PyTorch .as_tensor()

In PyTorch, the .as_tensor() function converts various data types, including NumPy arrays, into PyTorch tensors. When possible, it avoids copying data by sharing memory with the original array. This means that if a NumPy array is modified, the corresponding PyTorch tensor will also reflect those changes. However, if the NumPy array is not in a compatible format, PyTorch may create a copy instead.

Syntax

The syntax for the .as_tensor() function is as follows:

torch.as_tensor(data, dtype=None, device=None)

• data: The input data to be converted into a PyTorch tensor. It can be a NumPy array, PyTorch tensor, CuPy array, or Python list.
• dtype: The data type of the output tensor. If None, the data type is inferred from the input data.
• device: The device on which the output tensor will be stored (e.g., 'cpu' or 'cuda'). If None:
  • If data is a CuPy array, the tensor remains on the same GPU.
  • Otherwise, the tensor is placed on the CPU by default.

Example

In the following example, a NumPy array is converted into a PyTorch tensor using .as_tensor(), allowing both to share memory when possible:

import torch
import numpy as np

# Create a NumPy array
arr = np.array([1, 2, 3])

# Convert the NumPy array into a PyTorch tensor
tensor = torch.as_tensor(arr)

print(tensor)

Copy to clipboard
Copy to clipboard

It produces an output as follows:

tensor([1, 2, 3])

Copy to clipboard
Copy to clipboard

In this example, a NumPy array arr with values [1, 2, 3] is first created and then converted into a PyTorch tensor using the .as_tensor() function. The resulting tensor tensor shares memory with the original NumPy array when possible.

Additional Notes

• The .as_tensor() function helps avoid unnecessary data copies when converting a NumPy array into a PyTorch tensor, which can save memory and improve performance, especially with large datasets.
• The resulting PyTorch tensor shares memory with the original NumPy array when possible, meaning that modifying one affects the other.
• To create a separate copy of the data, .tensor() can be used, which always creates a new tensor with independent memory.