#static_code = FloatTensor(np.zeros((opt.n_classes ** 2, opt.code_dim))))
# ----------
# Training
# ----------
for epoch in range(opt.n_epochs):
i = 0
for batch_gexs in data.get_unsupervised_batch():
i += 1
batch_size = batch_gexs.shape[0]
# Adversarial ground truths
valid = Variable( # noisy labels
FloatTensor(batch_size, 1).fill_(1.0) - FloatTensor(np.random.uniform(0.0, 0.05, (batch_size, 1))), requires_grad=False
)
fake = Variable(
FloatTensor(batch_size, 1).fill_(0.0) + FloatTensor(np.random.uniform(0.0, 0.05, (batch_size, 1))), requires_grad=False
)
# Configure input
#real_imgs = Variable(batch_gex.type(FloatTensor)).view(-1, 1024)
real_gexs = FloatTensor(batch_gexs)
#labels = to_categorical(labels.numpy(), num_columns=opt.n_classes)
# -----------------
# Train Generator
# -----------------
optimizer_G.zero_grad()
test_split=0.0) # train this unsupervised model on all data
# Optimizers
optimizer_G = torch.optim.Adam(generator.parameters(),
lr=opt.lr,
betas=(opt.b1, opt.b2))
optimizer_D = torch.optim.Adam(discriminator.parameters(),
lr=opt.lr,
betas=(opt.b1, opt.b2))
optimizer_info = torch.optim.Adam(itertools.chain(generator.parameters(),
discriminator.parameters()),
lr=opt.lr,
betas=(opt.b1, opt.b2))
# Static generator inputs for sampling
static_z = Variable(FloatTensor(np.zeros((opt.n_classes**2, opt.latent_dim))))
static_label = to_categorical(np.array(
[num for _ in range(opt.n_classes) for num in range(opt.n_classes)]),
num_columns=opt.n_classes)
static_code = Variable(FloatTensor(np.zeros((opt.n_classes**2, opt.code_dim))))
# ----------
# Training
# ----------
for epoch in range(opt.n_epochs):
i = 0
for batch_gexs in data.get_unsupervised_batch():
i += 1
batch_size = batch_gexs.shape[0]
discriminator = Discriminator(categorical_size, train_config.code_dim,
train_dataset.n_genes)
discriminator_checkpoint = os.path.join(load_path, 'discriminator.pth')
discriminator.load_state_dict(torch.load(discriminator_checkpoint))
if cuda:
#generator = generator.cuda().eval()
discriminator = discriminator.cuda().eval()
print('CUDA is here')
else:
print('CPU using')
codes, hidden, clusters, tags, valids = [], [], [], [], []
for gex in tqdm(test_dataset.traverse_gexs()):
with torch.no_grad():
batch_gex = FloatTensor(gex)
valid, cluster, pred_code = discriminator(batch_gex)
#hi_z = discriminator.hidden(batch_gex)
#valids.append(valid.detach().cpu().numpy())
codes.append(pred_code.detach().cpu().numpy())
#clusters.append(cluster.detach().cpu().numpy().argmax(1))
#hidden.append(hi_z.detach().cpu().numpy())
codes = np.vstack(codes)
#valids = np.vstack(valids)
#hidden = np.vstack(hidden)
#clusters = np.hstack(clusters)
patient_idxs = test_dataset.gexs.loc[
np.invert(test_dataset.gexs['sample_name'].isnull().values),
['sample_name']].values.reshape(-1)
#patient_idxs.shape
# Training
# ----------
summary = SummaryWriter('summary/s{0:02d}_c{1:02d}_{2}'.format(
config.code_dim, categorical_size,
os.path.split(config.gexs_csv)[-1].split('.')[0]))
for epoch in range(config.n_epochs):
print('\repoch {0}'.format(epoch + 1), end='', flush=True)
i = 0
for batch_gexs in dataset.get_gexs_batch(config.batch_size):
i += 1
batch_size = batch_gexs.shape[0]
# Adversarial ground truths
valid = Variable( # noisy labels
FloatTensor(batch_size, 1).fill_(1.0) -
FloatTensor(np.random.uniform(0.0, 0.05, (batch_size, 1))),
requires_grad=False)
fake = Variable(
FloatTensor(batch_size, 1).fill_(0.0) +
FloatTensor(np.random.uniform(0.0, 0.05, (batch_size, 1))),
requires_grad=False)
# Configure input
real_gexs = FloatTensor(batch_gexs)
# -----------------
# Train Generator
# -----------------
optimizer_G.zero_grad()