def __init__(self, arguments):
self.arguments = arguments
input_dims = 28, 28, 1
self.ae = AutoEncoder(
input_dims=input_dims,
encoder_filters=arguments.encoder_filters,
encoder_conv_kernels=arguments.encoder_conv_kernels,
encoder_conv_strides=arguments.encoder_conv_strides,
decoder_filters=arguments.decoder_filters,
decoder_conv_kernels=arguments.decoder_conv_kernels,
decoder_conv_strides=arguments.decoder_conv_strides,
latent_dim=arguments.latent_dim,
use_batch_norm=arguments.use_batch_norm,
use_dropout=arguments.use_dropout).to(device)
self.ae.train()
self.criterion = nn.MSELoss()
self.encoder_optimizer = opt.Adam(params=self.ae.encoder.parameters(),
lr=arguments.learning_rate)
self.decoder_optimizer = opt.Adam(params=self.ae.decoder.parameters(),
lr=arguments.learning_rate)
self.writer = SummaryWriter(
logdir=os.path.join(self.arguments.log_dir, self.arguments.data),
comment='epoch_{0:03d}_batch_size_{1:03d}_lr_{2:.03f}'.format(
self.arguments.epochs - 1, self.arguments.batch_size,
self.arguments.learning_rate))
def run(data_obj, training_size):
data_obj.split_dataset(training_size)
data_obj.preprocess()
#-------------------------------- Autoencoder model
ae_model = AutoEncoder(data_obj.x_train_scaled.shape[1],
training_size, data_obj.name)
ae_model.train(data_obj.x_train_scaled, data_obj.x_val_scaled)
#-------------------------------- Encoded representation
x_train_encoded, x_val_encoded, x_test_encoded = ae_model.encoded_data(
data_obj.x_train_scaled, data_obj.x_val_scaled, data_obj.x_test_scaled)
#-------------------------------- Neural Network model
nn_model = NeuralNetwork(
data_obj.x_train_scaled.shape[1], data_obj.y_train.shape[1],
training_size, data_obj.name)
nn_model.train(
x_train_encoded, data_obj.y_train, x_val_encoded, data_obj.y_val)
nn_model.evaluate(x_test_encoded, data_obj.y_test)
#-------------------------------- reset data from memory
data_obj.reset_scalar()
return nn_model.result()
class trainAE(object):
def __init__(self, path, k, lr= 0.1, batch_size=1, loss='bce', n_epochs=100):
'''
Arguments:
path : path to training data
k : hidde unit's dimension
lr : learning rate
batch_size : batch_size for training, currently set to 1
loss : loss function (bce, rmse) to train AutoEncoder
n_epochs : number of epochs for training
'''
self.AE = AutoEncoder(path, k)
# Definne the autoencoder model
#self.AE.model()
self.AE.model_batch()
self.epochs = n_epochs
def train(self):
T = self.AE.T
# COnverting to csr format for indexing
T = T.tocsr()
#pdb.set_trace()
nonzero_indices = T.nonzero()
for epoch in xrange(self.epochs):
print("Running epoch %d"%(epoch))
for i in np.unique(nonzero_indices[0]):
# get indices of observed values from the user 'i' 's vector
indices = T[i, :].nonzero()[1]
#print indices
#indices = indices.reshape(indices.shape[0],)
# Get correspoding ratings
ratings = T[i, indices].toarray()
#print ratings
ratings = ratings.reshape(ratings.shape[1],)
# Convert inputs to theano datatype
indices = indices.astype(np.int32)
ratings = ratings.astype(np.int32)
#pdb.set_trace()
loss = self.AE.ae(indices, ratings)
print("Loss at epoch %d is %f"%(epoch, loss))
# Batch training method
def train_batch(self, batch_size):
T = self.AE.T
T = T.tocsr()
nonzero_indices = T.nonzero()
#pdb.set_trace()
n_users = len(np.unique(nonzero_indices[0]))
indices = np.unique(nonzero_indices[0])
for epoch in xrange(self.epochs):
for ind, i in enumerate(xrange(0, n_users, batch_size)):
ratings = T[indices[i:(i + batch_size)], :].toarray().astype(np.float32)
#print ratings
#pdb.set_trace()
loss = self.AE.ae_batch(ratings)
#loss = self.AE.debug(ratings)
print loss
def main(opt):
device = 'cuda' if torch.cuda.is_available() else 'cpu'
# Dataset
print('Dataset....')
transform = transforms.Compose([
transforms.Resize((600, 600)),
transforms.Grayscale(3),
transforms.ToTensor()
])
train_set = myDataset(image_path=opt.train_path, transform=transform)
val_set = myDataset(image_path=opt.val_path, transform=transform)
train_loader = DataLoader(train_set, batch_size=opt.train_batch_size)
val_loader = DataLoader(val_set, batch_size=opt.val_batch_size)
# Model
print('Model....')
model = AutoEncoder()
model.to(device)
# Optimizer
optimizer = optim.Adam(model.parameters(), lr=opt.lr)
loss_func = nn.MSELoss()
# Train
print('Training....')
train_epoch_loss = []
val_epoch_loss = []
train_iter_losses = []
val_iter_losses = []
for e in range(opt.epoch):
train_iter_loss = train(opt, model, train_loader, optimizer, loss_func,
device, e)
train_iter_losses += train_iter_loss
train_epoch_loss.append(sum(train_iter_loss))
val_iter_loss = val(opt, model, val_loader, loss_func, device, e)
val_iter_losses += val_iter_loss
val_epoch_loss.append(sum(val_iter_loss))
# save model
best = 10000
if val_epoch_loss[-1] < best:
print('Saving Model....')
