I'd like to have a "full training" mode in the trainer code that _when num_split==1, is does train a model given the rest of configuration, without k-fold evaluation, and save the checkpoint file, without printin/saving P/R/F scores. This will be helpful to train a model that uses the maximum amount of training data, once the hyperparameters are set after k-fold based experiments.
New Feature Summary
Current trainer code performs k-fold evaluation during training, and thus requires split size as a part of trainer configuration
https://github.com/clamsproject/app-swt-detection/blob/5925f029c6f13446a78e7144ae30f146354186a2/modeling/train.py#L164
I'd like to have a "full training" mode in the trainer code that _when
num_split==1, is does train a model given the rest of configuration, without k-fold evaluation, and save the checkpoint file, without printin/saving P/R/F scores. This will be helpful to train a model that uses the maximum amount of training data, once the hyperparameters are set after k-fold based experiments.Related
No response
Alternatives
No response
Additional context
No response