https://github.com/shawwn/sparkvis

Visualize tensors in a plain Python REPL using Sparklines
https://github.com/shawwn/sparkvis

Last synced: 4 months ago
JSON representation

Visualize tensors in a plain Python REPL using Sparklines

• Host: GitHub
• URL: https://github.com/shawwn/sparkvis
• Owner: shawwn
• License: other
• Created: 2021-05-13T05:13:34.000Z (about 3 years ago)
• Default Branch: master
• Last Pushed: 2021-05-13T07:17:41.000Z (about 3 years ago)
• Last Synced: 2024-01-16T10:46:11.092Z (5 months ago)
• Language: Python
• Homepage:
• Size: 15.6 KB
• Stars: 44
• Watchers: 3
• Forks: 3
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Lists

• awesome-stars - shawwn/sparkvis - Visualize tensors in a plain Python REPL using Sparklines (Python)

README

        
# sparkvis

See [here](https://twitter.com/theshawwn/status/1392730448682524672) for usage instructions. Maybe someday I'll write a real readme, but not at 2am on a Thursday.

Ok fine, I'll write a readme. This is a library for visualizing tensors in a plain Python REPL using sparklines. I was sick of having to install jupyter on servers just to see a damn tensor.

E.g. the FFT of MNIST looks like this:

![](https://pbs.twimg.com/media/E1P4TC3WEAApxDU?format=jpg&name=large)

## Quickstart

```

pip3 install -U sparkvis

python3

from sparkvis import sparkvis as vis

vis(foo)

```

`foo` can be a torch tensor, tf tensor, numpy array, etc. It supports anything with a .numpy() method.

`vis(a, b)` will put 'a' and 'b' side by side. For example,

```py

import numpy as np

from sparkvis import sparkvis as vis

x = np.random.rand(7,7)

vis(x, np.zeros_like(x), np.ones_like(x))

```

will print this:

```

▅▅▅▄▄▄▂▂▂▅▅▅▄▄▄▅▅▅▅▅▅▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█████████████████████

▄▄▄▃▃▃▃▃▃▆▆▆▁▁▁▃▃▃███▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█████████████████████

▆▆▆▇▇▇▆▆▆▂▂▂▇▇▇▅▅▅▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█████████████████████

███▇▇▇▃▃▃▇▇▇▄▄▄▂▂▂▂▂▂▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█████████████████████

▆▆▆▅▅▅▇▇▇▅▅▅███▆▆▆▄▄▄▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█████████████████████

▂▂▂▇▇▇▇▇▇▆▆▆▆▆▆▁▁▁▃▃▃▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█████████████████████

▅▅▅▇▇▇▆▆▆▅▅▅▅▅▅▁▁▁▇▇▇▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁█████████████████████

7x21 min=0.0 max=1.0

```

You can pass `to_string=True` if you want the string instead of

printing to stdout. Or you can pass `file=f` like the normal python

`print` function.

### Note on Tensorflow in Graph mode

Currently this library only supports Tensorflow in eager mode, since those are the only tensors that have a .numpy() method. Graph-based tensorflow tensors use .eval() rather than .numpy(). (Sorry, I'll get around to it sometime, otherwise PRs welcome.)

## License

MIT.