def __init__(self, config):
self.config = config
if torch.cuda.is_available():
self.use_cuda = True
torch.set_default_tensor_type('torch.cuda.FloatTensor')
else:
self.use_cuda = False
torch.set_default_tensor_type('torch.FloatTensor')
self.nz = config.nz
self.optimizer = config.optimizer
self.resl = 2 # we start from 2^2 = 4
self.max_resl = config.max_resl
self.trns_tick = config.trns_tick
self.stab_tick = config.stab_tick
self.TICK = config.TICK
self.globalIter = 0
self.globalTick = 0
self.kimgs = 0
self.stack = 0
self.epoch = 0
self.fadein = {'gen': None, 'dis': None}
self.complete = {'gen': 0, 'dis': 0}
self.phase = 'init'
self.flag_flush_gen = False
self.flag_flush_dis = False
self.trns_tick = self.config.trns_tick
self.stab_tick = self.config.stab_tick
# network and cirterion
self.G = net.Generator(config)
self.D = net.Discriminator(config)
print('Generator structure: ')
print(self.G.model)
print('Discriminator structure: ')
print(self.D.model)
self.mse = torch.nn.MSELoss()
if self.use_cuda:
self.mse = self.mse.cuda()
torch.cuda.manual_seed(config.random_seed)
if config.n_gpu == 1:
#self.G = self.G.cuda()
#self.D = self.D.cuda() # It seems simply call .cuda() on the model does not function. PyTorch bug when we use modules.
self.G = torch.nn.DataParallel(self.G).cuda(device_id=0)
self.D = torch.nn.DataParallel(self.D).cuda(device_id=0)
else:
gpus = []
for i in range(config.n_gpu):
gpus.append(i)
self.G = torch.nn.DataParallel(self.G, device_ids=gpus).cuda()
self.D = torch.nn.DataParallel(self.D, device_ids=gpus).cuda()
# define tensors, and get dataloader.
self.renew_everything()
# tensorboard
self.use_tb = config.use_tb
if self.use_tb:
self.tb = tensorboard.tf_recorder()
def __init__(self,config):
self.config=config
if torch.cuda.is_available():
self.use_cuda=True
torch.set_default_tensor_type('torch.cuda.FloatTensor')
else:
self.use_cuda=False
torch.set_default_tensor_type('torch.FloatTensor')
self.nz = config.nz
self.optimizer = config.optimizer
self.resl = 2 # we start from 2^2 = 4
self.lr = config.lr
self.eps_drift = config.eps_drift
self.smoothing = config.smoothing
self.max_resl = config.max_resl
self.trns_tick = config.trns_tick
self.stab_tick = config.stab_tick
self.TICK = config.TICK
self.globalIter = 0
self.globalTick = 0
self.kimgs = 0
self.stack = 0
self.epoch = 0
self.fadein = {'gen': None, 'dis': None}
self.complete = {'gen': 0, 'dis': 0}
self.phase = 'init'
self.flag_flush_gen = False
self.flag_flush_dis = False
self.flag_add_noise = self.config.flag_add_noise
self.flag_add_drift = self.config.flag_add_drift
self.loader=DL.dataloader(config)
self.LAMBDA=2
# network
self.G = network.Generator(config)
self.D = network.Discriminator(config)
print('Generator structure: ')
print(self.G.model)
print('Discriminator structure: ')
print(self.D.model)
if self.use_cuda:
torch.cuda.manual_seed(config.random_seed)
#self.G = self.G.cuda()
#self.D = self.D.cuda()
if config.n_gpu==1:
self.G=torch.nn.DataParallel(self.G).cuda(device=0)
self.D=torch.nn.DataParallel(self.D).cuda(device=0)
else:
gpus=[]
for i in range(config.n_gpu):
gpus.append(i)
self.G=torch.nn.DataParallel(self.G,device_ids=gpus).cuda()
self.D=torch.nn.DataParallel(self.D,device_ids=gpus).cuda()
self.renew_everything()
self.use_tb=config.use_tb
if self.use_tb:
self.tb=tensorboard.tf_recorder()
def __init__(self, config, loader, path):
self.config = config
if torch.cuda.is_available():
self.use_cuda = True
torch.set_default_tensor_type('torch.cuda.FloatTensor')
else:
self.use_cuda = False
torch.set_default_tensor_type('torch.FloatTensor')
self.nz = config.nz
self.resl = config.resl # we start from 2^2 = 4
self.lr = config.lr
self.eps_drift = config.eps_drift
self.smoothing = config.smoothing
self.max_resl = config.max_resl
self.trns_tick = config.trns_tick
self.stab_tick = config.stab_tick
self.TICK = config.TICK
self.globalIter = 0
self.globalTick = config.globalTick
self.kimgs = 0
self.stack = 0
self.epoch = 0
self.fadein = {'gen': None, 'dis': None}
self.complete = {'gen': 0, 'dis': 0}
self.phase = 'init'
self.flag_flush_gen = False
self.flag_flush_dis = False
self.flag_add_noise = self.config.flag_add_noise
self.flag_add_drift = self.config.flag_add_drift
self.load = config.load
self.batch = 1
# network and cirterion
self.D = net.Discriminator(config)
if self.use_cuda:
if config.n_gpu == 1:
self.D = torch.nn.DataParallel(self.D).cuda(device=0)
else:
gpus = []
for i in range(config.n_gpu):
gpus.append(i)
self.D = torch.nn.DataParallel(self.D, device_ids=gpus).cuda()
#self.renew_everything()
if self.load:
self.load_snapshot('repo/model')
self.renew_everything()
if self.use_cuda:
if config.n_gpu == 1:
self.D.cuda(device=0)
self.loader = loader
self.path = path
def __init__(self, config):
self.config = config
if torch.cuda.is_available():
self.use_cuda = True
torch.set_default_tensor_type('torch.cuda.FloatTensor')
else:
self.use_cuda = False
torch.set_default_tensor_type('torch.FloatTensor')
self.nz = config.nz
self.optimizer = config.optimizer
self.resl = config.start_res # we start from 2^2 = 4
self.lr = config.lr
self.eps_drift = config.eps_drift
self.smoothing = config.smoothing
self.max_resl = config.max_resl
self.trns_tick = config.trns_tick
self.stab_tick = config.stab_tick
self.TICK = config.TICK
self.globalIter = 0
self.globalTick = 0
self.kimgs = 0
self.stack = 0
self.epoch = 0
self.fadein = {'gen': None, 'dis': None}
self.complete = {'gen': 0, 'dis': 0}
self.phase = 'init'
self.flag_flush_gen = False
self.flag_flush_dis = False
self.flag_add_noise = self.config.flag_add_noise
self.flag_add_drift = self.config.flag_add_drift
# network and criterion
if config.start_res == 2:
self.G = net.Generator(config)
self.D = net.Discriminator(config)
else:
self.G, self.D = g_d_interpolated.recup_nets(config)
self.mse = torch.nn.MSELoss()
if self.use_cuda:
self.mse = self.mse.cuda()
torch.cuda.manual_seed(config.random_seed)
"""
if config.n_gpu==1:
self.G = torch.nn.DataParallel(self.G).cuda(device=0)
self.D = torch.nn.DataParallel(self.D).cuda(device=0)
else:
gpus = []
for i in range(config.n_gpu):
gpus.append(i)
self.G = torch.nn.DataParallel(self.G, device_ids=gpus).cuda()
self.D = torch.nn.DataParallel(self.D, device_ids=gpus).cuda()
"""
# define tensors, ship model to cuda, and get dataloader.
self.renew_everything()
def __init__(self, sample_dimension, noise_dimension, bs):
self.sample_dimension = sample_dimension
self.noise_dimension = noise_dimension
self.bs = bs
self.D = network.Discriminator(sample_dimension=sample_dimension)
self.G = network.Generator(noise_dimension=noise_dimension,
sample_dimension=sample_dimension)
self.optimizer_d = torch.optim.RMSprop(self.D.parameters())
self.optimizer_g = torch.optim.RMSprop(self.G.parameters())
self.criterion = nn.BCELoss()
def __init__(self, config):
self.config = config
if torch.cuda.is_available():
self.use_cuda = True
torch.set_default_tensor_type('torch.cuda.FloatTensor')
else:
self.use_cuda = False
torch.set_default_tensor_type('torch.FloatTensor')
self.nz = config.nz
self.optimizer = config.optimizer
self.resolution = 2 # we start from 2^2 = 4
self.lr = config.lr
self.eps_drift = config.eps_drift
self.smoothing = config.smoothing
self.max_resolution = config.max_resolution
self.transition_tick = config.transition_tick
self.stablize_tick = config.stablize_tick
self.TICK = config.TICK
self.globalIter = 0
self.globalTick = 0
self.kimgs = 0
self.stack = 0
self.epoch = 0
self.fadein = {'gen': None, 'dis': None}
self.complete = {'gen': 0, 'dis': 0}
self.phase = 'init'
self.flag_flush_gen = False
self.flag_flush_dis = False
self.flag_add_noise = self.config.flag_add_noise
self.flag_add_drift = self.config.flag_add_drift
# network and cirterion
self.G = nn.DataParallel(net.Generator(config))
self.D = nn.DataParallel(net.Discriminator(config))
print('Generator structure: ')
print(self.G.module.model)
print('Discriminator structure: ')
print(self.D.module.model)
self.mse = torch.nn.MSELoss()
self.renew_everything()
# tensorboard
self.use_tb = config.use_tb
if self.use_tb:
self.tb = tensorboard.tf_recorder()
if config.pretrained is not None:
self.load_pretrained(config.pretrained)
def __init__(self, opt):
super(TrainModel, self).__init__()
self.isTrain = opt.isTrain
if self.isTrain:
self.netE = network.Encoder(opt.input_nc, opt.ngf,
opt.n_downsampling)
self.netE.apply(network.weights_init)
self.netG = network.Decoder(opt.input_nc, opt.output_nc, opt.ngf,
opt.n_downsampling)
self.netG.apply(network.weights_init)
self.netD = network.Discriminator(opt.input_nc, opt.ngf,
opt.n_layer)
self.netD.apply(network.weights_init)
self.criterionGAN = nn.BCELoss()
self.criterionKL = network.KLLoss
self.criterionRecon = network.ReconLoss
else:
pass
def recup_nets(config):
