4725d18c4d23dfa1c598026a67c0620006a39dd9,rbm/run_rbm.py,,benchmark,#Any#Any#Any#Any#Any#Any#Any#,37

Before Change

    with open(output_file, "a") as f:
        writer = csv.DictWriter(f, ["L", "k", "time"])
        for L in sorted(dataset_sizes):
            for k_ in k:
                print("L = {}; k = {}".format(L, k_))
                train_set = load_train(L)

                num_hidden = (train_set.shape[-1]
                              if num_hidden is None
                              else num_hidden)

                rbm = RBM(num_visible=train_set.shape[-1],
                          num_hidden=num_hidden)

                time_elapsed = -time.perf_counter()
                rbm.train(train_set, epochs,
                          batch_size, k=k_,
                          lr=learning_rate,
                          momentum=momentum,
                          initial_gaussian_noise=0,
                          log_every=0,
                          progbar=False)
                time_elapsed += time.perf_counter()

                writer.writerow({"L": L,
                                 "k": k_,
                                 "time": time_elapsed})


@cli.command("train")
@click.option("--train-path", default="../data/Ising2d_L4.pkl.gz",
              show_default=True, type=click.Path(exists=True),
              help="path to the training data")

After Change

                     for f in listdir("/home/data/critical-2d-ising/")
                     if isdir(join("/home/data/critical-2d-ising/", f))]

    runs = list(product(dataset_sizes, k))

    shuffle(runs)

    with open(output_file, "a") as f:
        writer = csv.DictWriter(f, ["L", "k", "time"])
        for L, k_ in runs:
            print("L = {}; k = {}".format(L, k_))
            train_set = load_train(L)

            num_hidden = (train_set.shape[-1]
                          if num_hidden is None
                          else num_hidden)

            rbm = RBM(num_visible=train_set.shape[-1],
                      num_hidden=num_hidden)

            time_elapsed = -time.perf_counter()
            rbm.train(train_set, epochs,
                      batch_size, k=k_,
                      lr=learning_rate,
                      momentum=momentum,
                      initial_gaussian_noise=0,
                      log_every=0,
                      progbar=False)
            time_elapsed += time.perf_counter()
            torch.cuda.empty_cache()

            writer.writerow({"L": L,
                             "k": k_,
                             "time": time_elapsed})


@cli.command("train")
@click.option("--train-path", default="../data/Ising2d_L4.pkl.gz",
              show_default=True, type=click.Path(exists=True),
              help="path to the training data")

Instances