torch.save(model, 'weights/AutoEncoder_try1.pth')
best = val_epoch_loss[-1]
print('Saving Result')
plt.figure(figsize=(10, 10))
plt.plot(train_iter_losses)
plt.plot(val_iter_losses)
plt.legend(['Train_loss', 'Val_loss'])
plt.savefig('Result.jpg')
def main():
use_cuda = args.use_cuda
train_data = UnlabeledContact(data=args.data_dir)
print('Number of samples: {}'.format(len(train_data)))
trainloader = DataLoader(train_data, batch_size=args.batch_size)
# Contact matrices are 21x21
input_size = 441
img_height = 21
img_width = 21
vae = AutoEncoder(code_size=20,
imgsize=input_size,
height=img_height,
width=img_width)
criterion = nn.BCEWithLogitsLoss()
if use_cuda:
#vae = nn.DataParallel(vae)
vae = vae.cuda() #.half()
criterion = criterion.cuda()
optimizer = optim.SGD(vae.parameters(), lr=0.01)
clock = AverageMeter(name='clock32single', rank=0)
epoch_loss = 0
total_loss = 0
end = time.time()
for epoch in range(15):
for batch_idx, data in enumerate(trainloader):
inputs = data['cont_matrix']
inputs = inputs.resize_(args.batch_size, 1, 21, 21)
inputs = inputs.float()
if use_cuda:
inputs = inputs.cuda() #.half()
inputs = Variable(inputs)
optimizer.zero_grad()
output, code = vae(inputs)
loss = criterion(output, inputs)
loss.backward()
optimizer.step()
epoch_loss += loss.data[0]
clock.update(time.time() - end)
end = time.time()
if batch_idx % args.log_interval == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(
epoch, batch_idx * len(data), len(trainloader.dataset),
100. * batch_idx / len(trainloader), loss.data[0]))
clock.save(
path=
'/home/ygx/libraries/mds/molecules/molecules/conv_autoencoder/runtimes'
)
def train(output_filename, model_type, hidden_size, loss_type, norm_type,
sigma_noise):
train_data = torchvision.datasets.MNIST(
root='datasets/mnist/',
train=True,
transform=torchvision.transforms.ToTensor(),
download=False,
)
train_loader = Data.DataLoader(dataset=train_data,
batch_size=BATCH_SIZE,
shuffle=True)
if loss_type == 'l2':
loss_func = nn.MSELoss()
elif loss_type == 'cross_entropy':
loss_func = F.binary_cross_entropy
if model_type == 'AE':
model = AutoEncoder(hidden_size).cuda()
elif model_type == 'LTAE':
model = LatentAutoEncoder(hidden_size, norm_type,
sigma=sigma_noise).cuda()
model.set_device()
elif model_type == 'VAE':
model = VariationalAE(hidden_size).cuda()
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)
model.train()
for epoch in range(EPOCH):
for step, (x, _) in enumerate(train_loader):
optimizer.zero_grad()
x_batch = x.view(-1, 28 * 28).cuda()
y_batch = x.view(-1, 28 * 28).cuda()
if model_type == 'AE':
_, decoded = model(x_batch)
loss = loss_func(decoded, y_batch)
elif model_type == 'LTAE':
_, latent, transformed, decoded = model(x_batch)
loss = loss_func(decoded, y_batch)
loss += torch.nn.functional.mse_loss(transformed, latent)
elif model_type == 'VAE':
decoded, mu, logvar = model(x_batch)
loss = loss_func_vae(decoded, x_batch, mu, logvar, loss_type)
loss.backward()
optimizer.step()
if epoch % 10 == 0:
print('Epoch: ', epoch, '| train loss: %.4f' % loss.detach().cpu())
torch.save({'state_dict': model.state_dict()},
f'./saved_models/{output_filename}')
def __init__(self, path, k, lr= 0.1, batch_size=1, loss='bce', n_epochs=50):
'''
Arguments:
path : path to training data
k : hidde unit's dimension
lr : learning rate
batch_size : batch_size for training, currently set to 1
loss : loss function (bce, rmse) to train AutoEncoder
n_epochs : number of epochs for training
'''
self.AE = AutoEncoder(path, k)
# Definne the autoencoder model
#self.AE.model()
self.AE.model_batch()
self.epochs = n_epochs
def run_test():
model = AutoEncoder(n=2, dim_x=50, dim_hidden=2000)
device_id = 0
model_file = 'model/dae.pth'
net = load_pretrained_model(model, model_file, device_id)
x_test = pickle.load(open('data/manifold_test.pkl', 'rb'))
x_true = x_test[:]
noise_shape = x_true.shape[1:]
n_dim = np.prod(noise_shape)
sigma_train = 0.01
sigma_proposal = 0.01
sigma_noise_hat_init = 0.01
for noise in [0.01, 0.02, 0.03, 0.04]:
sigma_noise = [noise] * n_dim
sigma_noise = np.array(sigma_noise[:n_dim]).reshape(noise_shape)
rmse_mean, rmse_std, noise_mean, noise_std, variance_noise_hat_em = \
solve_agem(net=net, x_true=x_true, sigma_noise=sigma_noise,
sigma_train=sigma_train, sigma_noise_hat_init=sigma_noise_hat_init,
sigma_proposal=sigma_proposal, type_proposal='mala',
candidate='mean', em_epochs=10, sample_epochs=1000)
print('[AGEM] noise_gt: %.2f | rmse %.4f (%.4f), noise_est: %.4f (%.4f)' % (
noise, rmse_mean, rmse_std, noise_mean, noise_std
))
def build_net(self):
# define network
self.net = AutoEncoder(self.in_channel, self.image_size,
self.hidden_dim, self.output_dim)
if self.config.mode == 'test' and self.config.training_path == '':
print("[*] Enter model path!")
exit()
# if training model exists
if self.config.training_path != '':
self.net.load_state_dict(
torch.load(self.config.training_path,
map_location=lambda storage, loc: storage))
print("[*] Load weight from {}!".format(self.config.training_path))
self.net.to(self.device)
def generate(model_path,model_name, generate_path, generate_name, piece):
"""Synthesize audio from an array of embeddings.
Args:
encodings: Numpy array with shape [batch_size, time, dim].
save_paths: Iterable of output file names.
checkpoint_path: Location of the pretrained model. [model.ckpt-200000]
samples_per_save: Save files after every amount of generated samples.
