Refactor experiments to allow scenarios and reports to be parameteriz…

…ed by composable configurable components. Introduce a Configurable class which represents anything that has configurable attributes and can also be an injectable attribute type. For example, a scenario has a dataset attribute which is a configurable dataset object with its trainsize, valsize and testsize arguments. This approach is applied to datasets, models, pipelines and postprocessors. All these objects are refactored to have a load or construct class for returning actual objects that are usable in the scenario (e.g. a model.construct() returns a sklearn model). All scenarios are updated accordingly.
easeml · Apr 30, 2024 · 60f6357 · 60f6357
1 parent 6a2572a
commit 60f6357
Show file tree

Hide file tree

Showing 19 changed files with 1,798 additions and 1,316 deletions.
diff --git a/experiments/datascope/experiments/bench/__init__.py b/experiments/datascope/experiments/bench/__init__.py
@@ -1,4 +1,5 @@
 from .base import (
+    Configurable,
     Scenario,
     Study,
     Report,
@@ -20,6 +21,7 @@
 from .main import main
 
 __all__ = [
+    "Configurable",
     "Scenario",
     "Study",
     "Report",