def __init__(self, opt, phase, transform=None, quiet=False):
if not quiet:
print('DataLoader loading h5 question file: ' +
opt['h5_ques_file'])
h5_file = h5py.File(opt['h5_ques_file'], 'r')
if phase is 'train':
if not quiet:
print('DataLoader loading h5 image train file: ' +
opt['h5_img_file_train'])
self.image = np.array(
h5py.File(opt['h5_img_file_train'], 'r')['/images_train'])
self.ques = np.array(h5_file['/ques_train'])
self.ques_len = np.array(h5_file['/ques_len_train'])
self.img_pos = np.array(h5_file['/img_pos_train'])
self.ques_id = np.array(h5_file['/ques_id_train'])
self.ans = np.array(h5_file['/answers'])
self.split = np.array(h5_file['/split_train'])
else: # valid or test
if not quiet:
print('DataLoader loading h5 image test file: ' +
opt['h5_img_file_test'])
self.image = np.array(
h5py.File(opt['h5_img_file_test'], 'r')['/images_test'])
self.ques = np.array(h5_file['/ques_test'])
self.ques_len = np.array(h5_file['/ques_len_test'])
self.img_pos = np.array(h5_file['/img_pos_test'])
self.ques_id = np.array(h5_file['/ques_id_test'])
self.ans = np.array(h5_file['/ans_test'])
self.split = np.array(h5_file['/split_test'])
self.feature_type = opt['feature_type']
self.phase = phase
self.transform = transform
if not quiet:
print('DataLoader loading json file: %s' % opt['json_file'])
json_file = utils.read_json(opt['json_file'])
self.ix_to_word = json_file['ix_to_word']
self.ix_to_ans = json_file['ix_to_ans']
self.vocab_size = utils.count_key(self.ix_to_word)
self.seq_length = self.ques.shape[1]
def config_tasks_envs(eparams):
'''
Configure tasks parameters.
Envs params and task parameters based on pearl paper:
args like followings will be added:
n_train_tasks 2
n_eval_tasks 2
n_tasks 2
randomize_tasks true
low_gear false
forward_backward true
num_evals 4
num_steps_per_task 400
num_steps_per_eval 400
num_train_steps_per_itr 4000
'''
configs = read_json(eparams.env_configs)[eparams.env_name]
temp_params = vars(eparams)
for k, v in configs.items():
temp_params[k] = v