add_graph_to_summary=False,
train_dataset_text=train_text,
validation_datasets=dict(valid=valid_text),
batch_size=BATCH_SIZE)
for conf in confs:
build_hyperparameters = dict(init_parameter=conf['init_parameter'])
# other_hyperparameters={'dropout': [.3, .5, .7, .8, .9, .95]},
other_hyperparameters = dict(
learning_rate=dict(varying=dict(init=conf['learning_rate']),
fixed=dict(decay=.5,
max_no_progress_points=10,
path_to_target_metric_storage=('valid',
'loss')),
hp_type='built-in',
type='adaptive_change'))
tf.set_random_seed(1)
_, biggest_idx, _ = get_num_exps_and_res_files(save_path)
if biggest_idx is None:
initial_experiment_counter_value = 0
else:
initial_experiment_counter_value = biggest_idx + 1
env.grid_search(
evaluation,
kwargs_for_building,
build_hyperparameters=build_hyperparameters,
other_hyperparameters=other_hyperparameters,
initial_experiment_counter_value=initial_experiment_counter_value,
**launch_kwargs)
),
train_batch_kwargs=dict(
valid_size=VALID_SIZE
),
valid_batch_kwargs=dict(
valid_size=VALID_SIZE
),
# train_dataset_text='abc',
validation_datasets=dict(
valid='validation'
),
batch_size=BATCH_SIZE
)
for conf in confs:
build_hyperparameters = dict(
init_parameter=conf['init_parameter'],
rho=[1.-v for v in conf['1_minus_rho']],
)
# other_hyperparameters={'dropout': [.3, .5, .7, .8, .9, .95]},
other_hyperparameters = dict()
tf.set_random_seed(1)
env.grid_search(
evaluation,
kwargs_for_building,
build_hyperparameters=build_hyperparameters,
other_hyperparameters=other_hyperparameters,
**launch_kwargs
)
