search

sktime / mlaut

Other
24 stars 5 forks source link

Data pre-processing and transformation #18

Closed ViktorKaz closed 6 years ago

ViktorKaz commented 6 years ago

Data pre-processing and transformation should be moved to the estimator module and performed before training. Currently this is done when the datasets are saved in the HDF5 database.

ViktorKaz commented 6 years ago

This is now implemented. The datasets are saved in raw format in the HDF5 database. Data pre-processing is performed by the orchestrator class before training based on metadata attached to the estimator class.

Discussion point: Should the metadata about the pre-processing be attached to the estimator, the dataset or both?