type='exponential_decay' ) ) 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_for_meta( evaluation, kwargs_for_pupil_building, kwargs_for_optimizer_building, build_pupil_hyperparameters=build_pupil_hyperparameters, build_optimizer_hyperparameters=build_optimizer_hyperparameters, other_hyperparameters=other_hyperparameters, initial_experiment_counter_value=initial_experiment_counter_value, **launch_kwargs ) hp_names = get_hp_names_from_conf_file(parameter_set_file_name) for_plotting = get_optimizer_evaluation_results(save_path, hp_names, AVERAGING_NUMBER) best = get_best(for_plotting, 'optimizer') metric_res = best['adam_prep']['loss'] best_on_valid = metric_res['validation'] print(' ' * 2 + 'loss' + ':', best_on_valid[1])
# opt_inf_pupil_restore_paths={ # 'prelearn2000': 'lstm/test_res_net_1000_emb150_nl1_nn100_bs32_nu10/checkpoints/2000' # }, # opt_inf_additions_to_feed_dict=opt_inf_add_feed, # opt_inf_validation_dataset_texts=[valid_text], # opt_inf_train_dataset_texts=[train_text], # validation_additions_to_feed_dict=valid_add_feed, vocabulary=vocabulary, batch_size=32, num_unrollings=4, learning_rate={ 'type': 'exponential_decay', 'init': .002, 'decay': .5, 'period': 400 }, results_collect_interval=10, opt_inf_results_collect_interval=1, permute=False, summary=True, add_graph_to_summary=True) env.grid_search_for_meta( evaluation, kwargs_for_pupil_building, kwargs_for_optimizer_building, build_pupil_hyperparameters=build_pupil_hyperparameters, build_optimizer_hyperparameters=build_optimizer_hyperparameters, other_hyperparameters=other_hyperparameters, **launch_kwargs)