Rename models/pitch_model.txt to models/pitch_model.py

models/pitch_model.py

@@ -0,0 +1,59 @@
+from __future__ import annotations
+
+from typing import Any
+
+import numpy as np
+import pandas as pd
+
+
+def predict_pitch_outcome(
+    release_speed: float,
+    release_spin_rate: float,
+    pfx_x: float,
+    pfx_z: float,
+    batter_ev90: float,
+    batter_xwoba: float,
+) -> dict[str, float]:
+    """
+    Research baseline model.
+    This is deterministic and explainable, and can later be swapped
+    with a trained neural net without changing the app interface.
+    """
+    speed = 0.45 + max(0.0, (release_speed - 90.0)) * 0.012
+    spin = 0.18 + max(0.0, (release_spin_rate - 2100.0)) * 0.00008
+    movement = 0.10 + (abs(pfx_x) + abs(pfx_z)) * 0.01
+    hitter_damage = max(0.0, ((batter_ev90 - 92.0) * 0.01) + ((batter_xwoba - 0.320) * 0.8))
+
+    strike_prob = min(0.85, speed + (movement * 0.25))
+    whiff_prob = min(0.70, spin + (movement * 0.20) - (hitter_damage * 0.10))
+    damage_prob = min(0.75, 0.08 + hitter_damage - (speed * 0.10))
+
+    strike_prob = max(0.05, strike_prob)
+    whiff_prob = max(0.02, whiff_prob)
+    damage_prob = max(0.01, damage_prob)
+
+    ball_in_play_prob = max(0.01, 1.0 - whiff_prob - (strike_prob * 0.10))
+
+    return {
+        "strike_prob": float(min(1.0, strike_prob)),
+        "whiff_prob": float(min(1.0, whiff_prob)),
+        "damage_prob": float(min(1.0, damage_prob)),
+        "ball_in_play_prob": float(min(1.0, ball_in_play_prob)),
+    }
+
+
+def pitcher_baseline_from_events(df: pd.DataFrame) -> dict[str, float]:
+    if df.empty:
+        return {
+            "release_speed": 94.0,
+            "release_spin_rate": 2250.0,
+            "pfx_x": 0.0,
+            "pfx_z": 12.0,
+        }
+
+    return {
+        "release_speed": float(pd.to_numeric(df.get("release_speed"), errors="coerce").mean()),
+        "release_spin_rate": float(pd.to_numeric(df.get("release_spin_rate"), errors="coerce").mean()),
+        "pfx_x": float(pd.to_numeric(df.get("pfx_x"), errors="coerce").mean()),
+        "pfx_z": float(pd.to_numeric(df.get("pfx_z"), errors="coerce").mean()),
+    }