use_cuda = True
checkpoint_path_g = config.checkpoint_generator
checkpoint_path_d = config.checkpoint_discriminator
# load trained model.
model_g = net.Generator(config)
model_d = net.Discriminator(config)
if use_cuda:
torch.set_default_tensor_type('torch.cuda.FloatTensor')
model_g = torch.nn.DataParallel(model_g).cuda(device=0)
model_d = torch.nn.DataParallel(model_d).cuda(device=0)
else:
torch.set_default_tensor_type('torch.FloatTensor')
for resl in range(3, config.start_res + 1):
model_g.module.grow_network(resl)
model_d.module.grow_network(resl)
model_g.module.flush_network()
model_d.module.flush_network()
print('generator :')
print(model_g)
print('discriminator :')
print(model_d)
print('load generator from checkpoint ... {}'.format(checkpoint_path_g))
print(
'load discriminator from checkpoint ... {}'.format(checkpoint_path_d))
checkpoint_g = torch.load(os.path.join('repo/model', checkpoint_path_g))
checkpoint_d = torch.load(os.path.join('repo/model', checkpoint_path_d))
print(type(checkpoint_g['state_dict']))
print(type(checkpoint_d['state_dict']))
model_g.module.load_state_dict(checkpoint_g['state_dict'], False)
model_d.module.load_state_dict(checkpoint_d['state_dict'], False)
return model_g, model_d
def extract_feature(path):
args = parser()
dis = network.Discriminator(args.depth)
print('Loading Discirminator Model from ' + args.dis)
serializers.load_npz(args.dis, dis)
img_list = os.listdir(args.img)
img_list = sorted(img_list)
if img_list[0] == '.DS_Store':
del img_list[0]
IMG_PATH = [os.path.join(args.img, name) for name in img_list]
for i in tqdm(range(args.num)):
im = get_example(os.path.join(args.img, IMG_PATH[i]), args.depth, None)
im = im[np.newaxis, :, :, :]
out_dis, _ = dis(im, alpha=1.0)
out_dis = np.ravel(out_dis.data)
try:
output = np.vstack((output, out_dis))
except:
output = out_dis
return output
else:
train10, train20, label = iterator.get_next()
# set up learning rate.
'''
global_step_generator = tf.Variable(0, trainable=False)
global_step_discriminator = tf.Variable(0, trainable=False)
lr_discriminator = tf.train.exponential_decay(args.discrimator_learning_rate,
global_step_discriminator, 8e4, 0.5, staircase=False)
lr_generator = tf.train.exponential_decay(args.generator_learning_rate,
global_step_generator, 4e4, 0.5, staircase=False)
'''
# Set up models
discriminator = network.Discriminator(loss_type=args.gan_type,
image_size=args.patch_size,
batch_size=args.batch_size,
norm=args.norm,
run_60=args.run_60)
generator = network.Generator(adversarial_loss=args.gan_type,
content_loss=args.contentloss,
batch_size=args.batch_size,
discriminator=discriminator,
norm=False,
adv_weight=args.adv_weight,
relu=args.relu,
run_60=args.run_60)
# Generator
if args.run_60:
g_y_pred = generator.forward(train10, train20, train60)
else:
def __init__(self, config):
self.config = config
if torch.cuda.is_available():
self.use_cuda = True
torch.set_default_tensor_type('torch.cuda.FloatTensor')
else:
self.use_cuda = False
torch.set_default_tensor_type('torch.FloatTensor')
self.nz = config.nz
self.optimizer = config.optimizer
self.resl = 2 # we start from 2^2 = 4
self.lr = config.lr
self.eps_drift = config.eps_drift
self.smoothing = config.smoothing
self.max_resl = config.max_resl
self.trns_tick = config.trns_tick
self.stab_tick = config.stab_tick
self.TICK = config.TICK
self.globalIter = 0
self.globalTick = 0
self.kimgs = 0
self.stack = 0
self.epoch = 0
self.fadein = {'gen':None, 'dis':None}
self.complete = {'gen':0, 'dis':0}
self.phase = 'init'
self.flag_flush_gen = False
self.flag_flush_dis = False
self.flag_add_noise = self.config.flag_add_noise
self.flag_add_drift = self.config.flag_add_drift
self.use_captions = config.use_captions
self.gan_type = config.gan_type
self.lambda = config.lambda
if self.use_captions:
self.ncap = config.ncap
# network and cirterion
self.G = net.Generator(config, use_captions=self.use_captions)
self.D = net.Discriminator(config, use_captions=self.use_captions)
print ('Generator structure: ')
print(self.G.model)
print ('Discriminator structure: ')
print(self.D.model)
if self.gan_type == 'lsgan':
self.mse = torch.nn.MSELoss()
if self.use_cuda:
if self.gan_type == 'lsgan':
self.mse = self.mse.cuda()
torch.cuda.manual_seed(config.random_seed)
if config.n_gpu==1:
# if only on single GPU
self.G = torch.nn.DataParallel(self.G).cuda(device=0)
self.D = torch.nn.DataParallel(self.D).cuda(device=0)
else:
# if more we're doing multiple GPUs
gpus = []
for i in range(config.n_gpu):
gpus.append(i)
self.G = torch.nn.DataParallel(self.G, device_ids=gpus).cuda()
self.D = torch.nn.DataParallel(self.D, device_ids=gpus).cuda()
# define tensors, ship model to cuda, and get dataloader.
self.renew_everything()
# tensorboard
self.use_tb = config.use_tb
if self.use_tb:
self.tb = tensorboard.tf_recorder()
# construct DataLoader list
if cuda:
torch_dataset = torch.utils.data.TensorDataset(
torch.FloatTensor(gene_exp).cuda(), torch.LongTensor(labels).cuda())
else:
torch_dataset = torch.utils.data.TensorDataset(
torch.FloatTensor(gene_exp), torch.LongTensor(labels))
data_loader = torch.utils.data.DataLoader(torch_dataset, batch_size=batch_size,
shuffle=True, drop_last=True)
batch_loader_dict[i+1] = data_loader
# create model
encoder = models.Encoder(num_inputs=num_inputs)
decoder_a = models.Decoder_a(num_inputs=num_inputs)
decoder_b = models.Decoder_b(num_inputs=num_inputs)
discriminator = models.Discriminator(num_inputs=num_inputs)
if cuda:
encoder.cuda()
decoder_a.cuda()
decoder_b.cuda()
discriminator.cuda()
# training
loss_total_list = [] # list of total loss
loss_reconstruct_list = []
loss_transfer_list = []
loss_classifier_list = []
for epoch in range(1, num_epochs + 1):
log_interval = config.log_interval
base_lr = config.base_lr
print(opt)
img_shape = (opt.channels, opt.img_size, opt.img_size)
cuda = True if torch.cuda.is_available() else False
# Loss weight for gradient penalty
lambda_gp = 10
# Initialize generator and discriminator
generator = network.Generator(image_size=opt.img_size,
z_dim=opt.latent_dim,
conv_dim=opt.conv_dim,
selfattn=True)
discriminator = network.Discriminator(image_size=opt.img_size,
conv_dim=opt.conv_dim,
selfattn=True)
if cuda:
generator.cuda()
discriminator.cuda()
# Configure data loader
dataloader = DataLoader(datasets.CIFAR10(
train=True,
download=True,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
]),
),
def __init__(self,
timesteps=1,
nb_random_samples=10,
batch_size=1,
epochs=30,
lr=1e-3,
in_channels=1,
channels=10,
depth=1,
kernel_size=3,
reduced_size=10,
out_channels=10,
negative_penalty=1,
n_clusters=1,
n_init=10, **kwargs):
"""
EUDTR: Enhanced Unsupervised Deep Temporal Representation learning model
Args:
timesteps: Length of time series.
nb_random_samples: Number of randomly chosen intervals to select the
positiva and negative sample in the loss.
batch_size: Batch size used during the training of the network.
epochs: Number of optimization steps to perform for the training of
the network.
lr: Learning rate of the Adam optimizer used to train the network.
in_channels: Number of input channels.
channels: Number of channels manipulated in the causal CNN.
depth: Depth of the causal CNN.
kernel_size: Kernel size of the applied non-residual convolutions.
reduced_size: Fixed length to which the output time series of the
causal CNN is reduced.
out_channels: Number of features in the final output.
negative_penalty: Multiplicative coefficient for the negative sample
loss.
n_clusters: Number of clusters.
n_init: Number of time the k-means algorithm will be run with different
centroid seeds.
