h2=16,
embedding_dim=args.embedding_dim)
model.to(device=device)
# output dir
output_dir = "./results/{:%Y%m%d_%H%M%S}/".format(datetime.now())
output_path = output_dir + "model.weights"
if not os.path.exists(output_dir):
os.makedirs(output_dir)
# train
print("Training start ...")
start = time.time()
best_dev_score = train(model,
data,
output_path=output_path,
args=args,
loss_func=bpr_loss,
score_func=AUC)
print("Training process took {:.2f} seconds\n".format(time.time() - start))
# load best model and analysis on test set
model.load_state_dict(torch.load(output_path))
test_score = evaluate(model, data, AUC, args.batch_size, on='test')
dev_hit5, dev_hit10 = hit_eval(model, data, on='dev')
test_hit5, test_hit10 = hit_eval(model, data, on='test')
# write results
save_results('./' + args.dataset + '_deepfm.txt', args, best_dev_score,
test_score, dev_hit5, dev_hit10, test_hit5, test_hit10)
print('[init criterion]')
criterion = nn.BCEWithLogitsLoss()
print('[init engines]')
trainer = Trainer(model=model,
optimizer=optimizer,
criterion=criterion,
device=args.device)
evaluator = Evaluator(model=model, criterion=criterion, device=args.device)
print('[set handlers]')
set_handlers(trainer=trainer,
evaluator=evaluator,
valloader=valloader,
model=model,
optimizer=optimizer,
args=args)
if args.pretrained_path:
print('[load pretrained checkpoint]')
ckpt = torch.load(args.pretrained_path, map_location=args.device)
model.load_state_dict(ckpt['model'])
optimizer.load_state_dict(ckpt['optimizer'])
trainer.load_state_dict(ckpt['trainer'])
print('[start training]')
trainer.run(trainloader, max_epochs=args.max_epochs)
print('[done]')