edgepoint 3.5.0__tar.gz → 3.5.2__tar.gz

{edgepoint-3.5.0 → edgepoint-3.5.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: edgepoint
-Version: 3.5.0
+Version: 3.5.2
 Summary: Find the point in a numeric column above which a binary outcome becomes meaningfully and reliably better.
 Author: Henry
 License: MIT

{edgepoint-3.5.0 → edgepoint-3.5.2}/edgepoint/main.py

@@ -510,6 +510,9 @@ def search(df, outcome_col="hit", min_coverage=10, show_progress=True, top_combo
     # fold (see test_and_combo_val's fold_combos_df filtering). Those
     # aren't "no data", they're "didn't qualify", so they're excluded
     # here rather than shown as _te NaN rows.
+    # A further baseline check runs after the merge: any combo whose
+    # train or test hit_rate falls below baseline_full is disqualified
+    # here too, see below.
     stage_start = time.time()
     if show_progress:
         _stage_header(5, TOTAL_STAGES, "Merge Combos (train vs test)")
@@ -519,12 +522,39 @@ def search(df, outcome_col="hit", min_coverage=10, show_progress=True, top_combo
         on=["combo_num", "combo"], how="inner", suffixes=("_tr", "_te"),
     )
     merged_combos_df = _add_final_score(merged_combos_df)
+
+    # Snapshot the count right after the inner merge (i.e. how many combos
+    # actually passed test validation), BEFORE the baseline filter below
+    # removes any rows. Needed purely so the Stage 5 print message can show
+    # test-validation survivors, baseline-disqualified, and final remaining
+    # as one coherent chain instead of leaving an unaccounted-for gap
+    # between top_combos_df's total and the post-baseline-filter count.
+    n_test_qualified = len(merged_combos_df)
+    if show_progress:
+        print(f"{n_test_qualified}/{len(top_combos_df)} combo(s) entered (passed test validation).")
+        print()
+
+    # Disqualify any combo whose train OR test hit_rate falls below
+    # baseline_full - a combo that can't beat "doing nothing" on even
+    # one side isn't a real edge, no blended final_score should be able
+    # to paper over that. hit_rate_tr/te arrive as formatted strings
+    # (see recommend()'s combo_summary_df), so cast to float to compare.
+    n_disqualified = 0
+    if baseline_full is not None and not merged_combos_df.empty:
+        hr_tr = merged_combos_df["hit_rate_tr"].astype(float)
+        hr_te = merged_combos_df["hit_rate_te"].astype(float)
+        below_baseline_mask = (hr_tr < baseline_full) | (hr_te < baseline_full)
+        n_disqualified = int(below_baseline_mask.sum())
+        merged_combos_df = merged_combos_df[~below_baseline_mask].reset_index(drop=True)
+
     merged_combos_df = merged_combos_df.sort_values("final_score", ascending=False).reset_index(drop=True)
 
     if show_progress:
         display_combos_df, combos_legend = _abbreviate_table(
             merged_combos_df, extra_drop_bases=("miss_rate", "gap")
         )
+        print(f"{n_disqualified}/{n_test_qualified} sliced (below baseline {baseline_full:.2f}) -> {len(merged_combos_df)} remain.")
+        print()
         print(f"=== MERGED COMBOS (train vs test, test-qualified only) - {len(merged_combos_df)}/{len(top_combos_df)} combo(s) ===")
         print()
         _print_legend(combos_legend)
@@ -545,21 +575,4 @@ def search(df, outcome_col="hit", min_coverage=10, show_progress=True, top_combo
         _stage_footer(stage_start, run_start)
         _complete(run_start)
 
-    return edgepoints_df, top_combos
-
-"""
-if __name__ == "__main__":
-    rows, headers = MySql.Select(
-        "SELECT * FROM bucket WHERE pick = %s",
-        ("home_win",),
-        headers=True
-    )
-    df = pd.DataFrame(rows, columns=headers)
-
-    # Split ratio is auto-detected from row count inside search()'s
-    # Stage 1 (via split_router.split_dataset). Nothing else changes.
-    edgepoints_df, top_combos = search(
-        df, outcome_col="hit", min_coverage=10, show_progress=True,
-        top_combo_n=3,
-    )
-"""
+    return edgepoints_df, top_combos

{edgepoint-3.5.0 → edgepoint-3.5.2}/edgepoint.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: edgepoint
-Version: 3.5.0
+Version: 3.5.2
 Summary: Find the point in a numeric column above which a binary outcome becomes meaningfully and reliably better.
 Author: Henry
 License: MIT

{edgepoint-3.5.0 → edgepoint-3.5.2}/edgepoint.egg-info/SOURCES.txt

@@ -6,9 +6,7 @@ edgepoint/__init__.py
 edgepoint/aggregate_kfold_combos.py
 edgepoint/aggregate_kfold_edges.py
 edgepoint/core.py
-edgepoint/core_v1.py
 edgepoint/main.py
-edgepoint/main_old.py
 edgepoint/run_kfold_find.py
 edgepoint/run_kfold_search.py
 edgepoint/sanity_check.py

{edgepoint-3.5.0 → edgepoint-3.5.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "edgepoint"
-version = "3.5.0"
+version = "3.5.2"
 description = "Find the point in a numeric column above which a binary outcome becomes meaningfully and reliably better."
 readme = "README.md"
 requires-python = ">=3.9"

{edgepoint-3.5.0 → edgepoint-3.5.2}/setup.py

@@ -5,7 +5,7 @@ long_description = (Path(__file__).parent / "README.md").read_text(encoding="utf
 
 setup(
     name="edgepoint",
-    version="3.5.0",
+    version="3.5.2",
     author="Henry",
     description="Find where a metric, or combination of metrics, starts getting better while keeping the best coverage possible, then check if that holds on unseen data.",
     long_description=long_description,