"""
self.encoder = network.CausalCNNEncoder(in_channels, channels, depth, reduced_size, out_channels, kernel_size)
self.decoder = network.Discriminator(timesteps, out_channels, in_channels, reduced_size, kernel_size, stride=1, padding=1, out_padding=0)
self.device = "cuda" if torch.cuda.is_available() else "cpu"
self.encoder = self.encoder.double().to(self.device)
self.decoder = self.decoder.double().to(self.device)
self.mi_loss = reconstruction_loss.MILoss(self.device).to(self.device)
self.batch_size = batch_size
self.epochs = epochs
self.encoder_optimizer = torch.optim.Adam(self.encoder.parameters(), lr=lr)
self.decoder_optimizer = torch.optim.Adam(self.decoder.parameters(), lr=lr)
self.negative_penalty = negative_penalty
self.out_channels = out_channels
self.n_clusters = n_clusters
self.n_init = n_init
self.nb_random_samples = nb_random_samples
def __init__(self, config):
self.config = config
if torch.cuda.is_available():
self.use_cuda = True
torch.set_default_tensor_type('torch.cuda.FloatTensor')
tqdm.write('Using GPU.')
else:
self.use_cuda = False
torch.set_default_tensor_type('torch.FloatTensor')
self.nz = config.nz
self.optimizer = config.optimizer
self.minibatch_repeat = 4
self.resl = 2 # we start from 2^2 = 4
self.lr = config.lr
self.eps_drift = config.eps_drift
self.smoothing = config.smoothing
self.max_resl = config.max_resl
self.trns_tick = config.trns_tick
self.stab_tick = config.stab_tick
self.TICK = config.TICK
self.globalIter = 0
self.globalTick = 0
self.kimgs = 0
self.stack = 0
self.epoch = 0
self.fadein = {'gen':None, 'dis':None}
self.complete = {'gen':0, 'dis':0}
self.phase = 'init'
self.flag_flush_gen = False
self.flag_flush_dis = False
self.flag_add_noise = self.config.flag_add_noise
self.flag_add_drift = self.config.flag_add_drift
self.flag_wgan = self.config.flag_wgan
# network and cirterion
self.G = net.Generator(config)
self.Gs = net.Generator(config)
self.D = net.Discriminator(config)
self.mse = torch.nn.MSELoss()
if self.use_cuda:
self.mse = self.mse.cuda()
torch.cuda.manual_seed(int(time.time()))
if config.n_gpu==1:
self.G = torch.nn.DataParallel(self.G).cuda(device=0)
self.Gs = torch.nn.DataParallel(self.Gs).cuda(device=0)
self.D = torch.nn.DataParallel(self.D).cuda(device=0)
else:
gpus = []
for i in range(config.n_gpu):
gpus.append(i)
self.G = torch.nn.DataParallel(self.G, device_ids=gpus).cuda()
self.D = torch.nn.DataParallel(self.D, device_ids=gpus).cuda()
self.gen_ckpt = config.gen_ckpt
self.gs_ckpt = config.gs_ckpt
self.dis_ckpt = config.dis_ckpt
if self.gen_ckpt != '' and self.dis_ckpt != '':
pattern = '{}gen_R{}_T{}.pth.tar'
parsed = parse(pattern, self.gen_ckpt)
restore_resl = int(parsed[1])
restore_tick = int(parsed[2])
# Restore the network structure.
for resl in xrange(3, restore_resl+1):
self.G.module.grow_network(resl)
self.Gs.module.grow_network(resl)
self.D.module.grow_network(resl)
if resl < restore_resl:
self.G.module.flush_network()
self.Gs.module.flush_network()
self.D.module.flush_network()
# for _ in xrange(int(self.resl), restore_resl):
# self.lr = self.lr * float(self.config.lr_decay)
print(
"Restored resolution", restore_resl,
"Restored global tick", restore_tick,
"Restored learning rate", self.lr)
self.resl = restore_resl
self.globalTick = restore_tick
# Restore the network setting.
if self.resl != 2:
self.phase = 'stab'
# define tensors, ship model to cuda, and get dataloader.
self.renew_everything()
if self.gen_ckpt != '' and self.dis_ckpt != '':
self.G.module.flush_network()
self.Gs.module.flush_network()
self.D.module.flush_network()
self.globalIter = floor(self.globalTick * self.TICK / (self.loader.batchsize * self.minibatch_repeat))
gen_ckpt = torch.load(self.gen_ckpt)
gs_ckpt = torch.load(self.gs_ckpt)
dis_ckpt = torch.load(self.dis_ckpt)
self.opt_d.load_state_dict(dis_ckpt['optimizer'])
self.opt_g.load_state_dict(gen_ckpt['optimizer'])
print('Optimizer restored.')
self.resl = gen_ckpt['resl']
self.G.module.load_state_dict(gen_ckpt['state_dict'])
self.Gs.module.load_state_dict(gs_ckpt['state_dict'])
self.D.module.load_state_dict(dis_ckpt['state_dict'])
print('Model weights restored.')
gen_ckpt = None
dis_ckpt = None
gs_ckpt = None
print ('Generator structure: ')
print(self.G)
print ('Discriminator structure: ')
print(self.D)
# tensorboard
self.use_tb = config.use_tb
if self.use_tb:
self.tb = tensorboard.tf_recorder()
def train(enhancer, mode, param):
# create initial discriminator
disc = network.Discriminator(param['discriminator-size'])
assert disc.channels == enhancer.discriminator.channels
seed_size = param['image-size'] // param['zoom']
images = np.zeros(
(param['batch-size'], 3, param['image-size'], param['image-size']),
dtype=np.float32)
seeds = np.zeros((param['batch-size'], 3, seed_size, seed_size),
dtype=np.float32)
loader.copy(images, seeds)
# initial lr
lr = network.decay_learning_rate(param['learning-rate'], 75, 0.5)
# optimizer for generator
opt_gen = optim.Adam(enhancer.generator.parameters(), lr=0)
opt_disc = optim.Adam(disc.parameters(), lr=0)
try:
average, start = None, time.time()
for epoch in range(param['epochs']):
adversary_weight = 5e2
total, stats = None, None
l_r = next(lr)
if epoch >= param['generator-start']:
network.update_optimizer_lr(opt_gen, l_r)
if epoch >= param['discriminator-start']:
network.update_optimizer_lr(opt_disc, l_r)
for step in range(param['epoch-size']):
enhancer.zero_grad()
disc.zero_grad()
loader.copy(images, seeds)
# run full network once
gen_out, c12, c22, c32, c52, disc_out = enhancer(images, seeds)
# clone discriminator on the full network
enhancer.clone_discriminator_to(disc)
# output of new cloned network (maybe you can assert it to
# equal disc_out)
disc_out2 = disc(c12.detach(), c22.detach(), c32.detach())
disc_out_numpy = disc_out2.data.cpu().numpy(
) if torch.cuda.is_available() else disc_out2.data.numpy()
disc_out_mean = np.mean(disc_out_numpy, axis=(1, 2, 3))
stats = stats + disc_out_mean if stats is not None else disc_out_mean
# compute generator loss
if mode == 'pretrain':
gen_loss = network.loss_perceptual(c22[:param['batch-size']], c22[param['batch-size']:]) * param['perceptual-weight'] \
+ network.loss_total_variation(gen_out) * param['smoothness-weight'] \
+ network.loss_adversarial(disc_out[param['batch-size']:]) * adversary_weight
else:
gen_loss = network.loss_perceptual(c52[:param['batch-size']], c52[param['batch-size']:]) * param['perceptual-weight'] \
+ network.loss_total_variation(gen_out) * param['smoothness-weight'] \
+ network.loss_adversarial(disc_out[param['batch-size']:]) * adversary_weight
# compute discriminator loss
disc_loss = network.loss_discriminator(
disc_out2[:param['batch-size']],
disc_out2[param['batch-size']:])
gen_loss_data = gen_loss.data.cpu().numpy(
) if torch.cuda.is_available() else gen_loss.data.numpy()
total = total + gen_loss_data if total is not None else gen_loss_data
average = gen_loss_data if average is None else average * \
0.95 + 0.05 * gen_loss_data
print('↑' if gen_loss_data > average else '↓',
end='',
flush=True)
# update parameters step
gen_loss.backward()
disc_loss.backward()
torch.nn.utils.clip_grad_norm(disc.parameters(), 5)
opt_gen.step()
opt_disc.step()
# rebuild real discriminator from clone
enhancer.assign_back_discriminator(disc)
total /= param['epoch-size']
stats /= param['epoch-size']
print('\nOn Epoch: ' + str(epoch))
print('Generator Loss: ')
print(total)
real, fake = stats[:param['batch-size']], stats[
param['batch-size']:]
print(' - discriminator', real.mean(),
len(np.where(real > 0.5)[0]), fake.mean(),
len(np.where(fake < -0.5)[0]))
if epoch == param['adversarial-start'] - 1:
print(' - generator now optimizing against discriminator.')
