"float division by zero" exception during notebook-example on anaconda/windows about catalyst HOT 4 CLOSED

Hi @TezRomacH! It's totally ok.
Can't wait for documentation, by the way :)

from catalyst.

Hi,

Good catch! Looks like we need some additional Windows testing, cause all code works like a charm on Linux and MacOS. Just to be sure, I have tested your code with:

Python 3.6.4 :: Anaconda custom (64-bit)
torch.__version__==1.0.0
catalyst == 892d5e5

It's a bit weird when batch elapsed_time is == 0, but as a hotfix you can try to:

• turn off GPU
• increase batch size
• use large dataset, like finetune example

Somehow you process your batch so quick, that even python cannot make it out :)

from catalyst.

Just checked example on ubuntu machine (with much better hardware, lol). Works perfectly

from catalyst.

Hi, Davier. Thanks for reporting!

Currently, Catalyst is in alpha and we do a lot of breaking changes right now. Sorry for that!
Within a couple of weeks, we plan to release the stable version of Catalyst with a great number of new features!

Right now, you may add a zero-checker while we fixing it 😄

from catalyst.