def cycleclass()

in ebm_sandbox.py [0:0]

• 25 lines of code
• 7 McCabe index (conditional complexity)

def cycleclass(dataloader, weights, model, target_vars, logdir, sess):
    # X, Y_GT, X_final, X_targ = target_vars['X'], target_vars['Y_GT'], target_vars['X_final'], target_vars['X_targ']
    X, Y_GT, X_final = target_vars['X'], target_vars['Y_GT'], target_vars['X_final']
    for data_corrupt, data, label_gt in tqdm(dataloader):
        data, label_gt = data.numpy(), label_gt.numpy()
        data_corrupt = data_corrupt.numpy()


        data_mods = []
        x_curr = data_corrupt
        x_target = np.random.uniform(0, 1, data_corrupt.shape)
        # x_target = np.tile(x_target, (1, 32, 32, 1))


        for i in range(20):
            feed_dict = {X: x_curr, Y_GT: label_gt}
            x_curr_new = sess.run(X_final, feed_dict)
            x_curr = x_curr_new
            data_mods.append(x_curr_new)

            if i > 30:
                x_target = np.random.uniform(0, 1, data_corrupt.shape)

        data_corrupt, data = rescale_im(data_corrupt), rescale_im(data)

        data_mods = [rescale_im(data_mod) for data_mod in data_mods]

        panel_im = np.zeros((32*100, 32*(len(data_mods) + 2), 3)).astype(np.uint8)

        for i in range(100):
            panel_im[32*i:32*i+32, :32] = data_corrupt[i]

            for j in range(len(data_mods)):
                panel_im[32*i:32*i+32, 32*(j+1):32*(j+2)] = data_mods[j][i]

            panel_im[32*i:32*i+32, -32:] = data[i]

        imsave(osp.join(logdir, "cycleclass.png"), panel_im)
        assert False