"""
# Create directory for encoding
if os.path.exists(generate_path) is False:
os.makedirs(generate_path)
net = AutoEncoder()
net = load_model(net,model_path,model_name)
cuda_available = torch.cuda.is_available()
if cuda_available is True:
net = net.cuda()
net.eval()
# Load audio for encoding
piece = audio.load_wav(piece)
spec = audio.spectrogram(piece).astype(np.float32)
spec = torch.from_numpy(spec.T)
spec = torch.FloatTensor(spec)
spec = torch.unsqueeze(spec, 0)
spec = Variable(spec, volatile=True).contiguous()
if cuda_available is True:
spec = spec.cuda()
generated_spec = net(spec)
generated_spec = generated_spec.data.cpu().numpy()
generated_spec = np.squeeze(generated_spec)
waveform = audio.inv_spectrogram(generated_spec.T)
wav_name = generate_path + generate_name + '.wav'
audio.save_wav(waveform , wav_name)
def get_network(args_hiddens, logger=None):
"""
get autoencoder
:param args_hiddens: args.hiddens (comma separated)
:param logger: logger
:return: autoencoder model
"""
hidden_dims = [int(h) for h in args_hiddens.split(',')]
return AutoEncoder(hidden_dims, logger=logger)
def run(args):
# Create AutoEncoder
autoencoder = AutoEncoder(args['input_shape'],
args['z_dim'],
args['c_dim'],
learning_rate=args['learning_rate'])
# train
autoencoder.train(args['train_dir'], args['val_dir'], args['epochs'],
args['batch_size'], args['output_dir'])
# plot
x = autoencoder.sample_data()
plot_original(x, save_dir=args['output_dir'])
plot_reconstruction(x, autoencoder, save_dir=args['output_dir'])
plot_zvariation(x, autoencoder, save_dir=args['output_dir'])
plot_cvariation(x, autoencoder, save_dir=args['output_dir'])
plot_zsemireconstructed(x, autoencoder, save_dir=args['output_dir'])
plot_csemireconstructed(x, autoencoder, save_dir=args['output_dir'])
def gen_pic(ckpt, epoch):
device = torch.device('cuda')
model = AutoEncoder().to(device)
model.load_state_dict(torch.load(ckpt))
model.eval()
sample = torch.randn(64, 20).to(device)
sample = model.decoder(sample).cpu()
save_image(sample.view(64, 1, 28, 28),
'results/sample_e{:02}.png'.format(epoch))
def main(_):
create_dirs([FLAGS.summary_dir, FLAGS.checkpoint_dir])
config = auto_encoder_config()
model = AutoEncoder(config)
sess = tf.Session()
trainer = AutoEncoderTrainer(sess, model, FLAGS)
trainer.train()
def __init__(self, args):
self.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
self.id = args.id
self.dataset_path = args.dataset
model = AutoEncoder(args.E, args.D)
if args.load:
model.load_state_dict(torch.load(args.load))
self.model = model.to(self.device)
self.optimizer = optim.Adam(self.model.parameters(), lr=args.lr)
self.metric = nn.MSELoss()
self.epoch_num = args.epoch
self.batch_size = args.bs
self.cluster_num = args.cluster_num
self.save = args.save
self.testcase = args.testcase
self.csv = args.csv
self.record_file = None
self.best = {'epoch': 0, 'loss': 999}
self.test_images = get_test_image(args.dataset)
def main():
torch.manual_seed(1618)
device = 'cuda' if torch.cuda.is_available() else 'cpu'
print('Using PyTorch Device : {}'.format(device.upper()))
n_epochs = 800
LOGDIR = './runs/' + datetime.now().strftime('%Y%m%d_%H%M%S')
logger = Logger(log_dir=LOGDIR)
criterion = nn.MSELoss().to(device)
lr = 1e-4
model = AutoEncoder(connections=128).to(device)
optim = torch.optim.Adam(itertools.chain(model.encoder.parameters(),
model.decoder.parameters()),
lr=lr)
#model = AE_3D_200().to(device)
#optim = torch.optim.Adam(itertools.chain(model.encoder.parameters(), model.decoder.parameters()), lr=lr, weight_decay=1e-6)
train_batch, val_batch, test_batch = get_data_batches(device=device,
frac=1.0)
print(train_batch.size())
print(val_batch.size())
print(test_batch.size())
worst_case_loss = torch.FloatTensor([float('Inf')]).to(device)
pbar = tqdm(range(n_epochs), leave=True)
for e in pbar:
new_lr = lr * (0.2**((e + 1) // 100))
for param_group in optim.param_groups:
param_group['lr'] = new_lr
optim.zero_grad()
recon_batch = model(train_batch)
loss = criterion(recon_batch, train_batch)
loss.backward()
optim.step()
model.eval()
recon_val = model(val_batch)
val_loss = nn.MSELoss()(recon_val, val_batch)
recon_test = model(test_batch)
test_loss = nn.MSELoss()(recon_test, test_batch)
model.train()
info = {
'train_loss': loss.item(),
'val_loss': val_loss.item(),
'test_loss': test_loss.item()
}
for tag, value in info.items():
logger.scalar_summary(tag, value, e)
torch.save(model.encoder.state_dict(),
LOGDIR + '/encoder_epoch_{}.pt'.format(e))
torch.save(model.decoder.state_dict(),
LOGDIR + '/decoder_epoch_{}.pt'.format(e))
pbar.set_description(
'train_loss: {:.4f}, val_loss: {:.4f}, test_loss: {:.4f}'.format(
loss.item(), val_loss.item(), test_loss.item()))
def train(dataloader, parameters, device):
model = AutoEncoder(input_dim=1900,
nlayers=parameters.get('nlayers', 5),
latent=100)
model = model.to(device)
model.train()
train_loss = 0
optimizer = torch.optim.Adam(model.parameters(),
lr=parameters.get('lr', 1e-5),
weight_decay=parameters.get(
'weight_decay', 0.))
