I’m currently training a object detection model and wanted to use sweeps to do some hyperparameter optimization. A few of the hyperparameters are in the form of lists. E.g: data_preprocessors: [“random-flip”, “random-crop”, “random-expand”, etc.]
I would like sweep to take a subset from these values and pass them to my training script. However I could not find how to do this easily without a lot of custom wrapping code.
My current solution would be to have each value as a boolean and add some custom logic to convert that to the list I want, however this is not easily expandable/reusable. Is there something I am missing?
You seem to have a good idea for this. Assuming you set up your sweep with these parameters like this:
A sample config would look as follows:
"random_flip" : True,
"random_crop" : False,
"random_expand" : True
You should be use these boolean values as
wandb.config['random_crop'] in order to get your desired output.
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.
Weights & Biases
Hi @dikvangenuchten, 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!