Closed fkiraly closed 1 year ago
draft content sketch
Exposition – what are the key features? 1 jupyter notebook Using sktime as an example?
“objects” constructor get_params/set_params basic tags configs mention repr, pretty-printing
composition – simple show get_params/set_params for composition composition – heterogeneous, pipelines show get_params/set_params again
“estimators” Fitting, is_fitted Get_fitted_params Show atomic, composition simple, composition pipeline
Lookup all_objects aka sktime all_estimators all_tags
Testing Get_test_params Create_test_instance Create_test_instances_and_names Check_estimator
Showcase simple mock package – 5min Quick walkthrough on usage, parallel to part 1
Show codebase, check through the below – 5min Step-by-step instructions – 20min Import of BaseObject as parent Special methods in child package Tags in child package Configs in child package Testing (superficial)
2a search/retrieval – 5min All_objects interface
2b estimators – 5min Fit method Get_fitted_params
2c heterogeneous estimators – 10min Heterogeneous mixins Example composite
2d testing – 5min Importing BaseObject, BaseEstimator tests Extending tests
3 wrapup, summary, invite to contribute – 5min
We will need to prepare a tutorial/workshop for pydata Seattle 2023 - 90min incl Q/A, modus is live presentation (but can be prepared from notebooks or dev IDE)
To collect thoughts.
My thoughts:
skbase
, perhaps an "easy" use case and an "advanced" one.sklearn
/sktime
-like interface thatskbase
follows, before we get into implementation details.sktime
(assuming the refactor is released by then), and maybe another package