loss_func = torch.nn.MSELoss()
for epoch in range(parameters.get('epochs', 1000)):
for index, (data, ) in enumerate(dataloader, 1):
optimizer.zero_grad()
output = model(data)
loss = loss_func(output, data)
train_loss += loss.item()
loss.backward()
optimizer.step()
return model
def main(dataset, net_config, _run):
# Add all of the config into the helper class
for key in net_config:
setattr(a, key, net_config[key])
setattr(a, 'EXP_OUT', EXP_OUT)
setattr(a, 'RUN_id', _run._id)
output_dir = create_directories(_run._id, ex)
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.9)
with tf.Session(config=tf.ConfigProto(gpu_options=gpu_options)) as sess:
# load the dataset class
data = get_dataset(dataset['name'])
data = data(**dataset)
model = AutoEncoder(sess,
image_size=a.input_image_size,
batch_size=a.batch_size,
output_size=a.input_image_size,
dataset_name=dataset['name'],
checkpoint_dir=output_dir,
data=data,
momentum=a.batch_momentum,
aef_dim=a.naef,
noise_std_dev=a.noise_std_dev)
if a.mode == 'train':
tmp = model.train(a)
_run.info['predictions'] = tmp
_run.info['mean_predictions'] = np.mean(tmp, axis=0)
elif a.mode == 'valid':
tmp = model.validate(a)
_run.info['predictions'] = tmp
_run.info['mean_predictions'] = np.mean(tmp, axis=0)
else:
model.test(a)
def get_model(nNeuron, nLayer, in_dim, out_dim):
nNeuron = nNeuron + in_dim
mid1 = int((nNeuron + in_dim) / 2)
mid2 = int((nNeuron + out_dim) / 2)
mid = [nNeuron] * (nLayer - 2)
return ae.AutoEncoder(
encode_dim=[in_dim, mid1, *mid, mid2, out_dim],
decode_dim=[out_dim, mid2, mid1, in_dim],
activation=nn.LeakyReLU)
def main(args):
device = torch.device(
'cuda' if torch.cuda.is_available() and not args.cpu else 'cpu')
print('Using %s device.' % device)
world_size = int(
os.environ[args.env_size]) if args.env_size in os.environ else 1
local_rank = int(
os.environ[args.env_rank]) if args.env_rank in os.environ else 0
if local_rank == 0:
print(vars(args))
if world_size > 1:
print('rank: {}/{}'.format(local_rank + 1, world_size))
torch.distributed.init_process_group(backend='gloo',
init_method='file://%s' %
args.tmpname,
rank=local_rank,
world_size=world_size)
train_dataloader, test_dataloader = load_dataset(args, device, world_size)
net = AutoEncoder(input_dim=1900, nlayers=args.nlayers,
latent=100).to(device)
if world_size > 1:
net = torch.nn.parallel.DistributedDataParallel(net)
if args.modelfile:
net.load_state_dict(torch.load(args.modelfile))
# define our optimizer and loss function
optimizer = torch.optim.Adam(net.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
loss_func = nn.MSELoss(reduction='mean')
test_losses = []
for epoch in range(args.epochs):
epoch_start = timeit.default_timer()
train(train_dataloader, net, optimizer, loss_func, epoch)
test_loss = test(test_dataloader, net, loss_func)
print(' %5.2f sec' % (timeit.default_timer() - epoch_start))
test_losses.append(test_loss)
if test_loss <= min(test_losses):
torch.save(net.state_dict(), 'model/%5.3f.pth' % min(test_losses))
def main():
args = parser.parse_args()
filenames = glob.glob(os.path.join(args.dataset, '*.tfrecord'))
dataset = dataset_builder.build(filenames)
model = AutoEncoder()
optimizer = tf.keras.optimizers.Adam()
loss_object = tf.keras.losses.Huber(
reduction=tf.keras.losses.Reduction.NONE)
train_loss = tf.keras.metrics.Mean(name='train_loss')
ckpt = tf.train.Checkpoint(model=model)
ckpt_mgr = tf.train.CheckpointManager(ckpt, args.logdir, max_to_keep=5)
def train_step(origin, speaker_one_hot, target):
with tf.GradientTape() as tape:
convert = model(origin, speaker_one_hot)
loss = tf.reduce_mean(loss_object(convert, target))
gradients = tape.gradient(loss, model.trainable_variables)
optimizer.apply_gradients(zip(gradients, model.trainable_variables))
train_loss(loss)
EPOCHS = 1000
for epoch in range(EPOCHS):
# Reset the metrics at the start of the next epoch
train_loss.reset_states()
for feature in dataset:
speaker_id = feature['speaker_id']
speaker_one_hot = tf.one_hot(speaker_id, 2)
mel = feature['mel']
mel = tf.expand_dims(mel, axis=-1)
train_step(mel, speaker_one_hot, mel)
break
ckpt_mgr.save()
template = 'Epoch {}, Loss: {}'
print(template.format(epoch + 1, train_loss.result()))
ckpt_mgr.save()
def decode(model_name, encoding, decoder_name):
"""Synthesize audio from an array of embeddings.
Args:
encodings: Numpy array with shape [batch_size, time, dim].
save_paths: Iterable of output file names.
checkpoint_path: Location of the pretrained model. [model.ckpt-200000]
samples_per_save: Save files after every amount of generated samples.
