Update inference.py

inference.py

@@ -9,11 +9,7 @@ import warnings
 
 warnings.filterwarnings("ignore", category=UserWarning, module="sklearn")
 
-# ---------------------------------------------------------------------------
-# Compatibility patch — inject _RemainderColsList if the installed sklearn
-# version does not have it (added in sklearn 1.4+). This allows .skops files
-# saved with a newer sklearn to load correctly on older environments.
-# ---------------------------------------------------------------------------
+
 import sklearn.compose._column_transformer as _ct
 if not hasattr(_ct, "_RemainderColsList"):
     class _RemainderColsList(list):
@@ -26,9 +22,6 @@ if not hasattr(_ct, "_RemainderColsList"):
     sklearn.compose._RemainderColsList = _RemainderColsList
 
 
-# ---------------------------------------------------------------------------
-# Column / feature definitions
-# ---------------------------------------------------------------------------
 
 NUM_COLUMNS = ["AGE", "NACS2YR"]
 CATEG_COLUMNS = [
@@ -62,7 +55,7 @@ REPORTING_OUTCOMES = [
 OUTCOME_DESCRIPTIONS = {
     "OS":       "Overall Survival",
     "EFS":      "Event-Free Survival",
-    "DEAD":     "Total Mortality",
+    "DEAD":     "Death",
     "GF":       "Graft Failure",
     "AGVHD":    "Acute Graft-versus-Host Disease",
     "CGVHD":    "Chronic Graft-versus-Host Disease",
@@ -77,9 +70,7 @@ CONSENSUS_THRESHOLD = 0.5
 DEFAULT_N_BOOT_CI   = 500
 
 
-# ---------------------------------------------------------------------------
-# Model loading — skops
-# ---------------------------------------------------------------------------
+
 
 def _load_skops_model(fname):
     try:
@@ -108,7 +99,7 @@ consensus_thresholds  = {
     for o, d in classification_model_data.items()
 }
 
-# Calibrators — isotonic only; supports both old and new key names
+
 calibrators = {}
 for _o, _d in classification_model_data.items():
     _cal      = None
@@ -139,9 +130,7 @@ ohe_feature_names       = ohe.get_feature_names_out(CATEG_COLUMNS)
 processed_feature_names = np.concatenate([NUM_COLUMNS, ohe_feature_names])
 
 
-# ---------------------------------------------------------------------------
-# SHAP background data
-# ---------------------------------------------------------------------------
+
 
 np.random.seed(23)
 _n_background = 500
@@ -182,9 +171,6 @@ _X_background  = preprocessor.transform(_background_df)
 shap_background = shap.maskers.Independent(_X_background)
 
 
-# ---------------------------------------------------------------------------
-# Calibration helpers
-# ---------------------------------------------------------------------------
 
 def calibrate_probabilities_undersampling(p_s, beta):
     p_s         = np.asarray(p_s, dtype=float)
@@ -213,9 +199,7 @@ def predict_consensus_majority(ensemble_models, X_test, threshold=0.5):
     return avg_proba, individual_probas.flatten()
 
 
-# ---------------------------------------------------------------------------
-# Bootstrap CI
-# ---------------------------------------------------------------------------
+
 
 def bootstrap_ci_from_oof(
     point_estimate: float,
@@ -246,9 +230,7 @@ def bootstrap_ci_from_oof(
     return lo, hi
 
 
-# ---------------------------------------------------------------------------
-# Calibration dispatch
-# ---------------------------------------------------------------------------
+
 
 def _calibrate_point(outcome: str, raw_prob: float, use_calibration: bool) -> float:
     beta   = betas[outcome]
@@ -264,9 +246,7 @@ def _calibrate_point(outcome: str, raw_prob: float, use_calibration: bool) -> fl
     return float(cal.transform([p_beta])[0])
 
 
-# ---------------------------------------------------------------------------
-# Main prediction functions
-# ---------------------------------------------------------------------------
+
 
 def predict_all_outcomes(
     user_inputs,
@@ -311,7 +291,6 @@ def predict_all_outcomes(
         probs[o]     = event_prob
         intervals[o] = (lo, hi)
 
-    # OS = 1 - P(DEAD)
     if "DEAD" in probs:
         p_dead      = probs["DEAD"]
         probs["OS"] = float(1.0 - p_dead)
@@ -322,7 +301,6 @@ def predict_all_outcomes(
             float(np.clip(1.0 - dead_lo, 0, 1)),
         )
 
-    # EFS = 1 - P(DWOGF) - P(GF)
     if "DWOGF" in probs and "GF" in probs:
         p_dwogf      = probs["DWOGF"]
         p_gf         = probs["GF"]
@@ -368,9 +346,7 @@ def predict_with_comparison(user_inputs, n_boot_ci: int = DEFAULT_N_BOOT_CI):
     return (cal_probs, cal_intervals), (uncal_probs, uncal_intervals)
 
 
-# ---------------------------------------------------------------------------
-# SHAP helpers
-# ---------------------------------------------------------------------------
+
 
 def _get_shap_values_for_model_outcome(user_inputs, model_outcome, invert, X_proc):
     """Return per-model SHAP values (shape: n_models × n_processed_features)."""
@@ -515,15 +491,13 @@ def create_all_shap_plots(user_inputs, max_display=10):
     return {o: create_shap_plot(user_inputs, o, max_display) for o in SHAP_OUTCOMES}
 
 
-# ---------------------------------------------------------------------------
-# Icon Array (replaces Pie Charts)
-# ---------------------------------------------------------------------------
 
-EVENT_COLOR    = "#e53935"  # red  — event
-NO_EVENT_COLOR = "#43a047"  # green — no event
+
+EVENT_COLOR    = "#e53935"  
+NO_EVENT_COLOR = "#43a047" 
 
 OUTCOME_TITLES = {
-    "DEAD":     "Total Mortality",
+    "DEAD":     "TDeath",
     "GF":       "Graft Failure",
     "AGVHD":    "Acute GvHD",
     "CGVHD":    "Chronic GvHD",
@@ -531,7 +505,7 @@ OUTCOME_TITLES = {
     "STROKEHI": "Stroke Post-HCT",
 }
 
-# Short, equal-length label pairs so the legend stays uniformly sized
+
 OUTCOME_LABELS = {
     "DEAD":     ("Death",         "No Death"),
     "GF":       ("Graft Failure", "No Graft Failure"),
@@ -570,7 +544,6 @@ def create_icon_array_html(probability: float, outcome: str) -> str:
     n_no_event = 100 - n_event
     pct_str    = f"{probability * 100:.1f}%"
 
-    # --- grid: 10 rows × 10 cols ---
     rows_parts = []
     for row in range(10):
         cells = ""
@@ -656,9 +629,6 @@ def create_all_icon_arrays(calibrated_probs: dict) -> dict:
     return cards
 
 
-# ---------------------------------------------------------------------------
-# Backward-compatibility aliases
-# ---------------------------------------------------------------------------
 
 def create_pie_chart(probability, outcome):
     return create_icon_array_html(probability, outcome)