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.

Removing points from run

See original GitHub issue

Weights and Biases version: wandb, version 0.8.22
Python version: Python 3.7.6
Operating System: Linux

Description

How can I manually remove some points from the run?

What I Did

In documentaion said, that I can modify run.history, but, as far as I know, history does not attribute, it’s a method, so I can’t set a new value to it.

Issue Analytics

State:
Created 4 years ago
Comments:6 (2 by maintainers)

Top GitHub Comments

3reactions

joplaetecommented, Aug 1, 2022

Hey, was this functionality added since? Culling a window of logs feels like useful functionality to us as well and we also have a usecase where we would use this post model collapse to resume from a past iteration checkpoint. Thanks

0reactions

alex-petrenkocommented, Aug 25, 2022

I also have a use case for this. In population-based training you sometimes want to kill unsuccessful models in the population and restart them from an older checkpoint with a different set of hyperparameters. Sometimes you need to do it multiple times trying multiple sets of parameters. As a result, WandB logs become a mess with different attempts writing over each other.

Deleting all metrics in a specified [x1, x2] window would’ve been very useful!