Pytorch-lightning: Wandb Flatten Dict

[anthonytec2]

Wandb logger should flatten the dictionary of parameters before logging. Every other logger has the bellow pattern of code:

 params = self._convert_params(params)
 params = self._flatten_dict(params)

🐛 Bug

Wandb logger does not flatten parameters resulting in dictionaries being logged to Wandb, which are not searchable causing for some loss of features in wandb.

To Reproduce

Run the cpu_template with wandb logger, and log a nested dictionary.

Expected behavior

Solution, just call params = self._flatten_dict(params) this in the wandb logger.

Environment

• CUDA:
  
  
  
  • GPU:
  
  
  • available: False
  
  
  • version: None
  
  
• Packages:
  
  
  
  • numpy: 1.18.5
  
  
  • pyTorch_debug: False
  
  
  • pyTorch_version: 1.5.0
  
  
  • pytorch-lightning: 0.8.4
  
  
  • tensorboard: 2.2.2
  
  
  • tqdm: 4.46.1
  
  
• System:
  
  
  
  • OS: Darwin
  
  
  • architecture:
    
    
    
    
    
    • 64bit
      
      
      -
    
    
    
  
  
  • processor: i386
  
  
  • python: 3.7.7
  
  
  • version: Darwin Kernel Version 19.4.0: Wed Mar 4 22:28:40 PST 2020; root:xnu-6153.101.6~15/RELEASE_X86_64
  
  

[awaelchli]

Is this what you mean?
I agree, it would be nice if we could flatten that out.

[anthonytec2]

Exactly, this commit will make hparams look like:

[awaelchli]

perfect!