Modification of Learned Params to gain efficiency
To solve the issue presented in #148, we propose to:
- Compute each loss only once using a different weight for each slice.
To do so, we need to :
- Modify the loss function to fit this constraint (in particular
training.py
/_compute_loss_metrics
and classparams.py
/ParamSet
- Modify
param-set.yml
which is then referenced inpreprocessing.yml
andtraining.yml