NumPy bridge (from_numpy, numpy) in PyTorch

Introduction

We use NumPy bridge to easily switch data between PyTorch and NumPy. This helps when you want to use both tools together.

You have data in NumPy and want to use PyTorch models.

You want to convert PyTorch tensors back to NumPy arrays for analysis or plotting.

You want to share data between PyTorch and other libraries that use NumPy.

You want to speed up data preparation by using NumPy functions before training in PyTorch.

Syntax

PyTorch

import torch
import numpy as np

# Convert NumPy array to PyTorch tensor
tensor = torch.from_numpy(np_array)

# Convert PyTorch tensor to NumPy array
numpy_array = tensor.numpy()

from_numpy creates a tensor that shares memory with the NumPy array. Changing one changes the other.

numpy() returns a NumPy array that shares memory with the tensor. Changes affect both.

Examples

This converts a NumPy array to a PyTorch tensor.

PyTorch

import torch
import numpy as np

np_array = np.array([1, 2, 3])
tensor = torch.from_numpy(np_array)
print(tensor)

This converts a PyTorch tensor to a NumPy array.

PyTorch

import torch

tensor = torch.tensor([4, 5, 6])
np_array = tensor.numpy()
print(np_array)

Changing the tensor also changes the original NumPy array because they share memory.

PyTorch

import torch
import numpy as np

np_array = np.array([7, 8, 9])
tensor = torch.from_numpy(np_array)
tensor[0] = 10
print(np_array)

Sample Model

This program shows how to convert between NumPy arrays and PyTorch tensors. It also shows that they share memory, so changes in one affect the other.

PyTorch

import torch
import numpy as np

# Create a NumPy array
np_array = np.array([1.0, 2.0, 3.0])
print(f"Original NumPy array: {np_array}")

# Convert to PyTorch tensor
tensor = torch.from_numpy(np_array)
print(f"Converted to tensor: {tensor}")

# Modify tensor
tensor[1] = 20.0
print(f"Modified tensor: {tensor}")

# Check NumPy array after tensor change
print(f"NumPy array after tensor change: {np_array}")

# Convert tensor back to NumPy
np_array2 = tensor.numpy()
print(f"NumPy array from tensor: {np_array2}")

Important Notes

Both conversions share memory, so be careful when modifying data.

PyTorch tensors created from NumPy arrays keep the same data type by default.

If you want a copy instead of shared memory, use tensor.clone() or np.copy().

Summary

Use torch.from_numpy() to convert NumPy arrays to PyTorch tensors.

Use tensor.numpy() to convert PyTorch tensors back to NumPy arrays.

Both share memory, so changes in one affect the other.