adversary_weight = param['adversary-weight']
# Then save every several epochs
if epoch % 10 == 0:
torch.save(enhancer.state_dict(),
'model/model_' + mode + '.pth')
torch.save(enhancer.state_dict(), 'model/model_' + mode + '.pth')
print("Training ends after: " + str(float(time.time() - start)) +
" seconds")
except KeyboardInterrupt:
pass
def __init__(self, config):
self.config = config
# tensorboard
self.use_tb = config.use_tb
self.tb = None
# dir setup
self.save_dir = "log/"
if len(os.listdir(self.save_dir)) == 0:
os.mkdir(self.save_dir + 'act_{:05}'.format(0))
for i in range(1, 1000):
if "act_{:05}".format(i) in os.listdir(self.save_dir):
i += 1
continue
self.save_dir += "act_{:05}/".format(i)
os.mkdir(self.save_dir)
conf = self.save_dir + "conf.txt"
with open(conf, mode='w') as f:
for k, v in vars(config).items():
f.writelines('{} : {} \n'.format(k, v))
if self.use_tb:
self.tb = tensorboard.tf_recorder(config)
break
if not os.path.exists(self.save_dir + 'images'):
os.mkdir(self.save_dir + 'images')
if not os.path.exists(self.save_dir + 'model'):
os.mkdir(self.save_dir + 'model')
print(self.save_dir)
if torch.cuda.is_available():
print("cuda")
self.use_cuda = True
torch.set_default_tensor_type('torch.cuda.FloatTensor')
else:
self.use_cuda = False
torch.set_default_tensor_type('torch.FloatTensor')
self.nz = config.nz
self.optimizer = config.optimizer
self.resl = 2 # we start from 2^2 = 4
self.lr = config.lr
self.eps_drift = config.eps_drift
self.smoothing = config.smoothing
self.max_resl = config.max_resl
self.trns_tick = config.trns_tick
self.stab_tick = config.stab_tick
self.TICK = config.TICK
self.globalIter = 0
self.globalTick = 0
self.kimgs = 0
self.stack = 0
self.epoch = 0
self.fadein = {'gen': None, 'dis': None}
self.complete = {'gen': 0, 'dis': 0}
self.phase = 'init'
self.flag_flush_gen = False
self.flag_flush_dis = False
self.flag_add_noise = self.config.flag_add_noise
self.flag_add_drift = self.config.flag_add_drift
# network and cirterion
self.G = net.Generator(config)
self.D = net.Discriminator(config)
print('Generator structure: ')
print(self.G.model)
print('Discriminator structure: ')
print(self.D.model)
self.mse = torch.nn.MSELoss()
self.l1 = torch.nn.L1Loss()
if self.use_cuda:
self.mse = self.mse.cuda()
self.l1 = self.l1.cuda()
torch.cuda.manual_seed(config.random_seed)
if config.n_gpu == 1:
self.G = torch.nn.DataParallel(self.G).cuda(device=0)
self.D = torch.nn.DataParallel(self.D).cuda(device=0)
else:
gpus = []
for i in range(config.n_gpu):
gpus.append(i)
self.G = torch.nn.DataParallel(self.G, device_ids=gpus).cuda()
self.D = torch.nn.DataParallel(self.D, device_ids=gpus).cuda()
# define tensors, ship model to cuda, and get dataloader.
self.renew_everything()
transform=transforms.Compose([
transforms.Resize((img_size, img_size), Image.BICUBIC),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])),
batch_size=batch_size,
shuffle=True)
# GAN Loss function
adversarial_loss = nn.MSELoss()
# Initialize generator and discriminator
generator = network.Generator(latent_dim=latent_dim,
classes=n_classes,
channels=n_channels)
discriminator = network.Discriminator(classes=n_classes, channels=n_channels)
# Label embedding
label_emb = nn.Embedding(n_classes, n_classes)
if cuda:
generator.cuda()
discriminator.cuda()
adversarial_loss.cuda()
label_emb.cuda()
# Optimizers
optimizer_G = torch.optim.Adam(generator.parameters(), lr=lr, betas=(b1, b2))
optimizer_D = torch.optim.Adam(discriminator.parameters(),
lr=lr,
betas=(b1, b2))
def meta_train(device, dataset_path, continue_id):
run_start = datetime.now()
logging.info('===== META-TRAINING =====')
# GPU / CPU --------------------------------------------------------------------------------------------------------
if device is not None and device != 'cpu':
dtype = torch.cuda.FloatTensor
torch.cuda.set_device(device)
logging.info(f'Running on GPU: {torch.cuda.current_device()}.')
else:
dtype = torch.FloatTensor
logging.info(f'Running on CPU.')
# DATASET-----------------------------------------------------------------------------------------------------------
logging.info(f'Training using dataset located in {dataset_path}')
dataset = VoxCelebDataset(root=dataset_path,
extension='.vid',
shuffle=False,
shuffle_frames=True,
transform=transforms.Compose([
transforms.Resize(config.IMAGE_SIZE),
transforms.CenterCrop(config.IMAGE_SIZE),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225]),
]))
# NETWORK ----------------------------------------------------------------------------------------------------------
E = network.Embedder().type(dtype)
G = network.Generator().type(dtype)
D = network.Discriminator(143000).type(dtype)
if continue_id is not None:
E = load_model(E, continue_id)
G = load_model(G, continue_id)
D = load_model(D, continue_id)
optimizer_E_G = Adam(params=list(E.parameters()) + list(G.parameters()),
lr=config.LEARNING_RATE_E_G)
optimizer_D = Adam(params=D.parameters(), lr=config.LEARNING_RATE_D)
criterion_E_G = network.LossEG(device, feed_forward=True)
criterion_D = network.LossD(device)
# TRAINING LOOP ----------------------------------------------------------------------------------------------------
logging.info(
f'Starting training loop. Epochs: {config.EPOCHS} Dataset Size: {len(dataset)}'
)
for epoch in range(config.EPOCHS):
epoch_start = datetime.now()
batch_durations = []
E.train()
G.train()
D.train()
for batch_num, (i, video) in enumerate(dataset):
batch_start = datetime.now()
# Put one frame aside (frame t)
t = video.pop()
# Calculate average encoding vector for video
e_vectors = []
for s in video:
x_s = s['frame'].type(dtype)
y_s = s['landmarks'].type(dtype)
e_vectors.append(E(x_s, y_s))
e_hat = torch.stack(e_vectors).mean(dim=0)
# Generate frame using landmarks from frame t
x_t = t['frame'].type(dtype)
y_t = t['landmarks'].type(dtype)
x_hat = G(y_t, e_hat)
# Optimize E_G and D
r_x_hat, D_act_hat = D(x_hat, y_t, i)
r_x, D_act = D(x_t, y_t, i)
optimizer_E_G.zero_grad()
optimizer_D.zero_grad()
loss_E_G = criterion_E_G(x_t, x_hat, r_x_hat, e_hat, D.W[:, i],
D_act, D_act_hat)
loss_D = criterion_D(r_x, r_x_hat)
loss = loss_E_G + loss_D
loss.backward(retain_graph=True)
optimizer_E_G.step()
optimizer_D.step()
# Optimize D again
r_x_hat, D_act_hat = D(G(y_t, e_hat), y_t, i)
r_x, D_act = D(x_t, y_t, i)
optimizer_D.zero_grad()
loss_D = criterion_D(r_x, r_x_hat)
loss_D.backward()
optimizer_D.step()
batch_end = datetime.now()
batch_durations.append(batch_end - batch_start)
# SHOW PROGRESS --------------------------------------------------------------------------------------------
if (batch_num + 1) % 100 == 0 or batch_num == 0:
avg_time = sum(batch_durations,
timedelta(0)) / len(batch_durations)
logging.info(
f'Epoch {epoch+1}: [{batch_num + 1}/{len(dataset)}] | '
f'Avg Time: {avg_time} | '
f'Loss_E_G = {loss_E_G.item():.4} Loss_D {loss_D.item():.4}'
)
logging.debug(
f'D(x) = {r_x.item():.4} D(x_hat) = {r_x_hat.item():.4}')
# SAVE IMAGES ----------------------------------------------------------------------------------------------
if (batch_num + 1) % 100 == 0:
if not os.path.isdir(config.GENERATED_DIR):
os.makedirs(config.GENERATED_DIR)
save_image(
os.path.join(config.GENERATED_DIR,
f'{datetime.now():%Y%m%d_%H%M}_x.png'), x_t)
save_image(
os.path.join(config.GENERATED_DIR,
f'{datetime.now():%Y%m%d_%H%M}_x_hat.png'),
x_hat)
if (batch_num + 1) % 2000 == 0:
save_model(E, device)
save_model(G, device)
save_model(D, device)
# SAVE MODELS --------------------------------------------------------------------------------------------------
save_model(E, device, run_start)
save_model(G, device, run_start)
save_model(D, device, run_start)
epoch_end = datetime.now()
logging.info(
f'Epoch {epoch+1} finished in {epoch_end - epoch_start}. '
f'Average batch time: {sum(batch_durations, timedelta(0)) / len(batch_durations)}'
)
y_cat = np.zeros((y.shape[0], num_columns))
y_cat[range(y.shape[0]), y] = 1.
return Variable(FloatTensor(y_cat))
# GAN Loss function
adversarial_loss = nn.MSELoss()
categorical_loss = nn.CrossEntropyLoss()
continuous_loss = nn.MSELoss()
# Initialize generator and discriminator
generator = network.Generator(latent_dim=latent_dim,
categorical_dim=n_classes,
continuous_dim=code_dim)
discriminator = network.Discriminator(categorical_dim=n_classes)
if cuda:
generator.cuda()
discriminator.cuda()
adversarial_loss.cuda()
categorical_loss.cuda()
continuous_loss.cuda()
# Optimizers
optimizer_G = torch.optim.Adam(generator.parameters(), lr=lr, betas=(b1, b2))
optimizer_D = torch.optim.Adam(discriminator.parameters(),
lr=lr,
betas=(b1, b2))
# itertools.chain : iterable 객체를 연결, ex = itertools.chain([1, 2, 3], {'a', 'b', 'c'}) => ex = (1, 2, 3, 'a', 'b', 'c')
optimizer_info = torch.optim.Adam(itertools.chain(generator.parameters(),
def meta_train(gpu, dataset_path, continue_id):
run_start = datetime.now()
logging.info('===== META-TRAINING =====')
logging.info(f'Running on {"GPU" if gpu else "CPU"}.')
