train = True
test = True
model = None
if train:
# Create models:
model = my_models.model_2()
my_models.test_model_size(model, dtype) # test model size output:
optimizer = optim.Adadelta(model.parameters())
# Train model:
model, loss_data = my_models.train_model(model,
optimizer,
train_loader,
val_loader,
device,
dtype,
epoches=2)
# Save model to file:
torch.save(model.state_dict(), MODEL_PATH + MODEL_NAME)
# Save loss data to file:
np.savetxt(LOSS_PATH + 'data_part_4_train.csv', loss_data, delimiter=',')
fix_loss_data('data_part_4_train.csv')
print('Training is Finished !')
if test:
print('Checking model Accuracy over Test Set...')
if model is None:
# load model:
dtype = torch.float32 # we will be using float
train = False
train_loops = 20
loss_data = []
if train:
for tl in range(train_loops):
# Create models:
model = my_models.model_2()
#my_models.test_model_size(model, dtype) # test model size output:
optimizer = optim.Adadelta(model.parameters())
# Train model:
model, current_loss_data = my_models.train_model(model, optimizer, train_loader, test_loader, device, dtype, epoches=4, print_every=5)
loss_data.append(current_loss_data)
# Save model to file:
#torch.save(model.state_dict(), MODEL_PATH + MODEL_NAME)
print('Training Loop {}/{} is Finished !'.format(tl + 1, train_loops))
print()
# Add test accuracy and save data to file:
loss_data = np.asarray(loss_data)
loss_data = fix_loss_array(loss_data)
# Saving Accuracy to file
np.save(LOSS_PATH + 'data_part_2_acc.npy', loss_data)