def fit(self, args, model_dir):
init_variables(self.sess, self.saver, args.load_model, model_dir)
# Create evaluators and early_stoppings
# evaluators: for different eval_set
self.evaluators = {}
if args.eval_epoch is not None:
self.evaluators = create_evaluators(self.dataset, args.eval_types,
args.metrics, args.ks,
args.n_thread)
self.early_stopping = EarlyStopping(args.early_stop)
# Start training and evaluation.
print_seperate_line()
if args.eval_epoch is not None:
self.eval(args)
print_seperate_line()
self.int_weight_v = args.int_weight
self.pop_weight_v = args.pop_weight
for epoch in range(1, args.epoch + 1):
self.train_1_epoch(epoch, args)
self.int_weight_v *= args.loss_decay
self.pop_weight_v *= args.loss_decay
self.sampler.margin *= args.margin_decay
if args.eval_epoch is not None and epoch % args.eval_epoch == 0:
print_seperate_line()
self.eval(args)
print_seperate_line()
if self.early_stopping.check_stop(self.evaluators, epoch):
break
print(self.early_stopping)
print_seperate_line()
# Save model.
if args.save_model:
save_model(self.sess, self.saver, args.verbose_name, args.epoch,
model_dir)
import os
from recq.recommenders.dice import DICE
from recq.utils import Dataset
from recq.tools.parser import parse_args
from recq.tools.io import print_seperate_line
args = parse_args("dice")
print_seperate_line()
for key, value in vars(args).items():
print(key + '=' + str(value))
print_seperate_line()
curr_dir = os.path.dirname(__file__)
if args.gpu_id is not None:
os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu_id)
data_dir = os.path.join(curr_dir, "data", args.dataset)
model_dir = os.path.join(curr_dir, "output", "model")
dataset = Dataset(args, data_dir)
model = DICE(args, dataset)
model.fit(args, model_dir)