def significance_analysis(result_dir):
    top_num_to_include_plot_data = {}
    network_size_plot_data = {}
    lr_plot_data = {}

    for label in os.listdir(result_dir):

        this_run_dir = f"{result_dir}/{label}"
        log_dir = get_log_dir(this_run_dir)
        save_dir = get_save_dir(this_run_dir)
        if not os.path.exists(log_dir) or len(os.listdir(save_dir)) == 0:
            continue

        top_num_to_include, network_size, lr = extra(label)
        if top_num_to_include not in top_num_to_include_plot_data:
            top_num_to_include_plot_data[top_num_to_include] = {
                "log_dirs": [log_dir],
                "labels": [label]
            }
        else:
            top_num_to_include_plot_data[top_num_to_include][
                "log_dirs"].append(log_dir)
            top_num_to_include_plot_data[top_num_to_include]["labels"].append(
                label)

        if network_size not in network_size_plot_data:
            network_size_plot_data[network_size] = {
                "log_dirs": [log_dir],
                "labels": [label]
            }
        else:
            network_size_plot_data[network_size]["log_dirs"].append(log_dir)
            network_size_plot_data[network_size]["labels"].append(label)

        if lr not in lr_plot_data:
            lr_plot_data[lr] = {"log_dirs": [log_dir], "labels": [label]}
        else:
            lr_plot_data[lr]["log_dirs"].append(log_dir)
            lr_plot_data[lr]["labels"].append(label)

    for network_size, data in network_size_plot_data.items():
        for lr, data_lr in lr_plot_data.items():
            title = f"fix network_size {network_size} learning_rate {lr}"

            all_data_network_size_set = list(
                zip(data["labels"], data["log_dirs"]))
            all_data_lr_set = list(zip(data_lr["labels"], data_lr["log_dirs"]))

            final_all_data = set(all_data_network_size_set).intersection(
                set(all_data_lr_set))
            final_data = {}
            print(f"network and lr num of runs: {len(final_all_data)}")

            if len(final_all_data) == 0:
                continue

            final_data["labels"], final_data["log_dirs"] = zip(*final_all_data)

            _sig_analysis(final_data["labels"], final_data["log_dirs"],
                          aug_num_timesteps, result_dir, title)
Exemplo n.º 2
0
def get_results(result_dir):
    labels = []
    log_dirs = []
    for label in os.listdir(result_dir):
        this_run_dir = f"{result_dir}/{label}"
        log_dir = get_log_dir(this_run_dir)
        if not os.path.exists(log_dir):
            continue

        labels.append(label)
        log_dirs.append(log_dir)

    return labels, log_dirs