How can I plot pytorch tensor?

Question

I would like to plot pytorch gpu tensor:

input= torch.randn(100).to(device)
output = torch.where(input>=0, input, -input)

input = input.('cpu').detach().numpy().copy()
output = output.('cpu').detach().numpy().copy()

plt.plot(input,out)

However I try to convert those tensors into cpu, numpy, it does not work. How can I plot the tensors ?

anarchy · Accepted Answer · 2021-08-03 02:43:51Z

1

Does this work?

plt.plot(input.cpu().numpy(),output.cpu().numpy())

Alternatively you can try,

plt.plot(input.to('cpu').numpy(),output.to('cpu').numpy())

edited Aug 3, 2021 at 2:43

answered Aug 3, 2021 at 2:27

anarchy

5,2343 gold badges26 silver badges66 bronze badges

Sign up to request clarification or add additional context in comments.

4 Comments

Jenny I Over a year ago

Thank you. But they say 'can't convert non-cpu tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.'

anarchy Over a year ago

Hi @JennyRowland , I updated my answer, does that work?

Nivesh Gadipudi Over a year ago

plt.plot() accepts numpy arrays. The are sequence of operations to perform. First, assuming the tensor is on device(GPU), you need to copy it to CPU first by using .cpu(). Then the you need to change the data type from tensors to numpy by using .numpy(). so, it should be (a.cpu().numpy()).

xaviersjs Over a year ago

FYI, I use output.numpy(Force=True) for this. pytorch.org/docs/stable/generated/torch.Tensor.numpy.html

Luca · Accepted Answer · 2025-11-10 15:05:23Z

0

To address multiple dimensions I've just written the following function

import matplotlib.pyplot as plt
def plot_tensor(X: Tensor) -> None:
    X_copy = X.clone().to("cpu")
    print(f"The input tensor contains {X_copy.shape[0]} points in {X_copy.shape[1]} dimensions")
    print(X_copy.tolist())
    print(f"Tensor on device {X.device}, shape {X.shape}, type {X.dtype} ")
    
    if X.shape[1] == 1:
        plt.scatter(X_copy[:, 0], np.zeros_like(X_copy[:, 0]))
        plt.title("1D Tensor Values")
        plt.xlabel("X1")
        plt.ylabel("Value")  

    if X.shape[1] == 2:
        plt.scatter(X_copy[:, 0], X_copy[:, 1])
        plt.title("2D Tensor Values")
        plt.xlabel("X1")
        plt.ylabel("X2")

    if X.shape[1] == 3:
        fig = plt.figure()
        ax = fig.add_subplot(projection='3d')
        ax.scatter(X_copy[:, 0], X_copy[:, 1], X_copy[:, 2])
        ax.set_title("3D Tensor Values")
        ax.set_xlabel("X1")
        ax.set_ylabel("X2")
        ax.set_zlabel("X3")

    plt.show()

edited Nov 10 at 15:05

answered Nov 10 at 14:59

Luca

11 bronze badge

2 Comments

Viktor Sbruev Nov 10 at 15:42

I feel like this answer is a bit overcomplicated and doesn’t really solve the original problem directly.

Luca Nov 13 at 8:58

well, you need to move the tensors on the CPU anyways, don't you? Beside that I've just added the dimension handling just to ensure proper plotting.

Collectives™ on Stack Overflow

How can I plot pytorch tensor?

2 Answers 2

4 Comments

2 Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

4 Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Related