test_ext=repr_test_x_ext,
dtype=args.repr_x_dtype)
if decode:
repr_fold_y_splits = load_train_val_test_splits(args.repr_y,
dataset_name,
y_only=True,
train_ext=repr_train_y_ext,
valid_ext=repr_valid_y_ext,
test_ext=repr_test_y_ext,
dtype=args.repr_y_dtype)
#
# printing
logging.info('Original folds')
print_fold_splits_shapes(fold_splits)
logging.info('Repr X folds')
print_fold_splits_shapes(repr_fold_x_splits)
if decode:
logging.info('Repr Y folds')
print_fold_splits_shapes(repr_fold_y_splits)
#
# Opening the file for test prediction
#
if args.exp_name:
out_path = os.path.join(args.output, dataset_name + '_' + args.exp_name)
else:
date_string = datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
out_path = os.path.join(args.output, dataset_name + '_' + date_string)
n_folds,
train_ext=train_ext,
valid_ext=valid_ext,
test_ext=test_ext,
dtype=args.dtype)
else:
fold_splits = load_train_val_test_splits(args.dataset,
dataset_name,
x_only=False,
y_only=False,
train_ext=train_ext,
valid_ext=valid_ext,
test_ext=test_ext,
dtype=args.dtype)
print_fold_splits_shapes(fold_splits)
merged_fold_splits = []
for i, splits in enumerate(fold_splits):
logging.info('Processing fold {}\n'.format(i))
merged_splits = []
for j, split in enumerate(splits):
if split is not None:
split_x, split_y = split
logging.info('\tProcessing split {} ({}, {})'.format(
SPLIT_NAMES[j], split_x.shape, split_y.shape))
if args.n_classes:
logging.info('classes : {} -> {}'.format(
y_only=y_only,
dtype=args.dtype)
else:
fold_splits = load_train_val_test_splits(args.dataset,
dataset_name,
train_ext=train_ext,
valid_ext=valid_ext,
test_ext=test_ext,
x_only=x_only,
y_only=y_only,
dtype=args.dtype)
#
# printing
print_fold_splits_shapes(fold_splits)
# n_instances = train.shape[0]
# n_test_instances = test.shape[0]
#
# estimating the frequencies for the features
logging.info('Estimating features on training set...')
# freqs, features = dataset.data_2_freqs(train)
n_features = fold_splits[0][0].shape[1]
features = None
if args.feature_scheme is None:
features = numpy.array([2 for i in range(n_features)])
else:
raise ValueError('Loading feature schema not implemented yet')
#