feat: use encoded data consistently for similarity & distances (#187)

- use same numerical encoded data for similarity & distance sections
  - removed logic for `pull_data_for_embeddings`
- drop similarity from Data Report
  - no need to store PCA models to statistics_path anymore
  - drop `max_sample_size_embeddings` arg from `report_from_statistics`
- use `HistGradientBoostingClassifiers` for Discriminator AUC (instead of `LogisticRegression`)
- check distances for subsets of columns
  - only report the (sub)set with lowest DCR share
  - try all columns, subset of correlated columns, and subset of random columns

Author: mplatzer

docs/examples.md

@@ -17,10 +17,11 @@ from mostlyai import qa
 repo = "https://github.com/mostly-ai/mostlyai-qa"
 path = "/raw/refs/heads/main/examples/baseball-players"
 
+cols = ["birthDate", "deathDate", "weight", "height", "bats", "throws"]
 report_path, metrics = qa.report(
-    syn_tgt_data=pd.read_parquet(f"{repo}/{path}/synthetic-target.pqt"),
-    trn_tgt_data=pd.read_parquet(f"{repo}/{path}/training-target.pqt"),
-    hol_tgt_data=pd.read_parquet(f"{repo}/{path}/holdout-target.pqt"),
+    syn_tgt_data=pd.read_parquet(f"{repo}/{path}/synthetic-target.pqt")[cols],
+    trn_tgt_data=pd.read_parquet(f"{repo}/{path}/training-target.pqt")[cols],
+    hol_tgt_data=pd.read_parquet(f"{repo}/{path}/holdout-target.pqt")[cols],
     report_subtitle=" for Baseball Players",
     report_path="baseball-players.html",
 )

examples/baseball-players-with-context.html

@@ -44,15 +44,16 @@ plugins:
             show_root_heading: false
             show_object_full_path: true
             show_bases: false
-            show_docstring: true
             show_source: false
             show_signature: true
             separate_signature: true
-            show_docstring_examples: true
             docstring_section_style: table
             extensions:
             - griffe_fieldz
             docstring_style: google
+            extra:
+              show_docstring: true
+              show_docstring_examples: true
 
 markdown_extensions:
   - pymdownx.highlight: