search

OHDSI / DeepPatientLevelPrediction

An R package for performing patient level prediction using deep learning in an observational database in the OMOP Common Data Model.
https://ohdsi.github.io/DeepPatientLevelPrediction
10 stars 4 forks source link

Errors with ResNet and Transformer #89

Closed lhjohn closed 10 months ago

lhjohn commented 11 months ago

Training the final ResNet on the optimal hyperparamter combination seems to work, but there is an error afterwards when evaluating the model.

Training final model using optimal parameters
|00%|██████████| 58/58 [00:01<00:00, 29.46it/s]
-00%|██████████| 58/58 [00:01<00:00, 30.69it/s]
Calculating predictions on all train data.../s]
\00%|██████████| 59/59 [00:00<00:00, 71.76it/s]
Error in rbind(deparse.level, ...): numbers of columns of arguments do not match
Calculating covariate summary @ 2023-10-05 09:19:13

Training the Transformer throws an exception right on the first hyperparamter combination.

Running hyperparameter search for Transformer model
2023-10-05 05:29:55 [Main thread]   INFO    DeepPatientLevelPrediction      Running hyperparameter combination no 1
2023-10-05 05:29:55 [Main thread]   INFO    DeepPatientLevelPrediction      HyperParameters: 
2023-10-05 05:29:55 [Main thread]   INFO    DeepPatientLevelPrediction      numBlocks 2 | dimToken 256 | dimOut 1 | numHeads 4 | dimHidden 341 | attDropout 0.25 | ffnDropout 0.2 | resDropout 0.15 | estimator.weightDecay 0.001
2023-10-05 05:29:55 [Main thread]   INFO    DeepPatientLevelPrediction      Max fold: 3
2023-10-05 05:29:56 [Main thread]   ERROR   PatientLevelPrediction  3   The size of tensor a (2) must match the size of tensor b (256) at non-singleton dimension 1
2023-10-05 05:29:56 [Main thread]   INFO    PatientLevelPrediction      Calculating covariate summary @ 2023-10-05 05:29:56