This may already exist, but it would be great to be able to write plugins for different experiment logging methods, e.g. wandb (weights and biases), tensorboard, etc...
Motivation
For many ML experiments, something more organized than log files is useful, especially for metrics such as the loss, which are best graphed.
Pitch
Would love to be able to add something like python train.py log=wandb, (though maybe more options are needed so log.type="wandb" is better)
🚀 Feature
This may already exist, but it would be great to be able to write plugins for different experiment logging methods, e.g. wandb (weights and biases), tensorboard, etc...
Motivation
For many ML experiments, something more organized than log files is useful, especially for metrics such as the loss, which are best graphed.
Pitch
Would love to be able to add something like
python train.py log=wandb
, (though maybe more options are needed solog.type="wandb"
is better)