Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Custom Charts: Parameterize number of samples in pr_curve

See original GitHub issue

Requested feature Currently the number of points to plot when using wandb.plot.pr_curve is hardcoded to 20:

https://github.com/wandb/client/blob/488f6db75b36b2d44a0cb5a770040408487d05aa/wandb/plot/pr_curve.py#L70-L81

This is a bit frustrating because it leads to a very bad curve in some cases:

There is no point between ~ 0.8 and ~ 0.98 here. The problem is solved by just setting samples to a higher number, e. g. 100.

Suggested Solution This would be solved by simply making samples an argument to pr_curve and setting it to 20 by default.

If you are willing to change this I could open a PR with the change - would be only a couple of lines.

Issue Analytics

State:
Created 3 years ago
Comments:10 (2 by maintainers)

Top GitHub Comments

1reaction

issue-label-bot[bot]commented, Nov 26, 2020

Issue-Label Bot is automatically applying the label feature_request to this issue, with a confidence of 0.98. Please mark this comment with 👍 or 👎 to give our bot feedback!

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

0reactions

exalate-issue-sync[bot]commented, Jul 12, 2022

WandB Internal User commented: jerofad commented: Hi seems this issues isn’t fixed yet. I was also wondering why the plot using wandb pr_curve was not informative. What’s the work -around for this?

Top Results From Across the Web

Plot Precision Recall Curves With Weights & Biases - Wandb

This article provides usage and examples for plotting Precision Recall curves with Weights & Biases using wandb.plot.pr_curve().

sklearn.metrics.precision_recall_curve

Compute precision-recall pairs for different probability thresholds. Note: this implementation is restricted to the binary classification task. The precision is ...

scikit learn - Area under Precision-Recall Curve (AUC of PR ...

Short answer is: YES. Average Precision is a single number used to summarise a Precision-Recall curve: enter image description here.

ROCR: Visualizing the Performance of Scoring Classifiers

ROCR is a flexible tool for creating cutoff-parameterized 2D performance curves by freely combining two from over 25 performance measures (new.

Precision-recall curves – what are they and how are they used?

This article demonstrates that adding a lot of patients without disease and with low test results to a study may improve the ROC...