"""
decoder_path = './decoding/'
model_path = './restore/'
# Create directory for encoding
if os.path.exists(decoder_path) is False:
os.makedirs(decoder_path)
net = AutoEncoder()
net = load_model(net,model_path,model_name)
cuda_available = torch.cuda.is_available()
if cuda_available is True:
net = net.cuda()
net.eval()
# Load Encoding
encoding_ndarray = np.load(encoding)
encoding = torch.from_numpy(encoding_ndarray).float()
encoding = Variable(encoding, volatile=True)
generated_spec = net.decoder(encoding)
generated_spec = generated_spec.data.cpu().numpy()
generated_spec = np.squeeze(generated_spec)
dec_name = decoder_path + decoder_name
np.save(dec_name , generated_spec)
def encode(spec):
model_path = './restore/'
model_name = 'Autoencoder1.model'
net = AutoEncoder()
net = load_model(net, model_path, model_name)
cuda_available = torch.cuda.is_available()
if cuda_available is True:
net = net.cuda()
net.eval()
spec = torch.FloatTensor(torch.from_numpy(spec))
spec = torch.unsqueeze(spec, 0)
spec = Variable(spec, volatile=True).contiguous()
if cuda_available is True:
spec = spec.cuda()
# Pass input audio to net forward pass
out = net.encoder(spec)
out = out.data.cpu().numpy()
out = np.squeeze(out)
return out
def encode(model_name, piece, encoding_name):
model_path = './restore/'
encoding_path = './encoding/'
# Create directory for encoding
if os.path.exists(encoding_path) is False:
os.makedirs(encoding_path)
net = AutoEncoder()
net = load_model(net,model_path,model_name)
cuda_available = torch.cuda.is_available()
if cuda_available is True:
net = net.cuda()
net.eval()
# Load audio for encoding
piece = audio.load_wav(piece)
spec = audio.spectrogram(piece).astype(np.float32)
spec = torch.from_numpy(spec.T)
spec = torch.FloatTensor(spec)
spec = torch.unsqueeze(spec, 0)
spec = Variable(spec, volatile=True).contiguous()
if cuda_available is True:
spec = spec.cuda()
# Pass input audio to net forward pass
encoding = net.encoder(spec)
encoding = encoding.data.cpu().numpy()
#encoding = np.squeeze(encoding)
encoding_ndarray = encoding_path + encoding_name+ '.npy'
np.save(encoding_ndarray, encoding)
if __name__ == '__main__':
opts = get_args()
if not os.path.exists(opts.outDir):
os.makedirs(opts.outDir)
# mnist digits dataset
transforms_ = torchvision.transforms.ToTensor()
traindata = torchvision.datasets.MNIST(root=opts.root, train=True,
transform=transforms_, download=True)
trainset = DataLoader(dataset=traindata, batch_size=opts.bs, shuffle=False)
# getting the structure of your autoencoder
autoencoder = AutoEncoder(opts.nz)
# useful when you have a GPU (link pytorch to CUDA)
if torch.cuda.is_available():
autoencoder.cuda()
print("test the data")
test_dataset(traindata=traindata, trainset=trainset)
print("begin training ...")
train(autoencoder=autoencoder, outDir=opts.outDir, trainset=trainset,
traindata=traindata)
print("visualize latent space representation")
visualize(autoencoder=autoencoder, outDir=opts.outDir, trainset=trainset,
traindata=traindata)
def main(eval_args):
# ensures that weight initializations are all the same
logging = utils.Logger(eval_args.local_rank, eval_args.save)
# load a checkpoint
logging.info('loading the model at:')
logging.info(eval_args.checkpoint)
checkpoint = torch.load(eval_args.checkpoint, map_location='cpu')
args = checkpoint['args']
if not hasattr(args, 'ada_groups'):
logging.info('old model, no ada groups was found.')
args.ada_groups = False
if not hasattr(args, 'min_groups_per_scale'):
logging.info('old model, no min_groups_per_scale was found.')
args.min_groups_per_scale = 1
if not hasattr(args, 'num_mixture_dec'):
logging.info('old model, no num_mixture_dec was found.')
args.num_mixture_dec = 10
logging.info('loaded the model at epoch %d', checkpoint['epoch'])
arch_instance = utils.get_arch_cells(args.arch_instance)
model = AutoEncoder(args, None, arch_instance)
# Loading is not strict because of self.weight_normalized in Conv2D class in neural_operations. This variable
# is only used for computing the spectral normalization and it is safe not to load it. Some of our earlier models
# did not have this variable.
model.load_state_dict(checkpoint['state_dict'], strict=False)
model = model.cuda()
logging.info('args = %s', args)
logging.info('num conv layers: %d', len(model.all_conv_layers))
logging.info('param size = %fM ', utils.count_parameters_in_M(model))
if eval_args.eval_mode == 'evaluate':
# load train valid queue
args.data = eval_args.data
train_queue, valid_queue, num_classes = datasets.get_loaders(args)
if eval_args.eval_on_train:
logging.info('Using the training data for eval.')
valid_queue = train_queue
# get number of bits
num_output = utils.num_output(args.dataset)
bpd_coeff = 1. / np.log(2.) / num_output
valid_neg_log_p, valid_nelbo = test(
valid_queue,
model,
num_samples=eval_args.num_iw_samples,
args=args,
logging=logging)
logging.info('final valid nelbo %f', valid_nelbo)
logging.info('final valid neg log p %f', valid_neg_log_p)
logging.info('final valid nelbo in bpd %f', valid_nelbo * bpd_coeff)
logging.info('final valid neg log p in bpd %f',
valid_neg_log_p * bpd_coeff)
else:
bn_eval_mode = not eval_args.readjust_bn
num_samples = 16
with torch.no_grad():
n = int(np.floor(np.sqrt(num_samples)))
set_bn(model,
bn_eval_mode,
num_samples=36,
t=eval_args.temp,
iter=500)
for ind in range(10): # sampling is repeated.
