Keras tuner integration with WandB

Hi, I’m a new WandB user and I’m trying to integrate WandB with my keras tuner in order to keep track of my hyper parameter tunning.

I tried to follow an online guide I found but I’m getting errors when I try to load the models with the tuner.get_best_models() function

NotFoundError: Unsuccessful TensorSliceReader constructor: Failed to find any matching files for DAM_Aggr_tuning_129_cc/DAM_Aggr_tuning_129_cc_project/trial_2/checkpoint

Here I leave a link to my colab notebook, anyone that can help it would be really appreciated, I feel stuck…

Hi,
Someone on the team is having a look at this and will be in touch with some help ASAP.

1 Like

Hello,
Thank you very much for looking into my issue,

if there is anything I could provide you with please let me know.

It it would be amazing if there was a solution, thank you so much!

Hi Loannis,

Thank you for writing in, and apologies for the delay, we just happened to miss this request.I will review this material immediately and follow up by end of day tomorrow.

Regards,

Mohammad

Thank you very much! Let me also provide you with a link to my github folder in case that can help you more.
There is the full code suposed to be working along with the data i’m trying to use.

Hi @johnioanno ,

I verified that this error is not attributed to wandb. You are creating a python subclass, MyTuner(kt.Tuner) to customize the Keras tuner base class BaseTuner, see here for source, and you intend to save the model within this subclass. You must define the save_model method, see here within you subclass. Otherwise you do get the NotImplementedError reminding you to implement it.

This impacts your ability to load a model(s), resulting in the NotFoundError exception.

Hope this helps you to move forward with your implementation. Please do reach out again if you run into errors while logging with WandbCallback().

Regards,

Mohammad

Hi @johnioanno ,

As 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!

Regards,

Mohammad

ok, Thank you very much for the help!