How to convert a NumPy ndarray to a PyTorch Tensor and vice versa?

A PyTorch tensor is like numpy.ndarray. The difference between these two is that a tensor utilizes the GPUs to accelerate numeric computation. We convert a numpy.ndarray to a PyTorch tensor using the function torch.from_numpy(). And a tensor is converted to numpy.ndarray using the .numpy() method.

Steps

• Import the required libraries. Here, the required libraries are torch and numpy.

• Create a numpy.ndarray or a PyTorch tensor.

• Convert the numpy.ndarray to a PyTorch tensor using torch.from_numpy() function or convert the PyTorch tensor to numpy.ndarray using the .numpy() method.

• Finally, print the converted tensor or numpy.ndarray.

Converting NumPy Array to PyTorch Tensor

The following Python program converts a numpy.ndarray to a PyTorch tensor ?

# import the libraries
import torch
import numpy as np

# Create a numpy.ndarray "a"
a = np.array([[1,2,3],[2,1,3],[2,3,5],[5,6,4]])
print("a:\n", a)

print("Type of a :\n", type(a))
# Convert the numpy.ndarray to tensor
t = torch.from_numpy(a)
print("t:\n", t)
print("Type after conversion:\n", type(t))

a:
 [[1 2 3]
 [2 1 3]
 [2 3 5]
 [5 6 4]]
Type of a :
 <class 'numpy.ndarray'>
t:
 tensor([[1, 2, 3],
        [2, 1, 3],
        [2, 3, 5],
        [5, 6, 4]], dtype=torch.int32)
Type after conversion:
 <class 'torch.Tensor'>

Converting PyTorch Tensor to NumPy Array

The following Python program converts a PyTorch tensor to a numpy.ndarray ?

# import the libraries
import torch
import numpy as np

# Create a tensor "t"
t = torch.Tensor([[1,2,3],[2,1,3],[2,3,5],[5,6,4]])
print("t:\n", t)
print("Type of t :\n", type(t))

# Convert the tensor to numpy.ndarray
a = t.numpy()
print("a:\n", a)
print("Type after conversion:\n", type(a))

t:
 tensor([[1., 2., 3.],
        [2., 1., 3.],
        [2., 3., 5.],
        [5., 6., 4.]])
Type of t :
 <class 'torch.Tensor'>
a:
 [[1. 2. 3.]
 [2. 1. 3.]
 [2. 3. 5.]
 [5. 6. 4.]]
Type after conversion:
 <class 'numpy.ndarray'>

Key Points

• torch.from_numpy() creates a tensor that shares memory with the original NumPy array.

• The .numpy() method only works on tensors that are on the CPU, not GPU.

• Data types are automatically preserved during conversion when possible.

Conclusion

Use torch.from_numpy() to convert NumPy arrays to PyTorch tensors and .numpy() to convert tensors back to NumPy arrays. Both methods preserve the original data structure and share memory for efficient conversion.