataset.y_train
x_train, x_val, y_train, y_val = self.train_val_split(x_batch,
y_batch,
val_ratio=0.2)
x_train, y_train = _prepare_batch_data(
x_batch=x_train,
y_batch=y_train,
problem_type=self.config.problem_type,
is_gpu=self.config.is_gpu,
n_input=self.config.n_input,
n_output=self.config.n_output,
n_samples=1)
x_val, y_val = _prepare_batch_data(
x_batch=x_val,
y_batch=y_val,
problem_type=self.config.problem_type,
is_gpu=self.config.is_gpu,
n_input=self.config.n_input,
n_output=self.config.n_output,
n_samples=1)
if self.is_cuda:
x_train = x_train.cuda()
y_train = y_train.cuda()
x_val = x_val.cuda()
y_val = y_val.cuda()
# train
for epoch in range(self.n_epochs):
loss_train = self._train_one(x_train, y_train)
_, _, _, loss_val = self._evaluate(x_val,
y_val,
network=self.network)
self.backprop_report.report_epoch(epoch, loss_train, loss_val)
if loss_val < self.best_loss_val_rep:
self.best_loss_val_rep = loss_val
self.best_network_rep = copy.deepcopy(self.network)
self.last_update = epoch
print(f'New best network: {loss_val}')
if epoch - self.last_update > N_EPOCHS_WITHOUT_IMPROVING:
print(
f'Breaking training as not improving for {N_EPOCHS_WITHOUT_IMPROVING} epochs'
)
break
if epoch % 100 == 0:
print(f'Epoch = {epoch}. Error: {loss_train}')
print(f'Final Train Error: {loss_train}')
print(f'Best Val Error: {self.best_loss_val_rep}')
def evaluate(self):
x_batch, y_batch = self.dataset.x_test, self.dataset.y_test
x_batch, y_batch = _prepare_batch_data(
x_batch=x_batch,
y_batch=y_batch,
problem_type=self.config.problem_type,
is_gpu=self.config.is_gpu,
n_input=self.config.n_input,
n_output=self.config.n_output,
n_samples=1)
if self.is_cuda:
x_batch = x_batch.cuda()
y_batch = y_batch.cuda()
x, y_true, y_pred, _ = self._evaluate(x_batch,
y_batch,
network=self.best_network)
if self.is_cuda:
x = x.cpu()
y_pred = y_pred.cpu()
y_true = y_true.cpu()
return x, y_true, y_pred
def _evaluate_genome_parallel_jupyneat(genome: dict,
loss,
beta_type,
problem_type,
n_input,
n_output,
activation,
batch_size=10000,
n_samples=10,
is_gpu=False):
'''
Calculates: KL-Div(q(w)||p(w|D))
Uses the VariationalInferenceLoss class (not the alternative)
'''
kl_posterior = 0
# TODO: fix
# kl_qw_pw = compute_kl_qw_pw(genome=genome)
kl_qw_pw = 0.0
# setup network
network = ComplexStochasticNetworkJupyneat(genome=genome,
n_input=n_input,
n_output=n_output,
activation_type=activation)
if is_gpu:
network.cuda()
m = math.ceil(len(dataset.x) / batch_size)
network.eval()
# calculate Data log-likelihood (p(y*|x*,D))
x_batch, y_batch = dataset.x, dataset.y
x_batch, y_batch = _prepare_batch_data(x_batch=x_batch,
y_batch=y_batch,
problem_type=problem_type,
is_gpu=is_gpu,
n_input=n_input,
n_output=n_output,
n_samples=n_samples)
if is_gpu:
x_batch, y_batch = x_batch.cuda(), y_batch.cuda()
with torch.no_grad():
# forward pass
output, _ = network(x_batch)
beta = get_beta(beta_type=beta_type,
m=m,
batch_idx=0,
epoch=1,
n_epochs=1)
kl_posterior += loss(y_pred=output,
y_true=y_batch,
kl_qw_pw=kl_qw_pw,
beta=beta)
loss_value = kl_posterior.item()
return loss_value
def _evaluate_genome_parallel(genome: Genome,
loss,
beta_type,
problem_type,
is_testing,
batch_size=10000,
n_samples=10,
is_gpu=False):
'''
Calculates: KL-Div(q(w)||p(w|D))
Uses the VariationalInferenceLoss class (not the alternative)
'''
# kl_posterior = 0
#
# kl_qw_pw = compute_kl_qw_pw(genome=genome)
# setup network
network = ComplexStochasticNetwork(genome=genome)
if is_gpu:
network.cuda()
