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
kwargs['force_rebuild'] = False
kwargs['implicit'] = False
kwargs['save_local'] = True
kwargs['verbose'] = False
kwargs['split'] = False
datasets = [CiaoDVD, Delicious, LastFM, '100K', '1M', '10M']
URMs = []
path = os.path.join(DataReader.all_datasets_dir, 'stats')
if not os.path.exists(path):
os.makedirs(path)
for d in datasets:
if isinstance(d, str):
reader = Movielens(version=d, **kwargs)
else:
reader = d(**kwargs)
reader.describe(save_plots=True, path=path)
URMs.append(reader.get_URM_full().T.tocsr())
reader = Movielens('1M', **kwargs)
plot_long_tail(dataset=reader.get_URM_full().T.tocsr(),
label=reader.DATASET_NAME,
path=path)
plot_lorenz_curve(datasets=URMs,
labels=[
'MovieLens ' +
d if isinstance(d, str) else d.DATASET_NAME
for d in datasets
],
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
os.environ['KMP_WARNINGS'] = '0'
simplefilter(action='ignore', category=UserWarning)
simplefilter(action='ignore', category=FutureWarning)
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,