torch.cuda.synchronize()
start = time()
with autocast():
logits = model.sample(num_samples, eval_args.temp)
output = model.decoder_output(logits)
output_img = output.mean if isinstance(output, torch.distributions.bernoulli.Bernoulli) \
else output.sample()
torch.cuda.synchronize()
end = time()
output_tiled = utils.tile_image(output_img,
n).cpu().numpy().transpose(
1, 2, 0)
logging.info('sampling time per batch: %0.3f sec',
(end - start))
output_tiled = np.asarray(output_tiled * 255, dtype=np.uint8)
output_tiled = np.squeeze(output_tiled)
plt.imshow(output_tiled)
plt.show()
torch.backends.cudnn.deterministic = DEBUG
torch.backends.cudnn.benchmark = not DEBUG
print(f'Using device {DEVICE}')
# Data loaders
train_ds = PPFDataset(**TRAIN_DS_ARGS)
train_dl = DataLoader(train_ds, **TRAIN_DL_ARGS)
val_ds = PPFDataset(**VAL_DS_ARGS)
val_dl = DataLoader(val_ds, **VAL_DL_ARGS)
print('Training set: {} Validation set: {}\n'.format(
train_ds.__len__(), val_ds.__len__()
))
# Model
model = AutoEncoder(NUM_PTS_PER_PATCH)
model.apply(init_weights).to(DEVICE)
loss_func = ChamferLoss()
optimizer = Adam(model.parameters(), LR)
scheduler = OneCycleLR(
optimizer, MAX_LR, total_steps=len(train_dl)*TRAINER_ARGS.num_epochs
)
Path(TRAINER_ARGS.checkpoint_path).parent.mkdir(parents=True, exist_ok=True)
# Training
train(model, loss_func, optimizer, scheduler, (train_dl, val_dl), TRAINER_ARGS)
class trainAE(object):
def __init__(self,
path,
k,
lr=0.01,
batch_size=1,
loss='bce',
n_epochs=500):
'''
Arguments:
path : path to training data
k : hidde unit's dimension
lr : learning rate
batch_size : batch_size for training, currently set to 1
loss : loss function (bce, rmse) to train AutoEncoder
n_epochs : number of epochs for training
'''
self.AE = AutoEncoder(path, k)
# Definne the autoencoder model
#self.AE.model()
self.AE.model_batch()
self.epochs = n_epochs
def sigmoid(self, x):
return 1.0 / (1.0 + np.exp(-x))
def train(self):
T = self.AE.T
# COnverting to csr format for indexing
T = T.tocsr()
#pdb.set_trace()
nonzero_indices = T.nonzero()
for epoch in xrange(self.epochs):
print("Running epoch %d" % (epoch))
for i in np.unique(nonzero_indices[0]):
# get indices of observed values from the user 'i' 's vector
indices = T[i, :].nonzero()[1]
#print indices
#indices = indices.reshape(indices.shape[0],)
# Get correspoding ratings
ratings = T[i, indices].toarray()
#print ratings
ratings = ratings.reshape(ratings.shape[1], )
# Convert inputs to theano datatype
indices = indices.astype(np.int32)
ratings = ratings.astype(np.int32)
#pdb.set_trace()
loss = self.AE.ae(indices, ratings)
print("Loss at epoch %d is %f" % (epoch, loss))
print("RMSE after one epoch is %f" % (self.RMSE()))
# Batch training method
def train_batch(self, batch_size):
T = self.AE.T
T = T.tocsr()
nonzero_indices = T.nonzero()
#pdb.set_trace()
n_users = len(np.unique(nonzero_indices[0]))
indices = np.unique(nonzero_indices[0])
for epoch in xrange(self.epochs):
l = []
for ind, i in enumerate(xrange(0, n_users, batch_size)):
# CHECK : SEEMS BUGGY.
#------------------------
#ratings = T[indices[i:(i + batch_size)], :].toarray().astype(np.float32)
ratings = T[indices[i:(i + batch_size)], :].toarray().astype(
np.float32)
#------------------------
#print ratings
#pdb.set_trace()
loss = self.AE.ae_batch(ratings)
#loss = self.AE.debug(ratings)
#print loss
#pdb.set_trace()
l.append(loss)
m = np.mean(np.array(l))
print("mean Loss for epoch %d batch %d is %f" % (epoch, ind, m))
#rmse = self.RMSE_sparse()
rmse = self.RMSE()
print("RMSE after one epoch is %f" % (rmse))
f.write(str(rmse) + '\n')
def RMSE(self):
W, V, b, mu = self.AE.get_params()
print("testing process starts")
test = self.AE.test_ind[:5000]
rat = []
pred = []
rmse = 0
#try:
# Rt = self.AE.T[test[:,0]].todense()
# p = expit(np.dot(W, expit(np.dot(V, Rt) + mu)) + b)
# P = p[:, test[:,1]]
# pdb.set_trace()
#except:
# print "Exceptoin in batch test"
# pdb.set_trace()
for i, j in test:
Rt = self.AE.T[i, :].todense()
Rt1 = np.zeros(Rt.shape[1])
Rt1[:] = Rt[:]
#pdb.set_trace()
p = (np.dot(W, expit(np.dot(V, Rt1) + mu)) + b)
#p = np.tanh(np.dot(W, np.tanh(np.dot(V, Rt1) + mu)) + b)
p = p[j]
pred.append(p)
rat.append(self.AE.t[i, j])
try:
rat = np.array(rat)
pred = np.array(pred)
rmse = np.sqrt(np.mean((pred - rat)**2))
except:
print "exception"
pdb.set_trace()
np.save('test', test)
np.save('W', W)
np.save('V', V)
np.save('mu', mu)
np.save('b', b)
return rmse
#pdb.set_trace()
def RMSE_sparse(self):
W, V, b, mu = self.AE.get_params()
#pdb.set_trace()
mu = mu.reshape(100, 1)
print("testing process starts")
test = self.AE.test_ind
rat = []
pred = []
rmse = 0
for i, j in test:
#pdb.set_trace()
Rt = np.array(self.AE.T[i, :].todense().tolist()).astype(
np.float16)
#pdb.set_trace()
ind = np.where(Rt > 0)[1]
#pdb.set_trace()
Rt = Rt.T
#pdb.set_trace()
temp1 = V[:, ind]
temp2 = Rt[ind]
temp = expit(np.dot(temp1, temp2) + mu)
#del temp1
#del temp2
p = expit(np.dot(W, temp) + b)
#pdb.set_trace()
p = p[0, j]
pred.append(p)
rat.append(self.AE.t[i, j])
#gc.collect()
try:
rat = np.array(rat)
pred = np.array(pred)
rmse = np.sqrt(np.mean((pred - rat)**2))
except:
print "exception"
pdb.set_trace()
np.save('test', test)
np.save('W', W)
np.save('V', V)
np.save('mu', mu)
np.save('b', b)
return rmse
from torch.utils.data import TensorDataset, DataLoader
from model import AutoEncoder
import torch.nn as nn
import torch
import numpy as np
from tqdm import tqdm
model = AutoEncoder().cuda()
checkpoint = torch.load("autoencoder.pth")
criterion = nn.L1Loss()
model.load_state_dict(checkpoint)
