Scan_history() is empty

[chs20]

Hi, when I run run.history(), I get a sampled version of the history as expected (although the number of samples fluctuates). But when I run run.scan_history(), I get an empty object (i.e. 0 rows).

Any idea why this is happening or how it could be fixed?

[artsiom]

Hi Christian,
Happy Friday!

Could you send me some code I can try reproducing this with?

Cheers!
Artsiom

[chs20]

Hi Artsiom, thanks for your reply!

Sure, here is an API call to my project:

import wandb
api = wandb.Api()
runs = api.runs('chs20/scratch-public')
for run in runs:
    history = run.history()
    scan_history = run.scan_history()
    print(f'run: {run.name}')
    print(f'history columns: {len(history.columns)}')
    print(f'scan_history rows: {len(scan_history.rows)}')

This outputs for me:

run: 2023-02-07_10:48:43
history columns: 13
scan_history rows: 0

I would have expected the number of history columns to match the number of scan_history rows. In particular I would not expect scan_history to have zero rows.

[chs20]

Any ideas what is causing this?

[artsiom]

Sorry for the delay in response! So far I haven’t been able to reproduce this on my side so trying a few things

[chs20]

Does the code snippet yield a different output for you?

[artsiom]

Hi Christian, sorry it took so long to get back to you, had to try a few things out.
Can you see if something like this works for you?

import wandbapi = wandb.Api()runs = api.runs('chs20/scratch-public')for run in runs: history = run.scan_history() losses = [row for row in history] print(losses)

[artsiom]

Hi Christian,

We wanted to follow up with you regarding your support request as we have not heard back from you. Please let us know if we can be of further assistance or if your issue has been resolved.

Best,
Weights & Biases

[artsiom]

Hi Christian, since we have not heard back from you we are going to close this request. If you would like to re-open the conversation, please let us know!