def save_tensors(collected_tensors, output_directory):
filenames = []
for tensor_name, tensor_value in collected_tensors:
np_filename = os.path.join(output_directory, make_safe_filename(tensor_name) + ".npy")
np.save(np_filename, tensor_value.detach().cpu().numpy())
filenames.append(np_filename)
return filenames
def save_tensors(collected_tensors, experiment_dir_name):
filenames = []
for tensor_name, tensor_value in collected_tensors:
np_filename = os.path.join(experiment_dir_name,
make_safe_filename(tensor_name) + '.npy')
np.save(np_filename, tensor_value.numpy())
filenames.append(np_filename)
return filenames
def _save_as_numpy(predictions, output_directory, saved_keys, backend):
predictions = predictions[[
c for c in predictions.columns if c not in saved_keys
]]
npy_filename = os.path.join(output_directory, "{}.npy")
numpy_predictions = to_numpy_dataset(predictions, backend)
for k, v in numpy_predictions.items():
k = k.replace("<", "[").replace(
">", "]") # Replace <UNK> and <PAD> with [UNK], [PAD]
if k not in saved_keys:
if has_remote_protocol(output_directory):
with open_file(npy_filename.format(make_safe_filename(k)),
mode="wb") as f:
np.save(f, v)
else:
np.save(npy_filename.format(make_safe_filename(k)), v)
saved_keys.add(k)
def save_tensors(collected_tensors, experiment_dir_name):
for tensor_name, tensor_values in collected_tensors.items():
np_filename = os.path.join(experiment_dir_name,
make_safe_filename(tensor_name) + '.npy')
np.save(np_filename, tensor_values)
