print('model sayz:')
print('I am Loaded')
temp_list = [0.5, 1., 2.]
policy_list = ['greedy', 'temp']
seq_length = 111
alice_string = list('Alice')
# Generate some sentences by sampling from the model
for policy in policy_list:
for temperature in temp_list:
char_list = []
hidden = None
for alice in alice_string:
idx = dataset.convert_to_idx(alice)
char_list.append(idx)
generator = torch.tensor([idx]).unsqueeze(-1)
generator = generator.to(device)
generator, hidden = model.forward(generator, hidden)
for _ in range(seq_length):
if policy == 'greedy':
idx = torch.argmax(generator).item()
else:
temp = generator.squeeze() / temperature
soft = torch.softmax(temp, dim=0)
idx = torch.multinomial(soft, 1).item()
generator = torch.tensor([idx]).unsqueeze(-1)
generator = generator.to(device)
generator, hidden = model.forward(generator, hidden)
def train(config):
def acc(predictions, targets):
hotvec = predictions.argmax(-2) == targets
accuracy = torch.mean(hotvec.float())
return accuracy
# Initialize the device which to run the model on
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
# Initialize the dataset and data loader (note the +1)
dataset = TextDataset(config.txt_file, config.seq_length)
data_loader = DataLoader(dataset, config.batch_size, num_workers=0)
print('batch', config.batch_size)
vocabulary_size = dataset.vocab_size
print('vocab', vocabulary_size)
# Initialize the model that we are going to use
model = TextGenerationModel(config.batch_size,
config.seq_length,
vocabulary_size=vocabulary_size,
lstm_num_hidden=config.lstm_num_hidden,
lstm_num_layers=config.lstm_num_layers,
dropout=1 - config.dropout_keep_prob,
device=device)
model = model.to(device)
# Setup the loss and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(),
lr=config.learning_rate,
weight_decay=1e-5)
gamma = 1 - config.learning_rate_decay
lr_optim = torch.optim.lr_scheduler.StepLR(optimizer,
config.learning_rate_step,
gamma=gamma,
last_epoch=-1)
print('Hi')
acc_list = []
loss_list = []
step_list = []
text_list = []
epoch = 100
offset = 2380
temperature = 1
policy = 'greedy'
for e in range(epoch):
torch.save(model.state_dict(), str(e + 1) + 'tunedmodel.pt')
for step, (batch_inputs, batch_targets) in enumerate(data_loader):
# Only for time measurement of step through network
lr_optim.step()
optimizer.zero_grad()
t1 = time.time()
inputs = torch.stack([*batch_inputs], dim=1)
targets = torch.stack([*batch_targets], dim=1)
inputs = inputs.to(device)
targets = targets.to(device)
out = model.forward(inputs)[0]
out = out.permute(0, 2, 1)
loss = criterion(out, targets)
accuracy = acc(out, targets)
torch.nn.utils.clip_grad_norm(model.parameters(),
max_norm=config.max_norm)
loss.backward()
optimizer.step()
# Just for time measurement
t2 = time.time()
examples_per_second = config.batch_size / float(t2 - t1)
if step % config.print_every == 0:
print('accuracy, loss, step: \n',
np.around(accuracy.item(), 4), np.around(loss.item(),
4), step, '\n')
acc_list.append(accuracy.item())
loss_list.append(loss.item())
step_list.append(step + offset * e)
if step % config.sample_every == 0:
# Generate some sentences by sampling from the model
generator = torch.randint(low=0,
high=vocabulary_size,
size=(1, 1)).to(device)
hidden = None
char_list = []
for _ in range(config.seq_length):
generator, hidden = model.forward(generator, hidden)
if policy == 'greedy':
idx = torch.argmax(generator).item()
else:
pass
generator = torch.Tensor([idx]).unsqueeze(-1)
generator = generator.to(device)
char_list.append(idx)
char = dataset.convert_to_string(char_list)
with open("MyTunedBook.txt", "a") as text_file:
print('Epoch. ',
e,
'Stahp: ',
step,
'\n Output: ',
char,
file=text_file)
print('Epoch. ', e, 'Stahp: ', step, '\n Output: ', char)
text_list.append((str((step + offset * e)) + '\n' + char))
pass
if step == config.train_steps:
# If you receive a PyTorch data-loader error, check this bug report:
# https://github.com/pytorch/pytorch/pull/9655
break
print('Done training.')
with open('FinalTunedBook.txt', 'w+') as f:
for item in text_list:
f.write("%s\n" % item)
# save with pandas
header = ['accuracy', 'length', 'loss', 'step']
savefiles = zip(acc_list, [config.seq_length] * len(acc_list), loss_list,
step_list)
df = pd.DataFrame(list(savefiles), columns=header)
df.to_csv('GEN' + str(config.seq_length) + 'tunedlstm.csv')
print('I am Loaded')
temp_list = [0., 0.5, 1., 2.]
policy_list = ['greedy', 'temp']
seq_length = 111
alice_string = list('Alice')
# Generate some sentences by sampling from the model
for policy in policy_list:
for temperature in temp_list:
char_list = []
hidden = None
for alice in alice_string:
idx = dataset.convert_to_idx(alice)
char_list.append(idx)
generator = torch.tensor([idx]).unsqueeze(-1)
generator = generator.to(device)
generator, hidden = model.forward(generator, hidden)
for _ in range(seq_length):
if policy == 'greedy':
idx = torch.argmax(generator).item()
else:
temp = generator.squeeze() / temperature
soft = torch.softmax(temp, dim=0)
idx = torch.multinomial(soft, 1)[-1].item()
generator = torch.tensor([idx]).unsqueeze(-1)
generator = generator.to(device)
generator, hidden = model.forward(generator, hidden)
char_list.append(idx)
char = dataset.convert_to_string(char_list)
with open(
"BonusTemp" + str(int(np.floor(temperature))) + "Book.txt",
"w+") as text_file:
print(policy + ': ',
temperature,
'\n Output: ',
char,
file=text_file)
print(policy + ': ', temperature, '\n Output: ', char)
print('Finito!')
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
WonderLand = torch.load('model/49tunedmodelCP.pt',
map_location=torch.device(device))
print('I am Loaded')
temp_list = [0., 0.5, 1., 2.]
policy = 'temp'
seq_length = 30
txt_file = 'alice.txt'
dataset = TextDataset(txt_file, seq_length)
test_idx = dataset.convert_to_idx('A')
print(test_idx)
test_idx = dataset.convert_to_idx('Alice')
print(test_idx)
vocabulary_size = dataset.vocab_size
# Generate some sentences by sampling from the model
for temperature in temp_list:
generator = torch.randint(low=0, high=vocabulary_size,
size=(1, 1)).to(device)
hidden = None
char_list = [generator.item()]
for _ in range(seq_length):
generator, hidden = WonderLand.forward(generator, hidden)
if policy == 'greedy':
