As an alternative to MLP, we want to allow using a logistic regression head (+ grid search to find the best hyperparameters C1 & C2). TBD, but likely it would make sense to keep the dataset train/val splits and not introduce CV (since some datasets don't allow for different train/val splits due to data leakage).
ioangatop
commented
9 months ago
As an alternative to MLP, we want to allow using a logistic regression head (+ grid search to find the best hyperparameters C1 & C2). TBD, but likely it would make sense to keep the dataset train/val splits and not introduce CV (since some datasets don't allow for different train/val splits due to data leakage).