m = math.ceil(len(dataset.x) / batch_size)
network.eval()
# calculate Data log-likelihood (p(y*|x*,D))
if is_testing:
x_batch, y_batch = dataset.x_test, dataset.y_test
else:
x_batch, y_batch = dataset.x_train, dataset.y_train
x_batch, y_batch = _prepare_batch_data(x_batch=x_batch,
y_batch=y_batch,
problem_type=problem_type,
is_gpu=is_gpu,
n_input=genome.n_input,
n_output=genome.n_output,
n_samples=n_samples)
with torch.no_grad():
# forward pass
output, kl_qw_pw = network(x_batch)
output, _, y_batch = _process_output_data(output,
y_true=y_batch,
n_samples=n_samples,
n_output=genome.n_output,
problem_type=problem_type,
is_pass=True)
beta = get_beta(beta_type=beta_type,
m=m,
batch_idx=0,
epoch=1,
n_epochs=1)
kl_posterior = loss(y_pred=output,
y_true=y_batch,
kl_qw_pw=kl_qw_pw,
beta=beta)
loss_value = kl_posterior.item()
return loss_value
def calculate_prediction_distribution(network,
dataset,
problem_type,
is_testing,
n_samples=1000,
use_sigmoid=False):
'''
Calculate Predictive Distribution for a network and dataset
'''
# setup network
network.eval()
# calculate Data log-likelihood (p(y*|x*,D))
if is_testing:
x_batch, y_batch = dataset.x_test, dataset.y_test
else:
x_batch, y_batch = dataset.x_train, dataset.y_train
x_batch, y_batch = _prepare_batch_data(x_batch=x_batch,
y_batch=y_batch,
problem_type=problem_type,
is_gpu=False,
n_input=network.n_input,
n_output=network.n_output,
n_samples=n_samples)
chunks_x = []
chunks_output_distribution = []
chunks_y_true = []
with torch.no_grad():
# forward pass
output, _ = network(x_batch)
_, output_distribution, y_batch = _process_output_data(
output,
y_true=y_batch,
n_samples=n_samples,
n_output=network.n_output,
problem_type=problem_type,
is_pass=True)
chunks_x.append(x_batch)
chunks_output_distribution.append(output_distribution)
chunks_y_true.append(y_batch)
x = torch.cat(chunks_x, dim=0)
output_distribution = torch.cat(chunks_output_distribution, dim=0)
y_true = torch.cat(chunks_y_true, dim=0)
return x, y_true, output_distribution
def run(self) -> None:
'''
Calculates: KL-Div(q(w)||p(w|D))
Uses the VariationalInferenceLoss class (not the alternative)
'''
# from experiments.multiprocessing_utils import ForkedPdb; ForkedPdb().set_trace()
kl_posterior = 0
kl_qw_pw = compute_kl_qw_pw(genome=self.genome)
# setup network
network = ComplexStochasticNetwork(genome=self.genome)
# m = math.ceil(len(self.dataset) / self.batch_size)
m = math.ceil(len(self.x) / self.batch_size)
network.eval()
# calculate Data log-likelihood (p(y*|x*,D))
# x_batch, y_batch = self.dataset.x, self.dataset.y
x_batch, y_batch = self.x, self.y
x_batch, y_batch = _prepare_batch_data(x_batch=x_batch,
y_batch=y_batch,
problem_type=self.problem_type,
is_gpu=self.is_gpu,
n_input=self.genome.n_input,
n_output=self.genome.n_output,
n_samples=self.n_samples)
print('running forward pass')
with torch.no_grad():
# forward pass
output, _ = network(x_batch)
print('forward pass completed')
# print(self.config.beta_type)
beta = get_beta(beta_type=self.beta_type,
m=m,
batch_idx=0,
epoch=1,
n_epochs=1)
# print(f'Beta: {beta}')
kl_posterior += self.loss(y_pred=output,
y_true=y_batch,
kl_qw_pw=kl_qw_pw,
beta=beta)
loss_value = kl_posterior.item()
self.result = (self.genome.key, loss_value)
def evaluate_genome(genome: Genome, loss, beta_type, problem_type,
batch_size=10000, n_samples=10, is_gpu=False):
'''
Calculates: KL-Div(q(w)||p(w|D))
Uses the VariationalInferenceLoss class (not the alternative)
