Ask yourself the following:

• Are you using matplotlib.pyplot to plot pytorch tensors?
• Do you forget to call .cpu().detach().numpy() everytime you want to plot a tensor?

Then torchplot may be something for you. torchplot is a simple drop-in replacement for plotting pytorch tensors. We simply override every matplotlib.pyplot function such that pytorch tensors are automatically converted.

Simply just change your default matplotlib import statement:

Instead of

from matplotlib.pyplot import *

use

from torchplot import *

and instead of

import matplotlib.pyplot as plt

use

import torchplot as plt

Herafter, then you can remove every .cpu().detach().numpy() (or variations heroff) from your code and everything should just work. If you do not want to mix implementations, we recommend importing torchplot as seperaly package:

import torchplot as tp

Simple as

pip install torchplot

# lets make a scatter plot of two pytorch variables that are stored on gpu
import torch
import torchplot as plt
x = torch.randn(100, requires_grad=True, device='cuda')
y = torch.randn(100, requires_grad=True, device='cuda')
plt.plot(x, y, '.') # easy and simple

Tested using torch>=1.6 and matplotlib>=3.3.3 but should perfectly work with both earlier and later versions.

Please observe the Apache 2.0 license that is listed in this repository.

If you want to cite the framework feel free to use this (but only if you loved it 😊):

@article{detlefsen2021torchplot,
  title={TorchPlot},
  author={Detlefsen, Nicki S. and Hauberg, Søren},
  journal={GitHub. Note: https://github.com/MachineLearningLifeScience/torchplot},
  year={2021}
}