NumPy: Shape Operations

Q: What's the difference between .T and .transpose()?

.T is a shortcut for reversing all axes (equivalent to .transpose() with reversed order). For 2D arrays they're identical. For 3D+, .T reverses all axes, while .transpose(1,0,2) lets you specify the order.

Last updated: 2026-08-26

Shape Operations

1. What You'll Learn

❶ reshape doesn't copy data — it only changes metadata
❷ -1 auto-infers dimension size
❸ flatten vs ravel: copy vs view
❹ transpose / .T transposition
❺ newaxis for adding dimensions, squeeze for removing them

2. A Developer's True Story

(1) The Problem

Bob reshapes 12 data points from 3×4 to 4×3, thinking he's "rearranging memory" — but every reshape was copying data, killing his memory and speed.

(2) The Solution

Alice shows him a Mermaid diagram: "reshape only changes metadata — not a single byte of data moves. That's the secret to NumPy's speed." As long as an operation returns a view, no data is copied.

(3) The Payoff

Once Bob understood "view vs copy," his shape operation code used an order of magnitude less memory and ran several times faster. Only flatten truly copies; everything else — reshape, ravel, transpose, swapaxes — is a view.

3. Shape Operations in Detail

(1) How reshape Works

reshape changes the array's shape but does not copy data. It only modifies the array's metadata (shape and strides); the raw data in memory stays put.

PYTHON

import numpy as np

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

print(a.shape)   # (12,)
print(b.shape)   # (3, 4)
print(b.base is a)  # True — b is a view of a

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

⚠️ Note: The code below needs to run in a local Python environment.

PYTHON

# reshape does NOT copy data
b[0, 0] = 999
print(a[0])  # 999 — modifying b changes a!

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

reshape Memory Diagram

graph LR
    A["Original Array<br/>shape=(12,)"] -->|"reshape(3,4)"| B["View<br/>shape=(3,4)"]
    A -->|"reshape(4,3)"| C["View<br/>shape=(4,3)"]
    A -->|"reshape(2,6)"| D["View<br/>shape=(2,6)"]

    style A fill:#4CAF50,color:#fff
    style B fill:#2196F3,color:#fff
    style C fill:#FF9800,color:#fff
    style D fill:#9C27B0,color:#fff

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

Key point: A, B, C, and D all share the same memory — they just "interpret" it differently.

(2) -1 Auto-Inference

Use -1 in reshape to let NumPy automatically compute that dimension's size.

PYTHON

a = np.arange(12)

# NumPy infers: 12 / 3 = 4
b = a.reshape(3, -1)     # shape = (3, 4)
c = a.reshape(2, -1)     # shape = (2, 6)
d = a.reshape(-1, 6)     # shape = (2, 6)

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

⚠️ Note: The code below needs to run in a local Python environment.

(3) flatten vs ravel

Method	Returns	Copies?	Memory
`ndarray.flatten()`	1D array	Always copies	New memory
`ndarray.ravel()`	1D array	View when possible	Usually shared

PYTHON

import numpy as np

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

f = a.flatten()  # always a copy
r = a.ravel()    # usually a view

f[0] = 999
print(a[0, 0])   # 0 — f is a copy, a unchanged

r[0] = 999
print(a[0, 0])   # 999 — r is a view, a changed

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

⚠️ Note: The code below needs to run in a local Python environment.

(4) transpose and .T

PYTHON

import numpy as np

a = np.arange(12).reshape(3, 4)
print(a.shape)   # (3, 4)

b = a.T          # transpose — view, no copy
print(b.shape)   # (4, 3)

# For multi-dimensional arrays, transpose accepts an axis order
c = a.transpose(1, 0)
print(c.shape)   # (4, 3)

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

⚠️ Note: The code below needs to run in a local Python environment.

(5) newaxis and squeeze

PYTHON

import numpy as np

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

# newaxis adds a dimension
b = a[np.newaxis, :]           # shape (1, 3)
c = a[:, np.newaxis]           # shape (3, 1)
d = a[np.newaxis, :, np.newaxis]  # shape (1, 3, 1)

# squeeze removes dimensions of size 1
e = np.squeeze(d)              # shape (3,) — back to original

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

⚠️ Note: The code below needs to run in a local Python environment.

▶ Example

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

: reshape with -1 auto-inference (Difficulty ⭐)

PYTHON

import numpy as np

a = np.arange(24)

# Use -1 for auto-inference
b = a.reshape(2, -1)      # 2 x 12
c = a.reshape(3, -1)      # 3 x 8
d = a.reshape(4, -1)      # 4 x 6
e = a.reshape(6, -1)      # 6 x 4

print(f"a.shape={a.shape}")
print(f"b.shape={b.shape}")  # (2, 12)
print(f"c.shape={c.shape}")  # (3, 8)
print(f"d.shape={d.shape}")  # (4, 6)
print(f"e.shape={e.shape}")  # (6, 4)

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

⚠️ Note: The code below needs to run in a local Python environment.

