def make_dataset(dataset, specs):
set_seed(
seed
) # Need to set this so each dataset is created the same in any machine/order selected
if isinstance(dataset, str) and dataset in Movielens.urls.keys():
reader = Movielens(version=dataset, **specs)
else:
reader = dataset(**specs)
sets = []
URM_train = reader.get_URM_train()
URM_test = reader.get_URM_test()
URM_for_train, _, URM_validation = reader.split_urm(
URM_train.tocoo(),
split_ratio=[0.75, 0, 0.25],
save_local=False,
verbose=False,
min_ratings=1)
URM_train_small, _, URM_early_stop = reader.split_urm(
URM_for_train.tocoo(),
split_ratio=[0.85, 0, 0.15],
save_local=False,
verbose=False,
min_ratings=1)
sets.extend(
[URM_train, URM_test, URM_validation, URM_train_small, URM_early_stop])
for suf, urm in zip(URM_suffixes, sets):
sps.save_npz(os.path.join(exp_path, reader.DATASET_NAME + suf),
urm,
compressed=True)
return sets
use_gpu = False
verbose = False
only_build = False
transposed = False
if not use_gpu:
os.environ['CUDA_VISIBLE_DEVICES'] = ''
reader = Movielens(version='100K',
split_ratio=[0.6, 0.2, 0.2],
use_local=True,
implicit=True,
verbose=False,
seed=seed)
URM_train = reader.get_URM_train(transposed=transposed)
URM_validation = reader.get_URM_validation(transposed=transposed)
URM_test = reader.get_URM_test(transposed=transposed)
evaluator = EvaluatorHoldout(URM_test, [5, 20], exclude_seen=True)
evaluatorValidation = EvaluatorHoldout(URM_validation, [5], exclude_seen=True)
gan = GANMF(URM_train, mode='user')
gan.fit(num_factors=10,
emb_dim=128,
d_reg=1e-4,
g_reg=0,
epochs=300,
batch_size=128,
g_lr=1e-3,