def train(model, data_loader, parameters, name, save_best_ckpt):
# Train the given model and evaluate it
print("Start Train Model")
trainer = Trainer(model, data_loader, parameters['criterion'], parameters['lr'], parameters['betas'],
parameters['epochs'], parameters['batch_size'], parameters['num_classes'],
parameters['epsilon'], name, parameters['path'])
trainer.__train__(save_best_ckpt)
if save_best_ckpt:
# if len(trainer.model_states) > 0:
# trainer.model_states = sorted(trainer.model_states, key=lambda x: x['loss'])
# torch.save({'model_state_dict': trainer.model_states[0]['state_dict']}, trainer.ckpt)
# else:
trainer.model.save(trainer.ckpt)
plot(trainer.losses, "{} Loss".format(trainer.name), "loss", "epoch", parameters['path_plots_nn'])
plot(trainer.accuracies, "{} Accuracy".format(trainer.name), "accuracy", "epoch", parameters['path_plots_nn'])
print("End Train Model")
print("==============================================================================")
return trainer
def playing_with_learning_rate(train_loader, parameters):
print("Start part 2")
# Question 2 - Playing with learning rate
print("Playing with learning rate")
models = [SimpleModel() for _ in range(len(parameters['lrs']))]
for idx in range(len(parameters['lrs'])):
models[idx].load(parameters['pretrained_path'])
model_name = "model_{}".format(idx)
lr = parameters['lrs'][idx]
print("model: {}. lr: {}".format(model_name, lr))
trainer = Trainer(models[idx], train_loader, parameters['criterion'],
lr, parameters['betas'], parameters['epochs'],
parameters['batch_size'], parameters['num_classes'],
parameters['epsilon'], model_name,
parameters['path_lrs'])
trainer.__train__(False)
plot(trainer.losses, "{} Loss".format(trainer.name), "loss", "epoch",
parameters['path_lrs'])
plot(trainer.accuracies, "{} Accuracy".format(trainer.name),
"accuracy", "epoch", parameters['path_lrs'])
print("End Playing with learning rate")
print("End part 2")
