Problem with wandb.plot.pr_curve

Hello. My code for each training epoch:

wandb.log({
    "PR_curve":  wandb.plot.pr_curve(y_true, [(x, 1 - x) for x in y_predict])
})

But I had plots like that(the left part):


When I used pyplot I got the right part for some epoche:
As I understand curve for class “1” must be equal to one of curves from second picture.

I’ve changed predicted labels and now it’s working:

wandb.plot.pr_curve(y_true, [(1 - x, x) for x in y_predict])

Hi Miro,
Glad you were able to figure this out. Let us know if you need any additional help with this.

Thank you,
Nate

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.