- Add a default value from metadata for table models
- Add GRPC server timeout (in seconds) in var env
GRPC_SERVER_TIMEOUT
. Default to None.
- Upgrade sklearn version to 1.2.0
- Minimal python version required is now
3.10
- add new servicer for lgbm model outputting count per leaf instead of its target. The final value is an average of all counts from each tree
- Sklearn wrapper: unpickle models using cloudpickle library to support models pickled with their module
- Upgrade sklearn version to 1.1.0
- Add black + isort to CI and pre-commit configuration.
- Add support of CSV table, served as models
- Refactor parsing + checks of PredictRequest content
- Allow client to request model with string features
- Add support of scikit-learn models
- Allow setting treelite's Predictor
nthread
attribute by settingTREELITE_NTHREAD
environment variable. Defaults to1
.
- Add support of treelite compiled models
- Minimal python version required is now
3.8
base_path
field ofmodels.config
file should now point to the directory containing the model file and not the file itself