model.eval()
with open("../agent_code/rule_based_agent_auto_encode/states.npy", "rb") as f:
data = np.load(f, allow_pickle=False)
data = data[:2_500_000]
data_t = torch.Tensor(data)
data_t = data_t.cuda()
dataset = TensorDataset(data_t)
train_len = int(0.7 * len(dataset))
valid_len = (len(dataset) - train_len) // 2
test_len = len(dataset) - train_len - valid_len
train_dataset, valid_dataset, test_dataset = torch.utils.data.random_split(
dataset, [train_len, valid_len, test_len],
generator=torch.Generator().manual_seed(42))
batch_size = 32
help='Field of view angle in radians')
parser.add_argument('-l', '--light', nargs='+', type=float,
default=[.5, 30],
help='Light intensity from poseur')
parser.add_argument('-oc', '--num_occlusions', default=2, type=int,
help='Number of occlusions')
parser.add_argument('-sa', '--save_path',
default=os.path.join(
os.path.expanduser('~'), '.keras/paz/models'),
type=str, help='Path for writing model weights and logs')
args = parser.parse_args()
# setting optimizer and compiling model
latent_dimension = args.latent_dimension
model = AutoEncoder((args.image_size, args.image_size, 3), latent_dimension)
optimizer = Adam(args.learning_rate, amsgrad=True)
model.compile(optimizer, args.loss, metrics=['mse'])
model.summary()
# setting scene
renderer = SingleView(
args.obj_path, (args.image_size, args.image_size),
args.y_fov, args.depth, args.sphere, args.roll,
args.translation, args.shift, args.light)
# creating sequencer
image_paths = glob.glob(os.path.join(args.images_directory, '*.png'))
processor = DomainRandomization(
renderer, args.image_size, image_paths, args.num_occlusions)
sequence = GeneratingSequence(processor, args.batch_size, args.steps_per_epoch)
def main(args):
# ensures that weight initializations are all the same
torch.manual_seed(args.seed)
np.random.seed(args.seed)
torch.cuda.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
logging = utils.Logger(args.global_rank, args.save)
writer = utils.Writer(args.global_rank, args.save)
# Get data loaders.
train_queue, valid_queue, num_classes, _ = datasets.get_loaders(args)
args.num_total_iter = len(train_queue) * args.epochs
warmup_iters = len(train_queue) * args.warmup_epochs
swa_start = len(train_queue) * (args.epochs - 1)
arch_instance = utils.get_arch_cells(args.arch_instance)
model = AutoEncoder(args, writer, arch_instance)
model = model.cuda()
logging.info('args = %s', args)
logging.info('param size = %fM ', utils.count_parameters_in_M(model))
logging.info('groups per scale: %s, total_groups: %d',
model.groups_per_scale, sum(model.groups_per_scale))
if args.fast_adamax:
# Fast adamax has the same functionality as torch.optim.Adamax, except it is faster.
cnn_optimizer = Adamax(model.parameters(),
args.learning_rate,
weight_decay=args.weight_decay,
eps=1e-3)
else:
cnn_optimizer = torch.optim.Adamax(model.parameters(),
args.learning_rate,
weight_decay=args.weight_decay,
eps=1e-3)
cnn_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(
cnn_optimizer,
float(args.epochs - args.warmup_epochs - 1),
eta_min=args.learning_rate_min)
grad_scalar = GradScaler(2**10)
num_output = utils.num_output(args.dataset, args)
bpd_coeff = 1. / np.log(2.) / num_output
# if load
checkpoint_file = os.path.join(args.save, 'checkpoint.pt')
if args.cont_training:
logging.info('loading the model.')
checkpoint = torch.load(checkpoint_file, map_location='cpu')
init_epoch = checkpoint['epoch']
model.load_state_dict(checkpoint['state_dict'])
model = model.cuda()
cnn_optimizer.load_state_dict(checkpoint['optimizer'])
grad_scalar.load_state_dict(checkpoint['grad_scalar'])
cnn_scheduler.load_state_dict(checkpoint['scheduler'])
global_step = checkpoint['global_step']
else:
global_step, init_epoch = 0, 0
for epoch in range(init_epoch, args.epochs):
# update lrs.
if args.distributed:
train_queue.sampler.set_epoch(global_step + args.seed)
valid_queue.sampler.set_epoch(0)
if epoch > args.warmup_epochs:
cnn_scheduler.step()
# Logging.
logging.info('epoch %d', epoch)
# Training.
train_nelbo, global_step = train(train_queue, model, cnn_optimizer,
grad_scalar, global_step,
warmup_iters, writer, logging)
logging.info('train_nelbo %f', train_nelbo)
writer.add_scalar('train/nelbo', train_nelbo, global_step)
model.eval()
# generate samples less frequently
eval_freq = 1 if args.epochs <= 50 else 20
if epoch % eval_freq == 0 or epoch == (args.epochs - 1):
with torch.no_grad():
num_samples = 16
n = int(np.floor(np.sqrt(num_samples)))
for t in [0.7, 0.8, 0.9, 1.0]:
logits = model.sample(num_samples, t)
output = model.decoder_output(logits)
output_img = output.mean if isinstance(
output, torch.distributions.bernoulli.Bernoulli
) else output.sample(t)
output_tiled = utils.tile_image(output_img, n)
writer.add_image('generated_%0.1f' % t, output_tiled,
global_step)
valid_neg_log_p, valid_nelbo = test(valid_queue,
model,
num_samples=10,
args=args,
logging=logging)
logging.info('valid_nelbo %f', valid_nelbo)
logging.info('valid neg log p %f', valid_neg_log_p)
logging.info('valid bpd elbo %f', valid_nelbo * bpd_coeff)
logging.info('valid bpd log p %f', valid_neg_log_p * bpd_coeff)
writer.add_scalar('val/neg_log_p', valid_neg_log_p, epoch)
writer.add_scalar('val/nelbo', valid_nelbo, epoch)
writer.add_scalar('val/bpd_log_p', valid_neg_log_p * bpd_coeff,
epoch)
writer.add_scalar('val/bpd_elbo', valid_nelbo * bpd_coeff, epoch)
save_freq = int(np.ceil(args.epochs / 100))
if epoch % save_freq == 0 or epoch == (args.epochs - 1):
if args.global_rank == 0:
logging.info('saving the model.')
