sparkvis
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:
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,
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
<sparkvis.Sparkvis object at 0x10230f310>
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.)
