PR #51 adds all the base configs necessary to reproduce [wang 2022] but I noticed there were at least 1 setting that was non-trivial to specify in yaml and I just ignored (ex: the early stopping criteria). This issue is takes charge of comparing thijs' notebook with the configs and making sure all training params (optimizer, ...) are the same.
Tasks
[x] configs/callbacks/early_stopping_rmse.yaml (they have a custom condition based on a train to val rmse threshold, maybe it's not too important, but I didn't port this, just added a generic early stopping mechansim)
[x] configs/model/wang2022_convnet.yaml (almost certain this is well ported but double check
dgcnz
commented
1 month ago
PR #51 adds all the base configs necessary to reproduce [wang 2022] but I noticed there were at least 1 setting that was non-trivial to specify in yaml and I just ignored (ex: the early stopping criteria). This issue is takes charge of comparing thijs' notebook with the configs and making sure all training params (optimizer, ...) are the same.
Tasks
configs/callbacks/early_stopping_rmse.yaml(they have a custom condition based on a train to val rmse threshold, maybe it's not too important, but I didn't port this, just added a generic early stopping mechansim)configs/model/wang2022_convnet.yaml(almost certain this is well ported but double check