II. NumPy Array Operations (数组操作)#

Array operations (数组操作) let you change an array's shape (形状), dimensions (维度), and structure — without touching the underlying data values. These are the core tools for preparing data before computation.

1. `reshape()` — Change Shape Without Copying (改变形状)#

Core idea: Returns a new view of the same data with a different shape. Total elements must stay the same.

1
import numpy as np
2

3
a = np.arange(12)          # [0, 1, 2, ..., 11]
4
b = a.reshape(3, 4)        # 3 rows × 4 cols
5
c = a.reshape(2, 3, 2)     # 3-D: 2×3×2
6

7
# Use -1 to let NumPy infer one dimension
8
d = a.reshape(4, -1)       # → shape (4, 3)

Note: reshape returns a view — modifying b will also change a. Use .copy() if you need independence.

2. `resize()` — Reshape In-Place (原地改变形状)#

Core idea: Like reshape, but modifies the array in-place and can change total element count by repeating or truncating data.

1
a = np.array([1, 2, 3, 4])
2
a.resize(2, 3)   # repeats values to fill: [[1,2,3],[4,1,2]]
3
print(a)

Note: resize() modifies the original array permanently — use with caution.

3. `flatten()` — Collapse to 1-D (展平为一维)#

Core idea: Always returns a copy as a flat 1-D array.

1
b = np.array([[1, 2], [3, 4]])
2
print(b.flatten())    # [1 2 3 4]
3
print(b.ravel())      # [1 2 3 4] — same result, but returns a VIEW

Method	Returns	Modifies original?
`flatten()`	Copy	No
`ravel()`	View (usually)	Yes (if view)

4. `transpose()` — Swap Axes (转置)#

Core idea: Swap rows and columns (or any axes in higher dimensions). Shortcut: .T

1
a = np.array([[1, 2, 3],
2
              [4, 5, 6]])   # shape (2, 3)
3

4
print(a.T)                  # shape (3, 2)
5
print(a.transpose())        # same as a.T
6

7
# For 3-D: specify axis order
8
c = np.ones((2, 3, 4))
9
c.transpose(2, 0, 1)        # new shape: (4, 2, 3)

5. `concatenate()` — Join Arrays (拼接数组)#

Core idea: Join a sequence of arrays along an existing axis (轴).

1
a = np.array([[1, 2], [3, 4]])
2
b = np.array([[5, 6]])
3

4
# axis=0: stack rows (垂直拼接)
5
np.concatenate([a, b], axis=0)   # shape (3, 2)
6

7
# axis=1: stack columns (水平拼接)
8
c = np.array([[7], [8]])
9
np.concatenate([a, c], axis=1)   # shape (2, 3)

Note: Convenience wrappers: np.vstack() (vertical / axis=0) and np.hstack() (horizontal / axis=1).

6. `split()` — Divide an Array (分割数组)#

Core idea: Split an array into multiple sub-arrays along an axis.

1
a = np.arange(12).reshape(4, 3)
2

3
# Split into 2 equal halves along rows (axis=0)
4
parts = np.split(a, 2, axis=0)   # [shape(2,3), shape(2,3)]
5

6
# Split at specific indices
7
np.split(a, [1, 3], axis=0)      # rows 0, rows 1–2, rows 3

Note: np.vsplit(a, n) and np.hsplit(a, n) are shorthand for axis=0 and axis=1 splits.

7. Quick Comparison Table#

Function (函数)	In-place?	Returns
`reshape(shape)`	No	View (same data)
`resize(shape)`	Yes	None (modifies array)
`flatten()`	No	Copy, 1-D
`ravel()`	No	View, 1-D
`transpose()` / `.T`	No	View
`concatenate()`	No	New array
`split()`	No	List of views

💡 One-line Takeaway
Most shape-change operations return views (not copies) — changes propagate back to the original array, so use .copy() when you need an independent result.

II. NumPy Array Operations (数组操作)#

1. reshape() — Change Shape Without Copying (改变形状)#

2. resize() — Reshape In-Place (原地改变形状)#

3. flatten() — Collapse to 1-D (展平为一维)#

4. transpose() — Swap Axes (转置)#

5. concatenate() — Join Arrays (拼接数组)#

6. split() — Divide an Array (分割数组)#