Analyzing hyperparameters without actualy performing a sweep

Hy, I’m in love with wandb, but I have a problem…

I have a simple question…

How can I analyze hyperparameters…As seen in this picture, without actually creating a sweep.

In my own code…


I’m preforming learning and for every model i’m sending config with hyperparams…

project=f’{var.version} | {var.INPUT_DATASET}',

But in dashboard I dont see hyperparameters dashboard… And this makes me really sad !

This is my project view…


And this is my table view…


I can’t see the images above, but if you would like to create a parallel coordinates plot, you can do so using the UI by clicking “add panel” in your workspace and choosing Parallel Coordinates.

If you need to do this programmatically, one very recent feature would be to create a W&B Report using our Api. You can programatically define what plots show up. It is a very new feature so it’ll become better documented and more stable over time.

Here’s how you would create a Parallel Coordinates plot programmatically and save it in a report using Python.

import wandb
import wandb.apis.reports as wb
api = wandb.Api()
project = 'pytorch-sweeps-demo'
wandb.require('report-editing') # this is needed as of version 0.12.21 but will likely not be needed in future.
report = wb.Report(
    title='Sweep Results',
                     columns=[wb.reports.PCColumn('batch_size'), wb.reports.PCColumn('epoch'), wb.reports.PCColumn('loss')])
            ], runsets=[wb.RunSet(project=project)]),

This will then show up in the Reports tab on your project.
As this is a very fresh API, there may be issues or features that are not supported yet. I do apologise if that happens to you, I’ll be happy to follow up and provide help.

I have figured it out with the help of your support team, I would really suggest that you guys build the parallel coordinates plot by default since it’s really hard for newcomers to figure this out.

Now with this my soul is satisfied! Congratulation on such a great product!!!

1 Like