total_epoch=5,
after_scheduler=scheduler)
optimizer.zero_grad()
optimizer.step()
lr_scheduler.step()
util.check_dir('../data/models/')
best_model, loss_dict, top1_acc_dict, top5_acc_dict = train_model(
data_loaders,
data_sizes,
name,
model,
criterion,
optimizer,
lr_scheduler,
num_epochs=num_epochs,
device=device)
# 保存最好的模型参数
# util.save_model(best_model.cpu(), '../data/models/best_%s.pth' % name)
res_loss[name] = loss_dict
res_top1_acc[name] = top1_acc_dict
res_top5_acc[name] = top5_acc_dict
print('train %s done' % name)
print()
util.save_png('loss', res_loss)
util.save_png('top-1 acc', res_top1_acc)
util.save_png('top-5 acc', res_top5_acc)
# print(model)
model = model.to(device)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=1e-3)
lr_schduler = optim.lr_scheduler.StepLR(optimizer,
step_size=8,
gamma=0.96)
util.check_dir('./data/models/')
best_model, loss_dict, acc_dict = train_model(data_loaders,
data_sizes,
name,
model,
criterion,
optimizer,
lr_schduler,
num_epochs=100,
device=device)
# 保存最好的模型参数
util.save_model(best_model, './data/models/best_%s.pth' % name)
res_loss[name] = loss_dict
res_acc[name] = acc_dict
print('train %s done' % name)
print()
util.save_png('loss', res_loss)
util.save_png('acc', res_acc)