# region DATASET----------------------------------------------------------------------------------------------------
logging.info(f'Training using dataset located in {dataset_path}')
raw_dataset = VoxCelebDataset(
root=dataset_path,
extension='.vid',
shuffle_frames=True,
subset_size=config.SUBSET_SIZE,
transform=transforms.Compose([
transforms.Resize(config.IMAGE_SIZE),
transforms.CenterCrop(config.IMAGE_SIZE),
transforms.ToTensor(),
])
)
dataset = DataLoader(raw_dataset, batch_size=config.BATCH_SIZE, shuffle=True)
# endregion
# region NETWORK ---------------------------------------------------------------------------------------------------
E = network.Embedder(GPU['Embedder'])
G = network.Generator(GPU['Generator'])
D = network.Discriminator(len(raw_dataset), GPU['Discriminator'])
criterion_E_G = network.LossEG(config.FEED_FORWARD, GPU['LossEG'])
criterion_D = network.LossD(GPU['LossD'])
optimizer_E_G = Adam(
params=list(E.parameters()) + list(G.parameters()),
lr=config.LEARNING_RATE_E_G
)
optimizer_D = Adam(
params=D.parameters(),
lr=config.LEARNING_RATE_D
)
if continue_id is not None:
E = load_model(E, continue_id)
G = load_model(G, continue_id)
D = load_model(D, continue_id)
# endregion
# region TRAINING LOOP ---------------------------------------------------------------------------------------------
logging.info(f'Epochs: {config.EPOCHS} Batches: {len(dataset)} Batch Size: {config.BATCH_SIZE}')
for epoch in range(config.EPOCHS):
epoch_start = datetime.now()
E.train()
G.train()
D.train()
for batch_num, (i, video) in enumerate(dataset):
# region PROCESS BATCH -------------------------------------------------------------------------------------
batch_start = datetime.now()
# video [B, K+1, 2, C, W, H]
# Put one frame aside (frame t)
t = video[:, -1, ...] # [B, 2, C, W, H]
video = video[:, :-1, ...] # [B, K, 2, C, W, H]
dims = video.shape
# Calculate average encoding vector for video
e_in = .reshape(dims[0] * dims[1], dims[2], dims[3], dims[4], dims[5]) # [BxK, 2, C, W, H]
x, y = e_in[:, 0, ...], e_in[:, 1, ...]
e_vectors = E(x, y).reshape(dims[0], dims[1], -1) # B, K, len(e)
e_hat = e_vectors.mean(dim=1)
# Generate frame using landmarks from frame t
x_t, y_t = t[:, 0, ...], t[:, 1, ...]
x_hat = G(y_t, e_hat)
# Optimize E_G and D
r_x_hat, _ = D(x_hat, y_t, i)
r_x, _ = D(x_t, y_t, i)
optimizer_E_G.zero_grad()
optimizer_D.zero_grad()
loss_E_G = criterion_E_G(x_t, x_hat, r_x_hat, e_hat, D.W[:, i].transpose(1, 0))
loss_D = criterion_D(r_x, r_x_hat)
loss = loss_E_G + loss_D
loss.backward()
optimizer_E_G.step()
optimizer_D.step()
# Optimize D again
x_hat = G(y_t, e_hat).detach()
r_x_hat, D_act_hat = D(x_hat, y_t, i)
r_x, D_act = D(x_t, y_t, i)
optimizer_D.zero_grad()
loss_D = criterion_D(r_x, r_x_hat)
loss_D.backward()
optimizer_D.step()
batch_end = datetime.now()
# endregion
# region SHOW PROGRESS -------------------------------------------------------------------------------------
if (batch_num + 1) % 1 == 0 or batch_num == 0:
logging.info(f'Epoch {epoch + 1}: [{batch_num + 1}/{len(dataset)}] | '
f'Time: {batch_end - batch_start} | '
f'Loss_E_G = {loss_E_G.item():.4f} Loss_D = {loss_D.item():.4f}')
logging.debug(f'D(x) = {r_x.mean().item():.4f} D(x_hat) = {r_x_hat.mean().item():.4f}')
# endregion
# region SAVE ----------------------------------------------------------------------------------------------
save_image(os.path.join(config.GENERATED_DIR, f'last_result_x.png'), x_t[0])
save_image(os.path.join(config.GENERATED_DIR, f'last_result_x_hat.png'), x_hat[0])
if (batch_num + 1) % 100 == 0:
save_image(os.path.join(config.GENERATED_DIR, f'{datetime.now():%Y%m%d_%H%M%S%f}_x.png'), x_t[0])
save_image(os.path.join(config.GENERATED_DIR, f'{datetime.now():%Y%m%d_%H%M%S%f}_x_hat.png'), x_hat[0])
if (batch_num + 1) % 100 == 0:
save_model(E, gpu, run_start)
save_model(G, gpu, run_start)
save_model(D, gpu, run_start)
# endregion
# SAVE MODELS --------------------------------------------------------------------------------------------------
save_model(E, gpu, run_start)
save_model(G, gpu, run_start)
save_model(D, gpu, run_start)
epoch_end = datetime.now()
logging.info(f'Epoch {epoch + 1} finished in {epoch_end - epoch_start}. ')
def train():
parser = argparse.ArgumentParser()
parser.add_argument('--gpu', '-g', type=int, default=0)
parser.add_argument('--dir', type=str, default='./CelebA_Datasets/')
parser.add_argument('--gen', type=str, default=None)
parser.add_argument('--dis', type=str, default=None)
parser.add_argument('--optg', type=str, default=None)
parser.add_argument('--optd', type=str, default=None)
parser.add_argument('--epoch', '-e', type=int, default=3)
parser.add_argument('--lr', '-l', type=float, default=0.001)
parser.add_argument('--beta1', type=float, default=0)
parser.add_argument('--beta2', type=float, default=0.99)
parser.add_argument('--batch', '-b', type=int, default=16)
parser.add_argument('--depth', '-d', type=int, default=0)
parser.add_argument('--alpha', type=float, default=0)
parser.add_argument('--delta', type=float, default=0.00005)
parser.add_argument('--out', '-o', type=str, default='img/')
parser.add_argument('--num', '-n', type=int, default=10)
parser.add_argument('--im_num', '-i', type=int, default=200000)
args = parser.parse_args()
print('==================================================')
print('Depth: {}'.format(args.depth))
print('Num Minibatch Size: {}'.format(args.batch))
print('Num epoch: {}'.format(args.epoch))
print('==================================================')
gen = network.Generator(depth=args.depth)
if args.gen is not None:
print('loading generator model from ' + args.gen)
serializers.load_npz(args.gen, gen)
dis = network.Discriminator(depth=args.depth)
if args.dis is not None:
print('loading discriminator model from ' + args.dis)
serializers.load_npz(args.dis, dis)
if args.gpu >= 0:
cuda.get_device_from_id(args.gpu).use()
gen.to_gpu()
dis.to_gpu()
opt_g = optimizers.Adam(alpha=args.lr, beta1=args.beta1, beta2=args.beta2)
opt_g.setup(gen)
if args.optg is not None:
print('loading generator optimizer from ' + args.optg)
serializers.load_npz(args.optg, opt_g)
opt_d = optimizers.Adam(alpha=args.lr, beta1=args.beta1, beta2=args.beta2)
opt_d.setup(dis)
if args.optd is not None:
print('loading discriminator optimizer from ' + args.optd)
serializers.load_npz(args.optd, opt_d)
train = datasets.labelandID(args.dir, './../../GAN/text_file/out_attr.txt',
args.im_num, args.depth)
train = chainer.datasets.LabeledImageDataset(pairs=train,
root='resized_images')
train = chainermn.scatter_dataset(train, comm)
train_iter = iterators.SerialIterator(train, batch_size=args.batch)
updater = WganGpUpdater(alpha=args.alpha,
delta=args.delta,
models=(gen, dis),
iterator={'main': train_iter},
optimizer={
'gen': opt_g,
'dis': opt_d
},
device=args.gpu)
trainer = training.Trainer(updater, (args.epoch, 'epoch'), out='results')
if os.path.isdir(args.out):
shutil.rmtree(args.out)
os.makedirs(args.out)
def output_image(gen, depth, out, num):
@chainer.training.make_extension()
def make_image(trainer):
z = gen.z(num)
attribute = [1, 0, 0, 0, 1]
c = [cupy.asarray(attribute, cupy.float32) for i in range(num)]
c = cupy.asarray(c, dtype=cupy.float32)
x = gen(z, c, alpha=trainer.updater.alpha)
x = chainer.cuda.to_cpu(x.data)
for i in range(args.num):
img = x[i].copy()
filename = os.path.join(
out, '%d_%d.png' % (trainer.updater.epoch, i))
utils.save_image(img, filename)
return make_image
trainer.extend(extensions.LogReport(trigger=(1, 'epoch')))
trainer.extend(
extensions.PrintReport([
'epoch', 'gen_loss', 'loss_d', 'loss_l', 'loss_dr', 'dis_loss',
'alpha'
]))
trainer.extend(extensions.snapshot_object(gen, 'gen'),
trigger=(10, 'epoch'))
trainer.extend(extensions.snapshot_object(dis, 'dis'),
trigger=(10, 'epoch'))
trainer.extend(extensions.snapshot_object(opt_g, 'opt_g'),
trigger=(10, 'epoch'))
trainer.extend(extensions.snapshot_object(opt_d, 'opt_d'),
trigger=(10, 'epoch'))
trainer.extend(output_image(gen, args.depth, args.out, args.num),
trigger=(1, 'epoch'))
trainer.extend(extensions.ProgressBar(update_interval=1))
trainer.run()
modelname = './results/gen'
print('saving generator model to ' + modelname)
serializers.save_npz(modelname, gen)
modelname = './results/dis'
print('saving discriminator model to ' + modelname)
