Confused logit callback

[williamFalcon]

    Takes the logit predictions of a model and when the probabilities of two classes are very close, the model
    doesn't have high certainty that it should pick one vs the other class.

    This callback shows how the input would have to change to swing the model from one label prediction
    to the other.

[pep8speaks]

Hello @williamFalcon! Thanks for updating this PR.

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2020-07-31 12:42:27 UTC

[codecov]

Codecov Report

Merging #118 into master will decrease coverage by 0.83%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #118      +/-   ##
==========================================
- Coverage   92.25%   91.42%   -0.84%     
==========================================
  Files          79       82       +3     
  Lines        4031     4057      +26     
==========================================
- Hits         3719     3709      -10     
- Misses        312      348      +36     

Flag	Coverage Δ
#unittests	91.42% <100.00%> (-0.84%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
pl_bolts/callbacks/vision/__init__.py	100.00% <100.00%> (ø)
pl_bolts/callbacks/vision/confused_logit.py	100.00% <100.00%> (ø)
..._bolts/models/self_supervised/moco/moco2_module.py	85.71% <0.00%> (-0.73%)	⬇️
pl_bolts/models/self_supervised/cpc/cpc_module.py	87.50% <0.00%> (-0.53%)	⬇️
.../models/autoencoders/basic_vae/basic_vae_module.py	93.06% <0.00%> (-0.39%)	⬇️
...lts/models/self_supervised/simclr/simclr_module.py	91.26% <0.00%> (-0.27%)	⬇️
pl_bolts/models/gans/basic/basic_gan_module.py	93.02% <0.00%> (-0.16%)	⬇️
pl_bolts/models/vision/image_gpt/igpt_module.py	98.66% <0.00%> (-0.12%)	⬇️
pl_bolts/models/mnist_module.py	100.00% <0.00%> (ø)
pl_bolts/datamodules/__init__.py	100.00% <0.00%> (ø)
... and 7 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 9ef90b4...e2dd574. Read the comment docs.

[Borda]

use rather fig, axarr = plt.subplots(ncol=3, nrow=2)
and alter you just axarr[1, 2].imshow(...)

[williamFalcon]

i don't have a lot of time to mess with this. i don't know if this will work, but i know the current version works.
If you feel strongly about it, mind making the changes and posting a colab that shows this works?

[Borda]

ok I ll check I because the call to plt. takes the last pointer and if you plot two in parallel you will write everything into one