NumPy数组切片维度疑问：为何两种切片分别返回列与行？

Hey there! Let's break down your NumPy slicing questions clearly, step by step.

It’s not a fixed answer—it all depends on how you structure your slice, since NumPy operates around axes (axis 0 = rows, axis 1 = columns for 2D arrays):

• If you slice only along axis 0 (e.g., x[1:3]), you’ll get a subset of rows.
• If you slice only along axis 1 (e.g., x[:, 1:3]), you’ll get a subset of columns.
• The key detail is whether your slice preserves the original array’s dimensions—this is exactly what’s happening in your second question.

Let’s start by re-creating your setup to make things concrete:

import numpy as np
x = np.array([[ 0, 1, 2],[ 3, 4, 5],[ 6, 7, 8],[ 9, 10, 11]])

Case a: z = x[0:4:1,0:1:1] #

Here, you’re using slice ranges on both axes (0:4:1 for rows, 0:1:1 for columns). Even when a slice range only includes one element (like 0:1:1 for the first column), NumPy preserves the original axis’s dimension.

The result is a 2D array with shape (4, 1)—which looks like a column vector:

[[0]
 [3]
 [6]
 [9]]

You can confirm this with print(z.shape), which outputs (4, 1).

Case b: z = x[0:4 ,0] #

Here, you use a slice range on axis 0, but a single index (0) on axis 1. When you index a single position along an axis, NumPy automatically "squeezes" that axis out of the result—reducing the array’s dimensionality by 1.

The result is a 1D array with shape (4,), which looks like a row vector:

[0 3 6 9]

Check with print(z.shape) and you’ll see (4,).

Quick tip: How to get a column vector from case b? #

If you want to keep the 2D structure when selecting a single column, you have a few options:

• Use slice syntax for the column: x[:, 0:1] (same logic as case a)
• Add a new axis explicitly: x[:, 0, np.newaxis]
• Use list indexing: x[:, [0]] (list indexes preserve dimension even for single elements)

内容的提问来源于stack exchange，提问作者Daniel Garcia De Cceres Herran