▶ Example

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

: flatten vs ravel — copy vs view (Difficulty ⭐⭐)

PYTHON

import numpy as np

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

# flatten always copies
flat = a.flatten()
flat[0] = 999
print("After flat[0]=999, a[0,0]=", a[0,0])  # 0 (unchanged)

# ravel usually returns a view
rav = a.ravel()
rav[0] = 999
print("After rav[0]=999, a[0,0]=", a[0,0])  # 999 (changed!)

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

⚠️ Note: The code below needs to run in a local Python environment.

▶ Example

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

: transpose and newaxis (Difficulty ⭐⭐)

PYTHON

import numpy as np

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

# Transpose
print("Original shape:", a.shape)     # (3, 4)
print("Transposed shape:", a.T.shape)  # (4, 3)

# newaxis: add dimensions
row = np.array([1, 2, 3])
print("row shape:", row.shape)                # (3,)
print("row[np.newaxis,:] shape:", row[np.newaxis, :].shape)  # (1, 3)
print("row[:,np.newaxis] shape:", row[:, np.newaxis].shape)  # (3, 1)

TEXT 📖 Display only

> **Output:** Run NumPy 2.x in your local Python environment to see the ndarray output. The Piston server does not have NumPy pre-installed — install it locally (`pip install numpy`) and compare. Actual values may vary by NumPy version and random seed.

⚠️ Note: The code below needs to run in a local Python environment.

4. Comparison Tables

(1) Shape Operations Summary

Operation	Returns	Copies Data?	Use Case
`reshape`	View	No	Change shape
`flatten`	Copy	Yes	Safe 1D conversion
`ravel`	View (usually)	Rarely	Fast 1D
`T` / `transpose`	View	No	Axis reordering
`newaxis`	View	No	Add dimension
`squeeze`	View	No	Remove size-1 dims
`swapaxes`	View	No	Swap two axes

(2) View vs Copy

Feature	View	Copy
Memory	Shared	New allocation
Modification	Affects original	Independent
Speed	O(1)	O(n)
`base` attribute	Points to original	None
When to use	Read-only or intentional sharing	Safe modification

5. Principle Diagram

graph TB
    A["Original (12,)"] -->|reshape/flatten/ravel| B["New shape"]
    A --> C["data buffer: [0 1 2 ... 11]"]
    B --> C
    B -->|flatten| D["copy of data buffer"]
    B -->|ravel| C
    B -->|transpose| C
    B -->|newaxis| C
    
    style A fill:#4CAF50,color:#fff
    style C fill:#FF9800,color:#fff
    style D fill:#E91E63,color:#fff

❓ FAQ

Q Does reshape always return a view?

A Almost always — as long as the array is contiguous in memory. If the array is non-contiguous (e.g., after a transpose), reshape may need to copy. Use np.ascontiguousarray() first if needed.

Q What's the difference between .T and .transpose()?

A .T is a shortcut for reversing all axes (equivalent to .transpose() with reversed order). For 2D arrays they're identical. For 3D+, .T reverses all axes, while .transpose(1,0,2) lets you specify the order.

Q When should I use flatten vs ravel?

A If you're going to modify the 1D result and don't want to affect the original, use flatten (safe copy). If you're just reading or you want changes to propagate, use ravel (faster, usually a view).

Q What does -1 mean in reshape?

A It tells NumPy to "figure out this dimension automatically." Only one -1 is allowed. NumPy calculates it as total_size / product_of_other_dims.

📖 Summary

reshape only changes shape/strides metadata — no data copy, O(1) operation
-1 auto-infers dimension size from the total element count
flatten always copies; ravel usually returns a view (check .base)
transpose/T/swapaxes return views with modified strides, no data movement
newaxis adds a dimension of size 1; squeeze removes size-1 dimensions
Understanding view vs copy is key to writing memory-efficient NumPy code

📝 Exercises

Beginner (Difficulty ⭐): Create an array of 20 elements, reshape it to (4,5), (5,4), (2,10), and (10,2) using -1. Verify all shapes are correct.
Intermediate (Difficulty ⭐⭐): Create a 3x4 array, transpose it, then try to reshape the transposed result. Does reshape still avoid copying? Check with .base.
Advanced (Difficulty ⭐⭐⭐): Create a 3D array with shape (2,3,4). Use transpose(2,0,1) to reorder axes. Explain what each dimension represents in the result. Then use swapaxes to achieve the same result.

Web-Tutorial Tech Team

A team of developers maintaining programming tutorials. Each tutorial is written and reviewed by developers with expertise in that field. We work to keep our content accurate and reliable — if you spot an issue, please let us know.

About Us

NumPy: Shape Operations

Shape Operations

1. What You'll Learn

2. A Developer's True Story

(1) The Problem

(2) The Solution

(3) The Payoff

3. Shape Operations in Detail

(1) How reshape Works

reshape Memory Diagram

(2) -1 Auto-Inference

(3) flatten vs ravel

(4) transpose and .T

(5) newaxis and squeeze

▶ Example

▶ Example

▶ Example

4. Comparison Tables

(1) Shape Operations Summary

(2) View vs Copy

5. Principle Diagram

❓ FAQ

📖 Summary

📝 Exercises

🙏 帮我们做得更好