serializers.save_npz(modelname, dis)
optname = './results/opt_g'
print('saving generator optimizer to ' + optname)
serializers.save_npz(optname, opt_g)
optname = './results/opt_d'
print('saving generator optimizer to ' + optname)
serializers.save_npz(optname, opt_d)
def train():
parser = argparse.ArgumentParser()
parser.add_argument('--gpu', '-g', type=int, default=-1)
parser.add_argument('--dir', type=str, default='./train_images/')
parser.add_argument('--gen', type=str, default=None)
parser.add_argument('--dis', type=str, default=None)
parser.add_argument('--optg', type=str, default=None)
parser.add_argument('--optd', type=str, default=None)
parser.add_argument('--epoch', '-e', type=int, default=3)
parser.add_argument('--lr', '-l', type=float, default=0.001)
parser.add_argument('--beta1', type=float, default=0)
parser.add_argument('--beta2', type=float, default=0.99)
parser.add_argument('--batch', '-b', type=int, default=16)
parser.add_argument('--depth', '-d', type=int, default=0)
parser.add_argument('--alpha', type=float, default=0)
parser.add_argument('--delta', type=float, default=0.00005)
parser.add_argument('--out', '-o', type=str, default='img/')
parser.add_argument('--num', '-n', type=int, default=10)
parser.add_argument('--dim', type=int, default=5)
args = parser.parse_args()
train = dataset.YuiDataset(directory=args.dir, depth=args.depth)
train_iter = iterators.SerialIterator(train, batch_size=args.batch)
gen = network.LimGenerator(depth=args.depth, dim=args.dim)
if args.gen is not None:
print('loading generator model from ' + args.gen)
serializers.load_npz(args.gen, gen)
dis = network.Discriminator(depth=args.depth)
if args.dis is not None:
print('loading discriminator model from ' + args.dis)
serializers.load_npz(args.dis, dis)
if args.gpu >= 0:
cuda.get_device_from_id(0).use()
gen.to_gpu()
dis.to_gpu()
opt_g = optimizers.Adam(alpha=args.lr, beta1=args.beta1, beta2=args.beta2)
opt_g.setup(gen)
if args.optg is not None:
print('loading generator optimizer from ' + args.optg)
serializers.load_npz(args.optg, opt_g)
opt_d = optimizers.Adam(alpha=args.lr, beta1=args.beta1, beta2=args.beta2)
opt_d.setup(dis)
if args.optd is not None:
print('loading discriminator optimizer from ' + args.optd)
serializers.load_npz(args.optd, opt_d)
updater = WganGpUpdater(alpha=args.alpha,
delta=args.delta,
models=(gen, dis),
iterator={'main': train_iter},
optimizer={'gen': opt_g, 'dis': opt_d},
device=args.gpu)
trainer = training.Trainer(updater, (args.epoch, 'epoch'), out='results')
if os.path.isdir(args.out):
shutil.rmtree(args.out)
os.makedirs(args.out)
for i in range(args.num):
img = train.get_example(i)
filename = os.path.join(args.out, 'real_%d.png'%i)
utils.save_image(img, filename)
def output_image(gen, depth, out, num):
@chainer.training.make_extension()
def make_image(trainer):
z = gen.z(num)
x = gen(z, alpha=trainer.updater.alpha)
x = chainer.cuda.to_cpu(x.data)
for i in range(args.num):
img = x[i].copy()
filename = os.path.join(out, '%d_%d.png' % (trainer.updater.epoch, i))
utils.save_image(img, filename)
return make_image
trainer.extend(extensions.LogReport(trigger=(1, 'epoch')))
trainer.extend(extensions.PrintReport(['epoch', 'gen_loss', 'loss_d', 'loss_l', 'loss_dr', 'dis_loss', 'alpha']))
trainer.extend(extensions.snapshot_object(gen, 'gen'), trigger=(10, 'epoch'))
trainer.extend(extensions.snapshot_object(dis, 'dis'), trigger=(10, 'epoch'))
trainer.extend(extensions.snapshot_object(opt_g, 'opt_g'), trigger=(10, 'epoch'))
trainer.extend(extensions.snapshot_object(opt_d, 'opt_d'), trigger=(10, 'epoch'))
trainer.extend(output_image(gen, args.depth, args.out, args.num), trigger=(1, 'epoch'))
trainer.extend(extensions.ProgressBar(update_interval=1))
trainer.run()
modelname = './results/gen'
print( 'saving generator model to ' + modelname )
serializers.save_npz(modelname, gen)
modelname = './results/dis'
print( 'saving discriminator model to ' + modelname )
serializers.save_npz(modelname, dis)
optname = './results/opt_g'
print( 'saving generator optimizer to ' + optname )
serializers.save_npz(optname, opt_g)
optname = './results/opt_d'
print( 'saving generator optimizer to ' + optname )
serializers.save_npz(optname, opt_d)
train=True,
download=True,
transform=transforms.Compose([
transforms.Resize((img_size, img_size), Image.BICUBIC),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
])),
batch_size=batch_size,
shuffle=True)
# GAN Loss function
adversarial_loss = nn.BCELoss()
# Initialize generator and discriminator
generator = network.Generator(latent_dim=latent_dim, img_shape=img_shape)
discriminator = network.Discriminator(img_shape=img_shape)
if cuda:
generator.cuda()
discriminator.cuda()
adversarial_loss.cuda()
# Optimizers
optimizer_G = torch.optim.Adam(generator.parameters(), lr=lr, betas=(b1, b2))
optimizer_D = torch.optim.Adam(discriminator.parameters(),
lr=lr,
betas=(b1, b2))
# gpu or cpu
Tensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor
IfInitial = False
dp = data_processor.DataProcessor()
if IfInitial:
dp.init_data()
print('initial done!')
# get training data and test data
train_set_np = np.load("data/train_set.npy")
train_set_label_np = np.load("data/train_set_label.npy")
test_set_np = np.load("data/test_set.npy")
test_set_label_np = np.load("data/test_set_label.npy")
# network
G = nw.Generator().cuda()
D = nw.Discriminator().cuda()
G.weight_init(mean=0, std=0.01)
D.weight_init(mean=0, std=0.01)
# load train data
BatchSize = 32
train_set = torch.load('data/train_data_set.lib')
train_data = torch_data.DataLoader(
train_set,
batch_size=BatchSize,
shuffle=True,
num_workers=2,
)
def train(settings, output_root):
# directories
weights_root = output_root.joinpath("weights")
weights_root.mkdir()
# settings
amp_handle = amp.init(settings["use_apex"])
if settings["use_cuda"]:
device = torch.device("cuda:0")
else:
device = torch.device("cpu")
dtype = torch.float32
test_device = torch.device("cuda:0")
test_dtype = torch.float16
loss_type = settings["loss"]
z_dim = settings["network"]["z_dim"]
# model
label_size = len(settings["labels"])
generator = network.Generator(settings["network"],
label_size).to(device, dtype)
discriminator = network.Discriminator(settings["network"],
label_size).to(device, dtype)
# long-term average
gs = network.Generator(settings["network"], label_size).to(device, dtype)
gs.load_state_dict(generator.state_dict())
gs_beta = settings["gs_beta"]
lt_learning_rate = settings["learning_rates"]["latent_transformation"]
g_learning_rate = settings["learning_rates"]["generator"]
d_learning_rate = settings["learning_rates"]["discriminator"]
g_opt = optim.Adam([{
"params": generator.latent_transform.parameters(),
"lr": lt_learning_rate
}, {
"params": generator.synthesis_module.parameters()
}],
lr=g_learning_rate,
betas=(0.0, 0.99),
eps=1e-8)
d_opt = optim.Adam(discriminator.parameters(),
lr=d_learning_rate,
betas=(0.0, 0.99),
eps=1e-8)
# train data
loader = data_loader.LabeledDataLoader(settings)
if settings["use_yuv"]:
converter = image_converter.YUVConverter()
else:
converter = image_converter.RGBConverter()
# parameters
level = settings["start_level"]
generator.set_level(level)
discriminator.set_level(level)
gs.set_level(level)
fading = False
alpha = 1
step = 0
# log
writer = SummaryWriter(str(output_root))
test_rows = 12
test_cols = 6
test_zs = utils.create_test_z(test_rows, test_cols, z_dim)
test_z0 = torch.from_numpy(test_zs[0]).to(test_device, test_dtype)
test_z1 = torch.from_numpy(test_zs[1]).to(test_device, test_dtype)
test_labels0 = torch.randint(0, loader.label_size, (1, test_cols))
test_labels0 = test_labels0.repeat(test_rows, 1).to(device)
test_labels1 = torch.randint(0,
loader.label_size, (test_rows, test_cols),
device=test_device).view(-1)
for loop in range(9999999):
size = 2**(level + 1)
batch_size = settings["batch_sizes"][level - 1]
alpha_delta = batch_size / settings["num_images_in_stage"]
image_count = 0
for batch, labels in loader.generate(batch_size, size, size):
# pre train
step += 1
image_count += batch_size
if fading:
alpha = min(1.0, alpha + alpha_delta)
# data
batch = batch.transpose([0, 3, 1, 2])
batch = converter.to_train_data(batch)
trues = torch.from_numpy(batch).to(device, dtype)
labels = torch.from_numpy(labels).to(device)
# reset
g_opt.zero_grad()
