Changelog
Source:NEWS.md
DeepPatientLevelPrediction 2.0.2
- Ensure output from predict_proba is numeric instead of 1d array
- Refactoring: Move cross-validation to a separate function
- Refactoring: Move paramsToTune to a separate function
- linting: Enforcing HADES style
- Calculate AUC ourselves with torch, get rid of scikit-learn dependancy
- added Andromeda to dev dependencies
DeepPatientLevelPrediction 2.0.1
- Connection parameter fixed to be in line with newest polars
- Fixed a bug where LRFinder used a hardcoded batch size
- Seed is now used in LRFinder so it’s reproducible
- Fixed a bug in NumericalEmbedding
- Fixed a bug for Transformer and numerical features
- Fixed a bug when resuming from a full TrainingCache (thanks Zoey Jiang and Linying Zhang )
- Updated installation documentation after feedback from HADES hackathon
- Fixed a bug where order of numeric features wasn’t conserved between training and test set
- TrainingCache now only saves prediction dataframe for the best performing model
DeepPatientLevelPrediction 2.0.0
- New backend which uses pytorch through reticulate instead of torch in R
- All models ported over to python
- Dataset class now in python
- Estimator class in python
- Learning rate finder in python
- Added input checks and tests for wrong inputs
- Training-cache for single hyperparameter combination added
- Fixed empty test for training-cache
DeepPatientLevelPrediction 1.1.4
- Remove torchopt dependancy since adamw is now in torch
- Update torch dependency to >=0.10.0
- Allow device to be a function that resolves during Estimator initialization
DeepPatientLevelPrediction 1.1.3
- Fix actions after torch updated to v0.10 (#65)
DeepPatientLevelPrediction 1.1.2
- Fix bug introduced by removing modelType from attributes (#59)
DeepPatientLevelPrediction 1.1
- Check for if number of heads is compatible with embedding dimension fixed (#55)
- Now transformer width can be specified as a ratio of the embedding dimensions (dimToken), (#53)
- A custom metric can now be defined for earlyStopping and learning rate schedule (#51)
- Added a setEstimator function to configure the estimator (#51)
- Seed added for model weight initialization to improve reproducibility (#51)
- Added a learning rate finder for automatic calculatio of learning rate (#51)
- Add seed for sampling hyperparameters (#50)
- used vectorised torch operations to speed up data conversion in torch dataset
DeepPatientLevelPrediction 1.0.2
- Fix torch binaries issue when running tests from other github actions
- Fix link on website
- Fix tidyselect to silence warnings.