Chapter 10

NumPy → PyTorch

ndarray vs Tensor

A PyTorch Tensor is an n-dimensional array, just like a NumPy ndarray: the same grid, the same shape, the same axes. Everything from chapters 01–07 carries over.

The table puts each NumPy property next to its PyTorch spelling. The NumPy column is computed live. PyTorch cannot run in the browser, so its column shows what real PyTorch returns for the same data (recorded).

Try this

Remember: A tensor is an n-dimensional array with two extra powers: a device and gradient tracking.

a = dtype shape (2, 3)

Try

Spaces or commas between values · new line = new row · blank line = new 2-D layer (3-D array).

Run the PyTorch side in Colab

NumPy runs here for real; PyTorch does not run in the browser. PyTorch code is shown next to it, and every PyTorch result stated here (dtypes, errors) was recorded with PyTorch 2.14.0.

Starting Python… (first visit downloads ≈ 15 MB, then it is cached)

a NumPy ndarray · runs here

t — the same numbers as a PyTorch Tensor not run here

import torch

t = torch.tensor([[1, 2, 3],
                  [4, 5, 6]])

Same values, same shape, same axes: a tensor is PyTorch’s n-dimensional array.

Python not started

import numpy as np

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

a

Output

Output appears here.