# '''
# dataset = get_dataset(genome.genome_config.dataset, testing=True)
# dataset.generate_data()
kl_posterior = 0
kl_qw_pw = compute_kl_qw_pw(genome=genome)
# setup network
network = ComplexStochasticNetwork(genome=genome)
if is_gpu:
network.cuda()
m = math.ceil(len(dataset.x) / batch_size)
network.eval()
# calculate Data log-likelihood (p(y*|x*,D))
x_batch, y_batch = dataset.x, dataset.y
x_batch, y_batch = _prepare_batch_data(x_batch=x_batch,
y_batch=y_batch,
problem_type=problem_type,
is_gpu=is_gpu,
n_input=genome.n_input,
n_output=genome.n_output,
n_samples=n_samples)
if is_gpu:
x_batch, y_batch = x_batch.cuda(), y_batch.cuda()
with torch.no_grad():
# forward pass
output, _ = network(x_batch)
# print(self.config.beta_type)
beta = get_beta(beta_type=beta_type, m=m, batch_idx=0, epoch=1, n_epochs=1)
# print(f'Beta: {beta}')
kl_posterior += loss(y_pred=output, y_true=y_batch, kl_qw_pw=kl_qw_pw, beta=beta)
loss_value = kl_posterior.item()
return loss_value
def evaluate(self, is_testing=True):
if is_testing:
x_batch, y_batch = self.dataset.x_test, self.dataset.y_test
else:
x_batch, y_batch = self.dataset.x_train, self.dataset.y_train
x_batch, _ = _prepare_batch_data(x_batch=x_batch,
y_batch=y_batch,
problem_type=self.config.problem_type,
is_gpu=self.config.is_gpu,
n_input=self.config.n_input,
n_output=self.config.n_output,
n_samples=EVALUATION_N_SAMPLES)
network = self.best_network
# network = ProbabilisticFeedForward(n_input=self.config.n_input, n_output=self.config.n_output,
# is_cuda=self.is_cuda,
# n_neurons_per_layer=self.n_neurons_per_layer,
# n_hidden_layers=self.n_hidden_layers)
# network.load_state_dict(self.best_network)
if self.is_cuda:
network.cuda()
x_batch = x_batch.cuda()
y_batch = y_batch.cuda()
x, y_true, y_pred, loss = self._evaluate(x_batch,
y_batch,
network=network)
if self.is_cuda:
x = x.cpu()
y_true = y_true.cpu()
y_pred = y_pred.cpu()
return x, y_true, y_pred
def _run(self):
# TODO: remove (just debugging)
# torch.autograd.set_detect_anomaly(True)
self._initialize()
x_batch, y_batch = self.dataset.x_train, self.dataset.y_train
x_train, x_val, y_train, y_val = self.train_val_split(x_batch,
y_batch,
val_ratio=0.2)
x_train, _ = _prepare_batch_data(x_batch=x_train,
y_batch=y_train,
problem_type=self.config.problem_type,
is_gpu=self.config.is_gpu,
n_input=self.config.n_input,
n_output=self.config.n_output,
n_samples=self.config.n_samples)
x_val, _ = _prepare_batch_data(x_batch=x_val,
y_batch=y_val,
problem_type=self.config.problem_type,
is_gpu=self.config.is_gpu,
n_input=self.config.n_input,
n_output=self.config.n_output,
n_samples=EVALUATION_N_SAMPLES)
if self.is_cuda:
x_train = x_train.cuda()
y_train = y_train.cuda()
x_val = x_val.cuda()
y_val = y_val.cuda()
if self.config.problem_type == 'classification':
self._train_one = self._train_one_classification
elif self.config.problem_type == 'regression':
self._train_one = self._train_one_regression
# train
for epoch in range(self.n_epochs):
loss_train = self._train_one(x_train, y_train)
_, _, _, loss_val = self._evaluate(x_val,
y_val,
network=self.network)
self.backprop_report.report_epoch(epoch, loss_train, loss_val)
if loss_val < self.best_loss_val_rep:
self.best_loss_val_rep = loss_val
self.best_network_rep = copy.deepcopy(self.network)
self.last_update = epoch
print(f'New best network: {loss_val}')
if epoch - self.last_update > N_EPOCHS_WITHOUT_IMPROVING:
print(
f'Breaking training as not improving for {N_EPOCHS_WITHOUT_IMPROVING} epochs'