torch.save(
{
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'optimizer': cnn_optimizer.state_dict(),
'global_step': global_step,
'args': args,
'arch_instance': arch_instance,
'scheduler': cnn_scheduler.state_dict(),
'grad_scalar': grad_scalar.state_dict()
}, checkpoint_file)
# Final validation
valid_neg_log_p, valid_nelbo = test(valid_queue,
model,
num_samples=1000,
args=args,
logging=logging)
logging.info('final valid nelbo %f', valid_nelbo)
logging.info('final valid neg log p %f', valid_neg_log_p)
writer.add_scalar('val/neg_log_p', valid_neg_log_p, epoch + 1)
writer.add_scalar('val/nelbo', valid_nelbo, epoch + 1)
writer.add_scalar('val/bpd_log_p', valid_neg_log_p * bpd_coeff, epoch + 1)
writer.add_scalar('val/bpd_elbo', valid_nelbo * bpd_coeff, epoch + 1)
writer.close()
class trainAE(object):
def __init__(self, path, k, lr= 0.1, batch_size=1, loss='bce', n_epochs=50):
'''
Arguments:
path : path to training data
k : hidde unit's dimension
lr : learning rate
batch_size : batch_size for training, currently set to 1
loss : loss function (bce, rmse) to train AutoEncoder
n_epochs : number of epochs for training
'''
self.AE = AutoEncoder(path, k)
# Definne the autoencoder model
#self.AE.model()
self.AE.model_batch()
self.epochs = n_epochs
def sigmoid(self, x):
return 1.0 / (1.0 + np.exp(-x))
def train(self):
T = self.AE.T
# COnverting to csr format for indexing
T = T.tocsr()
#pdb.set_trace()
nonzero_indices = T.nonzero()
for epoch in xrange(self.epochs):
print("Running epoch %d"%(epoch))
for i in np.unique(nonzero_indices[0]):
# get indices of observed values from the user 'i' 's vector
indices = T[i, :].nonzero()[1]
#print indices
#indices = indices.reshape(indices.shape[0],)
# Get correspoding ratings
ratings = T[i, indices].toarray()
#print ratings
ratings = ratings.reshape(ratings.shape[1],)
# Convert inputs to theano datatype
indices = indices.astype(np.int32)
ratings = ratings.astype(np.int32)
#pdb.set_trace()
loss = self.AE.ae(indices, ratings)
print("Loss at epoch %d is %f"%(epoch, loss))
print("RMSE after one epoch is %f"%(self.RMSE()))
# Batch training method
def train_batch(self, batch_size):
T = self.AE.T
T = T.tocsr()
nonzero_indices = T.nonzero()
#pdb.set_trace()
n_users = len(np.unique(nonzero_indices[0]))
indices = np.unique(nonzero_indices[0])
for epoch in xrange(self.epochs):
for ind, i in enumerate(xrange(0, n_users, batch_size)):
# CHECK : SEEMS BUGGY.
#------------------------
#ratings = T[indices[i:(i + batch_size)], :].toarray().astype(np.float32)
ratings = T[indices[i:(i + batch_size)], :].toarray().astype(np.float32)
#------------------------
#print ratings
#pdb.set_trace()
loss = self.AE.ae_batch(ratings)
#loss = self.AE.debug(ratings)
#print loss
#pdb.set_trace()
print("Loss for epoch %d batch %d is %f"%(epoch, ind, loss))
print("RMSE after one epoch is %f"%(self.RMSE()))
def RMSE(self):
W, V, b, mu = self.AE.get_params()
print("testing process starts")
test = self.AE.test_ind
rat = []
pred = []
rmse = 0
for i,j in test:
Rt = self.AE.T[i, :].todense()
Rt1 = np.zeros(Rt.shape[1] +1)
Rt1[:6541] = Rt
#pdb.set_trace()
p = expit(np.dot(W, expit(np.dot(V, Rt1) + mu)) + b)
#p = np.tanh(np.dot(W, np.tanh(np.dot(V, Rt1) + mu)) + b)
p = p[j]
pred.append(p)
rat.append(self.AE.t[i, j])
try:
rat = np.array(rat)
pred = np.array(pred)
rmse = np.sqrt(np.mean((pred-rat)**2))
except:
print "exception"
pdb.set_trace()
np.save('test', test)
np.save('W', W)
np.save('V', V)
np.save('mu', mu)
np.save('b', b)
return rmse
#pdb.set_trace()
def RMSE_sparse(self):
W, V, mu, b = self.AE.get_params()
print("testing process starts")
test = self.AE.test_ind
rat = []
pred = []
for i,j in test:
Rt = self.AE.T[i, :].todense()
#pdb.set_trace()
ind = self.AE.T[i, :].nonzero()[1]
#pdb.set_trace()
Rt = Rt.T
temp = np.tanh(np.dot(V[:, ind], Rt[ind]) + mu.reshape(100,1))
p = np.tanh(np.dot(W, temp) + b)
p = p[j]
pred.append(p)
rat.append(self.AE.t[i, j])
try:
rat = np.array(rat)
pred = np.array(pred)
print np.sqrt(np.mean((pred-rat)*2))
except:
print "exception"
pdb.set_trace()
np.save('test', test)
np.save('W', W)
np.save('V', V)
np.save('mu', mu)
np.save('b', b)
pdb.set_trace()