parser.add_argument("--gpu", type=str, required=True)
args = parser.parse_args()
experiment_type = args.experiment_type
gpu = args.gpu
print('STARTING EXPERIMENTS <{}> WITH VAE'.format(experiment_type))
device = 'cuda:{}'.format(gpu) if torch.cuda.is_available() else 'cpu'
data_path = DOWN_DATA_PATH if experiment_type == 'down_sample' else DATA_PATH
demo_path = DOWN_DEMO_PATH if experiment_type == 'down_sample' else DEMO_PATH
for fold_n in trange(5, desc='folds'):
log_val_str = LOG_VAL_STR.format('vae', experiment_type, now, fold_n)
log_te_str = LOG_TE_STR.format('vae', experiment_type, now, fold_n)
ds = DataSplitter(data_path, demo_path, out_dir=OUT_DIR)
pandas_dir_path, scipy_dir_path, uids_dic_path, tids_path = ds.get_paths(
fold_n=fold_n)
if experiment_type == 'up_sample':
ds.up_sample_train_data_path(pandas_dir_path, scipy_dir_path,
'gender')
# Setting seed for reproducibility
reproducible(EXP_SEED)
# --- Data --- #
tr_loader = DataLoader(LFM2bDataset(
scipy_dir_path,
type=str,
required=True,
choices=['standard', 'up_sample', 'down_sample'])
args = parser.parse_args()
experiment_type = args.experiment_type
print('STARTING EXPERIMENTS <{}> WITH ITEMKNN'.format(experiment_type))
data_path = DOWN_DATA_PATH if experiment_type == 'down_sample' else DATA_PATH
demo_path = DOWN_DEMO_PATH if experiment_type == 'down_sample' else DEMO_PATH
for fold_n in trange(5, desc='folds'):
log_val_str = LOG_VAL_STR.format('itemknn', experiment_type, now,
fold_n)
log_te_str = LOG_TE_STR.format('itemknn', experiment_type, now, fold_n)
ds = DataSplitter(data_path, demo_path, out_dir=OUT_DIR)
pandas_dir_path, scipy_dir_path, uids_dic_path, tids_path = ds.get_paths(
fold_n=fold_n)
if experiment_type == 'up_sample':
up_tr_data_path, up_sp_tr_data_path = ds.up_sample_train_data_path(
pandas_dir_path, scipy_dir_path, 'gender')
# --- Data --- #
sp_tr_data = sp.load_npz(
up_sp_tr_data_path if experiment_type ==
'up_sample' else os.path.join(scipy_dir_path, 'sp_tr_data.npz'))
sp_vd_tr_data = sp.load_npz(
os.path.join(scipy_dir_path, 'sp_vd_tr_data.npz'))
from utils.eval import eval_proced
print('STARTING CONTROLLED EXPERIMENTS WITH ESAE')
print('SEEDS ARE: {}'.format(SEEDS))
grid = {'lam': [1, 1e1, 1e2, 5e2, 1e3, 1e4, 1e5, 1e6, 1e7]}
pg = ParameterGrid(grid)
now = datetime.now()
for trait in tqdm(TRAITS, desc='traits'):
print('WORKING ON TRAIT: ' + trait)
for seed in tqdm(SEEDS, desc='seeds'):
log_val_str = LOG_VAL_STR.format('ease', now, trait, seed)
log_te_str = LOG_TE_STR.format('ease', now, trait, seed)
ds = DataSplitter(DATA_PATH, PERS_PATH, out_dir=OUT_DIR)
pandas_dir_path, scipy_dir_path, uids_dic_path, tids_path = ds.get_paths(
seed, trait)
# Load data
sp_tr_data = sp.load_npz(os.path.join(scipy_dir_path,
'sp_tr_data.npz'))
sp_vd_tr_data = sp.load_npz(
os.path.join(scipy_dir_path, 'sp_vd_tr_data.npz'))
sp_vd_te_data = sp.load_npz(
os.path.join(scipy_dir_path, 'sp_vd_te_data.npz'))
sp_te_tr_data = sp.load_npz(
os.path.join(scipy_dir_path, 'sp_te_tr_data.npz'))
sp_te_te_data = sp.load_npz(
experiment_type = args.experiment_type
gpu = args.gpu
print('STARTING EVALUATION <{}> WITH VAE'.format(experiment_type))
device = 'cuda:{}'.format(gpu) if torch.cuda.is_available() else 'cpu'
data_path = DOWN_DATA_PATH if experiment_type == 'down_sample' else DATA_PATH
demo_path = DOWN_DEMO_PATH if experiment_type == 'down_sample' else DEMO_PATH
experiment_datetime = best_configs[experiment_type]
if not experiment_datetime:
raise ValueError('Configuration <{}> for <{}> not found!'.format(experiment_datetime, experiment_type))
for fold_n in trange(5, desc='folds'):
log_val_str = LOG_VAL_STR.format('vae', experiment_type, experiment_datetime, fold_n)
log_te_str = LOG_TE_STR.format('vae', experiment_type, experiment_datetime, fold_n)
ds = DataSplitter(data_path, demo_path, out_dir=OUT_DIR)
pandas_dir_path, scipy_dir_path, uids_dic_path, tids_path = ds.get_paths(fold_n=fold_n)
if experiment_type == 'up_sample':
ds.up_sample_train_data_path(pandas_dir_path, scipy_dir_path, 'gender')
# Setting seed for reproducibility
reproducible(EXP_SEED)
# --- Data --- #
tr_loader = DataLoader(LFM2bDataset(scipy_dir_path, which='train', up_sample=experiment_type == 'up_sample'),
batch_size=64, shuffle=True, num_workers=10)
te_loader = DataLoader(LFM2bDataset(scipy_dir_path, pandas_dir_path, uids_dic_path, which='test'),
batch_size=128,
grid = {
"alpha": [5e-1, 1e-1, 1e-2, 1e-3],
"l1_ratio": [1, 1e-1, 1e-2],
"max_iter": [500]
}
pg = ParameterGrid(grid)
now = datetime.now()
for trait in tqdm(TRAITS, desc='traits'):
print('WORKING ON TRAIT: ' + trait)
for seed in tqdm(SEEDS, desc='seeds'):
log_val_str = LOG_VAL_STR.format('slim', now, trait, seed)
log_te_str = LOG_TE_STR.format('slim', now, trait, seed)
ds = DataSplitter(DATA_PATH, PERS_PATH, out_dir=OUT_DIR)
pandas_dir_path, scipy_dir_path, uids_dic_path, tids_path = ds.get_paths(
seed, trait)
# Load data
sp_tr_data = sp.load_npz(os.path.join(scipy_dir_path,
'sp_tr_data.npz'))
sp_vd_tr_data = sp.load_npz(
os.path.join(scipy_dir_path, 'sp_vd_tr_data.npz'))
sp_vd_te_data = sp.load_npz(
os.path.join(scipy_dir_path, 'sp_vd_te_data.npz'))
sp_te_tr_data = sp.load_npz(
os.path.join(scipy_dir_path, 'sp_te_tr_data.npz'))
sp_te_te_data = sp.load_npz(