Resolved: Feature/#15

[chrstnkgn]

Motivation 🤔

• Resolved Features: WandB logging part as a Hydra option  #15 (partially)
• While executing multiple experiments, found out that the seed was not fixed for the whole process 😦

Key Changes 🔑

• WandB Issue
  👍 WandB is applied for our default running!

  1. The name of each experiment is saved as Y-M-D_H-M-S (same as the hydra logdir name)

  2. val_loss, batch_id, epochs, val_score, and best_val_score are logged
     

  3. Different configs that we want to identify for each run are logged as a table
     

  👎 Could not implement these settings to Ray

  • When applying hyperparameter search using Ray Tune, the WandB logger is still applied using ray-embedded WandBCallbackLogger but the output log is too much in detail which we are not fond of.

• Seed fixing issue

  👍 Implemented seed fixing code to train.py

To Reviewers 🙏

• Ray update needed (pip install -U ray[default]==2.2.0) if you want to test with ray
• For testing purposes: you can modify 'project' value in config/logging/wandb.yaml file so you can control and keep track of your experimental trials.

resolved: #15
references:

[juppak]

Thanks a lot!
I updated ray and tested it as you said. 😁

[chrstnkgn]

If you're all good with this pr I'll just merge it!

[kdg1993]

Beautiful work!! 👍

While I'm trying to test these new changes, a pip install error occurred.

I used the exact command that @chrstnkgn mentioned in her PR "pip install -U ray[default]==2.2.0"

There has been no other error since then, but I'm leaving it for recording just in case 😄