def objective(trial):
start = time.time()
import gc
gc.collect()
score_sum = 0
data_dirs = [
'../' + data_path + '/valid1/bpr', '../' + data_path + '/valid2/bpr'
]
for data_dir in data_dirs:
dataset = dataloader.AmazonDataset(data_dir)
embedding_dim = trial.suggest_discrete_uniform('embedding_dim', 16,
128, 16)
user_size = len(dataset.user_list)
item_size = len(dataset.item_list)
layer_size = trial.suggest_int('layer_size', 1, 3)
nfm = model.NFM(int(embedding_dim), user_size, item_size,
layer_size).to(device)
batch_size = int(
trial.suggest_discrete_uniform('batch_size', 128, 512, 128))
lr = trial.suggest_loguniform('lr', 1e-4, 1e-2)
weight_decay = trial.suggest_loguniform('weight_decay', 1e-6, 1e-2)
warmup = 350
lr_decay_every = 2
lr_decay_rate = trial.suggest_uniform('lr_decay_rate', 0.5, 1)
iterater = training.TrainIterater(batch_size=batch_size,
data_dir=data_dir)
score = iterater.iterate_epoch(nfm,
lr,
epoch=2000,
weight_decay=weight_decay,
warmup=warmup,
lr_decay_rate=lr_decay_rate,
lr_decay_every=lr_decay_every,
eval_every=1e+5,
early_stop=True)
torch.cuda.empty_cache()
score_sum += score
mi, sec = time_since(time.time() - start)
print('{}m{}sec'.format(mi, sec))
return -1 * score_sum / 2
dataset = dataloader.AmazonDataset(data_dir)
#print(dataset.user_items_test_dict)
embedding_dim = params['embedding_dim']
user_size = len(dataset.user_list)
item_size = len(dataset.item_list)
layer_size = int(params['layer_size'])
batch_size = int(params['batch_size'])
lr = params['lr']
weight_decay = params['weight_decay']
warmup = 350
lr_decay_every = 2
lr_decay_rate = params['lr_decay_rate']
nfm = model.NFM(int(embedding_dim), user_size, item_size,
layer_size).to(device)
iterater = training.TrainIterater(batch_size=batch_size, data_dir=data_dir)
score = iterater.iterate_epoch(nfm,
lr,
epoch=300,
weight_decay=weight_decay,
warmup=warmup,
lr_decay_rate=lr_decay_rate,
lr_decay_every=lr_decay_every,
eval_every=3)
#evaluater = evaluate.Evaluater(data_dir)
#score = evaluater.topn_map(nfm)
#print(score)
batch_size=args.batch_size, shuffle=False, num_workers=0) ############################## CREATE MODEL ########################### if args.pre_train: assert os.path.exists(config.FM_model_path), 'lack of FM model' assert config.model == 'NFM', 'only support NFM for now' FM_model = torch.load(config.FM_model_path) else: FM_model = None if config.model == 'FM': model = model.FM(num_features, args.hidden_factor, args.batch_norm, eval(args.dropout)) else: model = model.NFM( num_features, args.hidden_factor, config.activation_function, eval(args.layers), args.batch_norm, eval(args.dropout), FM_model) model.cuda() if config.optimizer == 'Adagrad': optimizer = optim.Adagrad( model.parameters(), lr=args.lr, initial_accumulator_value=1e-8) elif config.optimizer == 'Adam': optimizer = optim.Adam(model.parameters(), lr=args.lr) elif config.optimizer == 'SGD': optimizer = optim.SGD(model.parameters(), lr=args.lr) elif config.optimizer == 'Momentum': optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=0.95) if config.loss_type == 'square_loss': criterion = nn.MSELoss(reduction='sum') else: # log_loss
return loss_total / len(self.user_item_train_df)
def valid_loss(self, batch, loss_func, model):
with torch.no_grad():
batch, y_train = batch
user_tensor = torch.tensor(batch[:, 0],
dtype=torch.long,
device=device)
item_tensor = torch.tensor(batch[:, 1],
dtype=torch.long,
device=device)
y_train = torch.tensor(y_train, dtype=torch.float, device=device)
pred = model(user_tensor, item_tensor)
loss = loss_func(pred, y_train)
return loss
def valid_metric(self, model):
return 0
if __name__ == '__main__':
import model
dataset = AmazonDataset('../data_beauty_2core_es/valid1/bpr/')
user_size = len(dataset.user_list)
item_size = len(dataset.item_list)
nfm = model.NFM(32, user_size, item_size, 2)
es = EarlyStop('../data_beauty_2core_es/early_stopping/bpr', 10)
es.early_stop(nfm)