d_opt.zero_grad()
# === train discriminator ===
z = utils.create_z(batch_size, z_dim)
z = torch.from_numpy(z).to(device, dtype)
fakes = generator.forward(z, labels, alpha)
fakes_nograd = fakes.detach()
for param in discriminator.parameters():
param.requires_grad_(True)
if loss_type == "wgan":
d_loss, wd = loss.d_wgan_loss(discriminator, trues,
fakes_nograd, labels, alpha)
elif loss_type == "lsgan":
d_loss = loss.d_lsgan_loss(discriminator, trues, fakes_nograd,
labels, alpha)
elif loss_type == "logistic":
d_loss = loss.d_logistic_loss(discriminator, trues,
fakes_nograd, labels, alpha)
else:
raise Exception(f"Invalid loss: {loss_type}")
with amp_handle.scale_loss(d_loss, d_opt) as scaled_loss:
scaled_loss.backward()
d_opt.step()
# === train generator ===
z = utils.create_z(batch_size, z_dim)
z = torch.from_numpy(z).to(device, dtype)
fakes = generator.forward(z, labels, alpha)
for param in discriminator.parameters():
param.requires_grad_(False)
if loss_type == "wgan":
g_loss = loss.g_wgan_loss(discriminator, fakes, labels, alpha)
elif loss_type == "lsgan":
g_loss = loss.g_lsgan_loss(discriminator, fakes, labels, alpha)
elif loss_type == "logistic":
g_loss = loss.g_logistic_loss(discriminator, fakes, labels,
alpha)
else:
raise Exception(f"Invalid loss: {loss_type}")
with amp_handle.scale_loss(g_loss, g_opt) as scaled_loss:
scaled_loss.backward()
del scaled_loss
g_opt.step()
del trues, fakes, fakes_nograd
# update gs
for gparam, gsparam in zip(generator.parameters(),
gs.parameters()):
gsparam.data = (1 -
gs_beta) * gsparam.data + gs_beta * gparam.data
gs.w_average.data = (
1 - gs_beta
) * gs.w_average.data + gs_beta * generator.w_average.data
# log
if step % 1 == 0:
print(f"lv{level}-{step}: "
f"a: {alpha:.5f} "
f"g: {g_loss.item():.7f} "
f"d: {d_loss.item():.7f} ")
writer.add_scalar(f"lv{level}/loss_gen",
g_loss.item(),
global_step=step)
writer.add_scalar(f"lv{level}/loss_disc",
d_loss.item(),
global_step=step)
if loss_type == "wgan":
writer.add_scalar(f"lv{level}/wd", wd, global_step=step)
del d_loss, g_loss
# histogram
if settings["save_steps"]["histogram"] > 0 and step % settings[
"save_steps"]["histogram"] == 0:
gs.write_histogram(writer, step)
for name, param in discriminator.named_parameters():
writer.add_histogram(f"disc/{name}",
param.cpu().data.numpy(), step)
# image
if step % settings["save_steps"]["image"] == 0 or alpha == 0:
fading_text = "fading" if fading else "stabilizing"
with torch.no_grad():
eval_gen = network.Generator(settings["network"],
label_size).to(
test_device,
test_dtype).eval()
eval_gen.load_state_dict(gs.state_dict())
eval_gen.synthesis_module.set_noise_fixed(True)
fakes = eval_gen.forward(test_z0, test_labels0, alpha)
fakes = torchvision.utils.make_grid(fakes,
nrow=test_cols,
padding=0)
fakes = fakes.to(torch.float32).cpu().numpy()
fakes = converter.from_generator_output(fakes)
writer.add_image(f"lv{level}_{fading_text}/intpl",
torch.from_numpy(fakes), step)
fakes = eval_gen.forward(test_z1, test_labels1, alpha)
fakes = torchvision.utils.make_grid(fakes,
nrow=test_cols,
padding=0)
fakes = fakes.to(torch.float32).cpu().numpy()
fakes = converter.from_generator_output(fakes)
writer.add_image(f"lv{level}_{fading_text}/random",
torch.from_numpy(fakes), step)
del eval_gen
# memory usage
writer.add_scalar("memory_allocated(MB)",
torch.cuda.memory_allocated() /
(1024 * 1024),
global_step=step)
# model save
if step % settings["save_steps"][
"model"] == 0 and level >= 5 and not fading:
savedir = weights_root.joinpath(f"{step}_lv{level}")
savedir.mkdir()
torch.save(generator.state_dict(), savedir.joinpath("gen.pth"))
torch.save(generator.state_dict(), savedir.joinpath("gs.pth"))
torch.save(discriminator.state_dict(),
savedir.joinpath("disc.pth"))
# switch fading/stabilizing
if image_count > settings["num_images_in_stage"]:
if fading:
print("start stabilizing")
fading = False
alpha = 1
image_count = 0
elif level < settings["max_level"]:
print(f"end lv: {level}")
break
# level up
if level < settings["max_level"]:
level = level + 1
generator.set_level(level)
discriminator.set_level(level)
gs.set_level(level)
fading = True
alpha = 0
print(f"lv up: {level}")
if settings["reset_optimizer"]:
g_opt = optim.Adam(
[{
"params": generator.latent_transform.parameters(),
"lr": lt_learning_rate
}, {
"params": generator.synthesis_module.parameters()
}],
lr=g_learning_rate,
betas=(0.0, 0.99),
eps=1e-8)
d_opt = optim.Adam(discriminator.parameters(),
lr=d_learning_rate,
betas=(0.0, 0.99),
eps=1e-8)
def __init__(self, config):
self.config = config
if torch.cuda.is_available():
self.use_cuda = True
torch.set_default_tensor_type("torch.cuda.FloatTensor")
else:
self.use_cuda = False
torch.set_default_tensor_type("torch.FloatTensor")
self.nz = config.nz
self.optimizer = config.optimizer
self.resl = 2 # we start from 2^2 = 4
self.lr = config.lr
self.eps_drift = config.eps_drift
self.smoothing = config.smoothing
self.max_resl = config.max_resl
self.accelerate = 1
self.wgan_target = 1.0
self.trns_tick = config.trns_tick
self.stab_tick = config.stab_tick
self.TICK = config.TICK
self.skip = False
self.globalIter = 0
self.globalTick = 0
self.wgan_epsilon = 0.001
self.stack = 0
self.wgan_lambda = 10.0
self.just_passed = False
if self.config.resume:
saved_models = os.listdir("repo/model/")
iterations = list(
map(lambda x: int(x.split("_")[-1].split(".")[0][1:]),
saved_models))
self.last_iteration = max(iterations)
selected_indexes = np.where(
[x == self.last_iteration for x in iterations])[0]
G_last_model = [
saved_models[x] for x in selected_indexes
if "gen" in saved_models[x]
][0]
D_last_model = [
saved_models[x] for x in selected_indexes
if "dis" in saved_models[x]
][0]
saved_grids = os.listdir("repo/save/grid")
global_iterations = list(
map(lambda x: int(x.split("_")[0]), saved_grids))
self.globalIter = self.config.save_img_every * max(
global_iterations)
print("Resuming after " + str(self.last_iteration) +
" ticks and " + str(self.globalIter) + " iterations")
G_weights = torch.load("repo/model/" + G_last_model)
D_weights = torch.load("repo/model/" + D_last_model)
self.resuming = True
else:
self.resuming = False
self.kimgs = 0
self.stack = 0
self.epoch = 0
self.fadein = {"gen": None, "dis": None}
self.complete = {"gen": 0, "dis": 0}
self.phase = "init"
self.flag_flush_gen = False
self.flag_flush_dis = False
self.flag_add_noise = self.config.flag_add_noise
self.flag_add_drift = self.config.flag_add_drift
# network and cirterion
self.G = net.Generator(config)
self.D = net.Discriminator(config)
print("Generator structure: ")
print(self.G.model)
print("Discriminator structure: ")
print(self.D.model)
self.mse = torch.nn.MSELoss()
if self.use_cuda:
self.mse = self.mse.cuda()
torch.cuda.manual_seed(config.random_seed)
self.G = torch.nn.DataParallel(self.G,
device_ids=[0]).cuda(device=0)
self.D = torch.nn.DataParallel(self.D,
device_ids=[0]).cuda(device=0)
# define tensors, ship model to cuda, and get dataloader.
self.renew_everything()
if self.resuming:
self.resl = G_weights["resl"]
self.globalIter = G_weights["globalIter"]
self.globalTick = G_weights["globalTick"]
self.kimgs = G_weights["kimgs"]
self.epoch = G_weights["epoch"]
self.phase = G_weights["phase"]
self.fadein = G_weights["fadein"]
self.complete = G_weights["complete"]
self.flag_flush_gen = G_weights["flag_flush_gen"]
self.flag_flush_dis = G_weights["flag_flush_dis"]
self.stack = G_weights["stack"]
print("Resuming at " + str(self.resl) + " definition after " +
str(self.epoch) + " epochs")
self.G.module.load_state_dict(G_weights["state_dict"])
self.D.module.load_state_dict(D_weights["state_dict"])
self.opt_g.load_state_dict(G_weights["optimizer"])
self.opt_d.load_state_dict(D_weights["optimizer"])
# tensorboard
self.use_tb = config.use_tb
if self.use_tb:
self.tb = tensorboard.tf_recorder()
def meta_train(gpu, dataset_path, continue_id):
run_start = datetime.now()
logging.info('===== META-TRAINING =====')
# GPU / CPU --------------------------------------------------------------------------------------------------------
if gpu:
dtype = torch.cuda.FloatTensor
torch.set_default_tensor_type(dtype)
logging.info(f'Running on GPU: {torch.cuda.current_device()}.')
else:
dtype = torch.FloatTensor
torch.set_default_tensor_type(dtype)
logging.info(f'Running on CPU.')