)
break
if epoch % 100 == 0:
print(f'Epoch = {epoch}. Error: {loss_train}')
print(f'Final Train Error: {loss_train}')
print(f'Best Val Error: {self.best_loss_val_rep}')
def evaluate_genome_with_dataloader(genome: Genome,
data_loader,
loss,
beta_type,
problem_type,
batch_size=10000,
n_samples=10,
is_gpu=False,
return_all=False):
'''
Calculates: KL-Div(q(w)||p(w|D))
Uses the VariationalInferenceLoss class (not the alternative)
'''
kl_posterior = 0
kl_qw_pw = compute_kl_qw_pw(genome=genome)
# setup network
network = ComplexStochasticNetwork(genome=genome)
if is_gpu:
network.cuda()
m = math.ceil(len(data_loader) / batch_size)
network.eval()
chunks_x = []
chunks_y_pred = []
chunks_y_true = []
# calculate Data log-likelihood (p(y*|x*,D))
for batch_idx, (x_batch, y_batch) in enumerate(data_loader):
x_batch, y_batch = _prepare_batch_data(x_batch=x_batch,
y_batch=y_batch,
problem_type=problem_type,
is_gpu=is_gpu,
n_input=genome.n_input,
n_output=genome.n_output,
n_samples=n_samples)
with torch.no_grad():
# forward pass
output, _ = network(x_batch)
beta = get_beta(beta_type=beta_type,
m=m,
batch_idx=batch_idx,
epoch=1,
n_epochs=1)
kl_posterior += loss(y_pred=output,
y_true=y_batch,
kl_qw_pw=kl_qw_pw,
beta=beta)
if return_all:
chunks_x.append(x_batch)
chunks_y_pred.append(output)
chunks_y_true.append(y_batch)
loss_value = kl_posterior.item()
if return_all:
x = torch.cat(chunks_x, dim=0)
y_pred = torch.cat(chunks_y_pred, dim=0)
y_true = torch.cat(chunks_y_true, dim=0)
return x, y_true, y_pred, loss_value
return loss_value
def train(self, genome):
kl_qw_pw = compute_kl_qw_pw(genome=genome)
# setup network
self.network = ComplexStochasticNetwork(genome=genome,
is_trainable=True,
is_cuda=self.is_cuda)
self.criterion = get_loss(problem_type=self.problem_type)
if self.is_cuda:
self.network.cuda()
self.criterion.cuda()
self.optimizer = Adam(self.network.parameters(),
lr=self.lr,
weight_decay=self.weight_decay)
x_batch, y_batch = self.dataset.x_train, self.dataset.y_train
x_train, x_val, y_train, y_val = self.train_val_split(
x_batch, y_batch, problem_type=self.problem_type, val_ratio=0.2)
x_train, _ = _prepare_batch_data(
x_batch=x_train,
y_batch=y_train,
problem_type=self.problem_type,
is_gpu=False, # this could be removed
n_input=genome.n_input,
n_output=genome.n_output,
n_samples=self.n_samples)
x_val, _ = _prepare_batch_data(x_batch=x_val,
y_batch=y_val,
problem_type=self.problem_type,
is_gpu=False,
n_input=genome.n_input,
n_output=genome.n_output,
n_samples=self.n_samples)
if self.is_cuda:
x_train = x_train.cuda()
y_train = y_train.cuda()
x_val = x_val.cuda()
y_val = y_val.cuda()
self.network.train()
for epoch in range(self.n_epochs):
loss_epoch = self._train_one(x_train, y_train, kl_qw_pw)
# if epoch % 10 == 0:
_, _, _, loss_val = self._evaluate(x_val,
y_val,
network=self.network)
if loss_val < self.best_loss_val:
self.best_loss_val = loss_val
self.best_network_state = copy.deepcopy(
self.network.state_dict())
self.last_update = epoch
if epoch - self.last_update > N_EPOCHS_WITHOUT_IMPROVING:
print(
f'Breaking training as not improving for {N_EPOCHS_WITHOUT_IMPROVING} epochs'
)
break
if epoch % 200 == 0:
print(f'Epoch = {epoch}. Training Loss: {loss_epoch}. '
f'Best Val. Loss: {self.best_loss_val}')
self.network.clear_non_existing_weights(
clear_grad=False) # reset non-existing weights
self.final_loss = loss_epoch
print(f'Final Epoch = {epoch}. Training Error: {self.final_loss}')
