Repeatedly logging confusion matrix results in overlapping numbers

wandb --version && python --version && uname

• Weights and Biases version: wandb, version 0.9.3
• Python version: Python 3.7.6
• Operating System: Linux

Description

What I want: log confusion matrix for multiclass predictions on the validation set after every epoch, and have them logged separately so I can flip through them

What happened: The numbers come up overlapped on a single plot:

What I Did

I’m using it through the pytorch-lightning API. The validation_epoch_end method returns

{ 'log': {'val_loss': mean_losses,
             'confusion_matrix': confusion_matrix(y, pred, atom_syms)}}

and the ‘log’ key is simply logged via the wandb run object. confusion_matrix is from wandb.sklearn.

[issue-label-bot[bot]]

Issue-Label Bot is automatically applying the label bug to this issue, with a confidence of 0.68. Please mark this comment with 👍 or 👎 to give our bot feedback!

Links: app homepage, dashboard and code for this bot.

[LucaMalagutti]

Hi Sooheon, logging confusion matrices across steps isn’t support natively yet. The plotly way recommended by Mathïs is the recommended way to log them across steps right now.

Are there updates on this? I’m also looking to log a different confusion matrix for every epoch/step