# DATASET-----------------------------------------------------------------------------------------------------------
logging.info(f'Training using dataset located in {dataset_path}')
raw_dataset = VoxCelebDataset(
root=dataset_path,
extension='.vid',
shuffle_frames=True,
# subset_size=1,
transform=transforms.Compose([
transforms.Resize(config.IMAGE_SIZE),
transforms.CenterCrop(config.IMAGE_SIZE),
transforms.ToTensor(),
]))
dataset = DataLoader(raw_dataset,
batch_size=config.BATCH_SIZE,
shuffle=True)
# NETWORK ----------------------------------------------------------------------------------------------------------
E = network.Embedder().type(dtype)
G = network.Generator().type(dtype)
D = network.Discriminator(len(raw_dataset)).type(dtype)
optimizer_E_G = Adam(params=list(E.parameters()) + list(G.parameters()),
lr=config.LEARNING_RATE_E_G)
optimizer_D = Adam(params=D.parameters(), lr=config.LEARNING_RATE_D)
criterion_E_G = network.LossEG(feed_forward=True)
criterion_D = network.LossD()
if gpu:
E = DataParallel(E)
G = DataParallel(G)
D = ParallelDiscriminator(D)
criterion_E_G = DataParallel(criterion_E_G)
criterion_D = DataParallel(criterion_D)
if continue_id is not None:
E = load_model(E, 'Embedder', continue_id)
G = load_model(G, 'Generator', continue_id)
D = load_model(D, 'Discriminator', continue_id)
# TRAINING LOOP ----------------------------------------------------------------------------------------------------
logging.info(f'Starting training loop. '
f'Epochs: {config.EPOCHS} '
f'Batches: {len(dataset)} '
f'Batch Size: {config.BATCH_SIZE}')
for epoch in range(config.EPOCHS):
epoch_start = datetime.now()
batch_durations = []
E.train()
G.train()
D.train()
for batch_num, (i, video) in enumerate(dataset):
batch_start = datetime.now()
video = video.type(dtype) # [B, K+1, 2, C, W, H]
# Put one frame aside (frame t)
t = video[:, -1, ...] # [B, 2, C, W, H]
video = video[:, :-1, ...] # [B, K, C, W, H]
dims = video.shape
# Calculate average encoding vector for video
e_in = video.reshape(dims[0] * dims[1], dims[2], dims[3], dims[4],
dims[5]) # [BxK, 2, C, W, H]
x, y = e_in[:, 0, ...], e_in[:, 1, ...]
e_vectors = E(x, y).reshape(dims[0], dims[1], -1) # B, K, len(e)
e_hat = e_vectors.mean(dim=1)
# Generate frame using landmarks from frame t
x_t, y_t = t[:, 0, ...], t[:, 1, ...]
x_hat = G(y_t, e_hat)
# Optimize E_G and D
r_x_hat, D_act_hat = D(x_hat, y_t, i)
r_x, D_act = D(x_t, y_t, i)
optimizer_E_G.zero_grad()
optimizer_D.zero_grad()
loss_E_G = criterion_E_G(x_t, x_hat, r_x_hat, e_hat,
D.W[:, i].transpose(1, 0), D_act,
D_act_hat).mean()
loss_D = criterion_D(r_x, r_x_hat).mean()
loss = loss_E_G + loss_D
loss.backward()
optimizer_E_G.step()
optimizer_D.step()
# Optimize D again
x_hat = G(y_t, e_hat).detach()
r_x_hat, D_act_hat = D(x_hat, y_t, i)
r_x, D_act = D(x_t, y_t, i)
optimizer_D.zero_grad()
loss_D = criterion_D(r_x, r_x_hat).mean()
loss_D.backward()
optimizer_D.step()
batch_end = datetime.now()
batch_duration = batch_end - batch_start
batch_durations.append(batch_duration)
# SHOW PROGRESS --------------------------------------------------------------------------------------------
if (batch_num + 1) % 1 == 0 or batch_num == 0:
logging.info(
f'Epoch {epoch + 1}: [{batch_num + 1}/{len(dataset)}] | '
f'Time: {batch_duration} | '
f'Loss_E_G = {loss_E_G.item():.4} Loss_D {loss_D.item():.4}'
)
logging.debug(
f'D(x) = {r_x.mean().item():.4} D(x_hat) = {r_x_hat.mean().item():.4}'
)
# SAVE IMAGES ----------------------------------------------------------------------------------------------
save_image(
os.path.join(config.GENERATED_DIR, f'last_result_x.png'),
x_t[0])
save_image(
os.path.join(config.GENERATED_DIR, f'last_result_x_hat.png'),
x_hat[0])
if (batch_num + 1) % 1000 == 0:
save_image(
os.path.join(config.GENERATED_DIR,
f'{datetime.now():%Y%m%d_%H%M%S%f}_x.png'),
x_t[0])
save_image(
os.path.join(
config.GENERATED_DIR,
f'{datetime.now():%Y%m%d_%H%M%S%f}_x_hat.png'),
x_hat[0])
# SAVE MODELS ----------------------------------------------------------------------------------------------
if (batch_num + 1) % 100 == 0:
save_model(E, 'Embedder', gpu, run_start)
save_model(G, 'Generator', gpu, run_start)
save_model(D, 'Discriminator', gpu, run_start)
# SAVE MODELS --------------------------------------------------------------------------------------------------
save_model(E, 'Embedder', gpu, run_start)
save_model(G, 'Generator', gpu, run_start)
save_model(D, 'Discriminator', gpu, run_start)
epoch_end = datetime.now()
logging.info(
f'Epoch {epoch + 1} finished in {epoch_end - epoch_start}. '
f'Average batch time: {sum(batch_durations, timedelta(0)) / len(batch_durations)}'
)
def __init__(self, config):
nvidia_smi.nvmlInit()
self.handle = nvidia_smi.nvmlDeviceGetHandleByIndex(0)
self.config = config
if torch.cuda.is_available():
self.use_cuda = True
ngpu = config.n_gpu
# prepare folders
if not os.path.exists('./checkpoint_dir/' + config.model_name):
os.mkdir('./checkpoint_dir/' + config.model_name)
if not os.path.exists('./images/' + config.model_name):
os.mkdir('./images/' + config.model_name)
if not os.path.exists('./tb_log/' + config.model_name):
os.mkdir('./tb_log/' + config.model_name)
if not os.path.exists('./code_backup/' + config.model_name):
os.mkdir('./code_backup/' + config.model_name)
os.system('cp *.py ' + './code_backup/' + config.model_name)
# network
self.G = net.Generator(config).cuda()
self.D = net.Discriminator(config).cuda()
print('Generator structure: ')
print(self.G.model)
print('Discriminator structure: ')
print(self.D.model)
devices = [i for i in range(ngpu)]
self.G = MyDataParallel(self.G, device_ids=devices)
self.D = MyDataParallel(self.D, device_ids=devices)
self.start_resl = config.start_resl
self.max_resl = config.max_resl
self.load_model(G_pth=config.G_pth, D_pth=config.D_pth)
self.nz = config.nz
self.optimizer = config.optimizer
self.lr = config.lr
self.fadein = {'gen': None, 'dis': None}
self.upsam_mode = self.config.G_upsam_mode # either 'nearest' or 'tri-linear'
self.batchSize = {
2: 64 * ngpu,
3: 64 * ngpu,
4: 64 * ngpu,
5: 64 * ngpu,
6: 48 * ngpu,
7: 12 * ngpu
}
self.fadeInEpochs = {2: 0, 3: 1, 4: 1, 5: 2000, 6: 2000, 7: 2000}
self.stableEpochs = {2: 0, 3: 0, 4: 3510, 5: 10100, 6: 10600, 7: 50000}
self.ncritic = {2: 5, 3: 5, 4: 5, 5: 3, 6: 3, 7: 3}
# size 16 need 5000-7000 enough
# size 32 need 16000-30000 enough
self.global_batch_done = 0
# define all dataloaders into a dictionary
self.dataloaders = {}
for resl in range(self.start_resl, self.max_resl + 1):
self.dataloaders[resl] = DataLoader(
DL.Data(config.train_data_root + 'resl{}/'.format(2**resl)),
batch_size=self.batchSize[resl],
shuffle=True,
drop_last=True)
# ship new model to cuda, and update optimizer
self.renew_everything()
def main(input_path, training_mode='mle-gan'):
#Preprocessing data
data_obj = pr.Preprocessing()
data_obj.training_mode = training_mode
data_obj.run(input_path)
#Creating network object
#Parameters
input_size = len(data_obj.selected_columns)
batch = data_obj.batch_size
hidden_size = 200
num_layers = 5
num_directions = 1 # It should be 2 if we use bidirectional
beam_width = [1, 3, 5, 7, 10, 15] #Window size of beam search
#Creating Networks
enc = nw.Encoder(input_size, batch, hidden_size, num_layers,
num_directions).cuda()
dec = nw.Decoder(input_size, batch, hidden_size, num_layers,
dropout=.3).cuda()
dec.duration_time_loc = data_obj.duration_time_loc
rnnD = nw.Discriminator(input_size,
batch,
hidden_size,
num_layers,
dropout=.3).cuda()
model = nw.Seq2Seq(enc, dec).cuda()
#Initializing model parameters
model.apply(nw.init_weights)
rnnD.apply(nw.init_weights)
#Creating optimizers
optimizerG = torch.optim.RMSprop(model.parameters(), lr=5e-5)
optimizerD = torch.optim.RMSprop(rnnD.parameters(), lr=5e-5)
#Lets try several GPU
# if torch.cuda.device_count() > 1:
# print("Let's use", torch.cuda.device_count(), "GPUs!")
# model = torch.nn.DataParallel(model, device_ids= range(0, torch.cuda.device_count()))
# enc = torch.nn.DataParallel(enc, device_ids= range(0, torch.cuda.device_count()))
# dec = torch.nn.DataParallel(dec, device_ids= range(0, torch.cuda.device_count()))
# rnnD = torch.nn.DataParallel(rnnD, device_ids= range(0, torch.cuda.device_count()))
#--------------------------------------------
if (training_mode == 'mle'):
print("Training via MLE")
nw.train_mle(model, optimizerG, data_obj)
#Loading the best model saved during training
path = os.path.join(data_obj.output_dir, 'rnnG(validation entropy).m')
model.load_state_dict(torch.load(path))
nw.model_eval_test(model, data_obj, mode='test')
elif (training_mode == 'mle-gan'):
print("Training via MLE-GAN")
#Training via MLE-GAN
nw.train_gan(model, rnnD, optimizerG, optimizerD, data_obj)
# Loading the best model saved during training
path = os.path.join(data_obj.output_dir,
'rnnG(validation entropy gan).m')
model.load_state_dict(torch.load(path))
nw.model_eval_test(model, data_obj, mode='test')
#-------------------------------------------
#Generating suffixes
print("start generating suffixes using beam search!")
for i in beam_width:
sf.suffix_generate(model, data_obj, candidate_num=i)
sf.suffix_similarity(data_obj, beam_size=i)
return data_obj, model