matlab、numpy的indexing差异

用 index 数组从数据集里抽取一批数据，matlab 的写法与 python numpy 的写法不同：matlab 除了在 sample 那一维放索引数组，还要在其余维度加冒号 :（表示该维全选？）。

numpy

import numpy as np


images = np.zeros([5, 32, 32, 3])  # [n, H, W, C]
labels = np.ones([5, 10])  # [n, c]
indices = np.array([0, 2, 1])  # 0-base
print("indices:", indices)  # [0 2 1]

image_batch = images[indices]
label_batch = labels[indices]
print("image batch:", image_batch.shape)  # (3, 32, 32, 3)
print("label batch:", label_batch.shape)  # (3, 10)

matlab

images = zeros(5, 32, 32, 3);  % [n, H, W, C]
labels = ones(5, 10);  % [n, c]
indices = int8([0, 2, 1])' + 1;  % 1-base
fprintf("indices:"), disp(indices);  % [1, 3, 2]


image_batch = images(indices);
label_batch = labels(indices);
fprintf("image batch 1:"), disp(size(image_batch));  % (3, 1)
fprintf("label batch 1:"), disp(size(label_batch));  % (3, 1)

image_batch = images(indices, :);
label_batch = labels(indices, :);
fprintf("image batch 2:"), disp(size(image_batch));  % (3, 3072), 其中 3072 = 32 * 32 * 3
fprintf("label batch 2:"), disp(size(label_batch));  % (3, 10)

image_batch = images(indices, :, :);
fprintf("image batch 3:"), disp(size(image_batch));  % (3, 32, 96), 其中 96 = 32 * 3

image_batch = images(indices, :, :, :);
fprintf("image batch 4:"), disp(size(image_batch));  % (3, 32, 32, 3)