workbench 0.8.205__py3-none-any.whl → 0.8.213__py3-none-any.whl

workbench/model_scripts/uq_models/generated_model_script.py

@@ -1,198 +1,45 @@
-# Model: XGBoost for point predictions + LightGBM with MAPIE for conformalized intervals
-from mapie.regression import ConformalizedQuantileRegressor
-from lightgbm import LGBMRegressor
+# Model: XGBoost for point predictions + MAPIE UQ Harness for conformalized intervals
 from xgboost import XGBRegressor
 from sklearn.model_selection import train_test_split
 
-# Model Performance Scores
-from sklearn.metrics import mean_absolute_error, median_absolute_error, r2_score, root_mean_squared_error
-from scipy.stats import spearmanr
-
-from io import StringIO
 import json
 import argparse
 import joblib
 import os
 import numpy as np
 import pandas as pd
-from typing import List, Tuple, Optional, Dict
+
+# Shared model script utilities
+from model_script_utils import (
+    check_dataframe,
+    match_features_case_insensitive,
+    convert_categorical_types,
+    decompress_features,
+    input_fn,
+    output_fn,
+    compute_regression_metrics,
+    print_regression_metrics,
+)
+
+# UQ Harness for uncertainty quantification
+from uq_harness import (
+    train_uq_models,
+    save_uq_models,
+    load_uq_models,
+    predict_intervals,
+    compute_confidence,
+)
 
 # Template Placeholders
 TEMPLATE_PARAMS = {
-    "target": "udm_asy_res_efflux_ratio",
-    "features": ['smr_vsa4', 'tpsa', 'nhohcount', 'peoe_vsa1', 'mollogp', 'vsa_estate3', 'xc_4dv', 'smr_vsa3', 'tertiary_amine_count', 'peoe_vsa8', 'minpartialcharge', 'nitrogen_span', 'vsa_estate2', 'chi1v', 'hba_hbd_ratio', 'molecular_axis_length', 'molmr', 'vsa_estate4', 'num_s_centers', 'vsa_estate6', 'qed', 'numhdonors', 'mi', 'estate_vsa4', 'axp_7d', 'kappa3', 'asphericity', 'estate_vsa8', 'estate_vsa2', 'estate_vsa3', 'peoe_vsa3', 'xp_6dv', 'bcut2d_logphi', 'vsa_estate8', 'amphiphilic_moment', 'type_ii_pattern_count', 'minestateindex', 'charge_centroid_distance', 'molecular_asymmetry', 'molecular_volume_3d', 'bcut2d_mrlow', 'axp_1d', 'vsa_estate9', 'aromatic_interaction_score', 'xp_7dv', 'bcut2d_mwlow', 'axp_7dv', 'slogp_vsa1', 'maxestateindex', 'fr_al_oh', 'nbase', 'xp_2dv', 'radius_of_gyration', 'sps', 'xch_7d', 'bcut2d_mrhi', 'axp_0dv', 'vsa_estate5', 'hallkieralpha', 'xp_0dv', 'fr_nhpyrrole', 'smr_vsa1', 'smr_vsa6', 'chi2v', 'bcut2d_mwhi', 'estate_vsa6', 'bcut2d_logplow', 'peoe_vsa2', 'fractioncsp3', 'slogp_vsa2', 'c3sp3', 'peoe_vsa7', 'estate_vsa9', 'peoe_vsa9', 'avgipc', 'smr_vsa9', 'xpc_4dv', 'balabanj', 'axp_1dv', 'mv', 'minabsestateindex', 'bcut2d_chglo', 'fpdensitymorgan2', 'axp_4d', 'numsaturatedheterocycles', 'fpdensitymorgan1', 'axp_3dv', 'axp_5d', 'smr_vsa5', 'bcut2d_chghi', 'axp_3d', 'xpc_5dv', 'chi4n', 'peoe_vsa10', 'vsa_estate7', 'peoe_vsa11', 'estate_vsa10', 'xp_7d', 'slogp_vsa5', 'xch_7dv', 'vsa_estate10', 'labuteasa', 'estate_vsa5', 'xp_3d', 'chi1', 'xch_4dv', 'xp_6d', 'estate_vsa1', 'axp_4dv', 'phi', 'xp_3dv', 'xch_6dv', 'smr_vsa10', 'num_r_centers', 'xc_5d', 'maxpartialcharge', 'xc_3d', 'peoe_vsa6', 'fr_imidazole', 'axp_2d', 'slogp_vsa3', 'mz', 'axp_6dv', 'xch_6d', 'mm', 'numatomstereocenters', 'c1sp3', 'chi1n', 'fpdensitymorgan3', 'xp_5dv', 'chi3v', 'slogp_vsa4', 'fr_ether', 'xp_2d', 'chi3n', 'xch_5dv', 'axp_6d', 'xc_5dv', 'numheterocycles', 'mpe', 'fr_hoccn', 'xc_3dv', 'type_i_pattern_count', 'chi0v', 'xch_4d', 'numsaturatedcarbocycles', 'mp', 'xch_5d', 'maxabspartialcharge', 'axp_2dv', 'bertzct', 'sse', 'xpc_6dv', 'sv', 'xpc_4d', 'si', 'chi0n', 'mse', 'xpc_6d', 'peoe_vsa12', 'xpc_5d', 'kappa2', 'axp_5dv', 'kappa1', 'chi2n', 'intramolecular_hbond_potential', 'fr_nh0', 'numaliphaticheterocycles', 'smr_vsa7', 'mare', 'fr_priamide', 'vsa_estate1', 'num_stereocenters', 'fr_nh1', 'estate_vsa7', 'fr_piperzine', 'c1sp2', 'slogp_vsa6', 'xp_5d', 'fr_aryl_methyl', 'molwt', 'chi4v', 'xc_6dv', 'heavyatommolwt', 'xp_4d', 'sp', 'slogp_vsa7', 'numhacceptors', 'c2sp3', 'peoe_vsa4', 'slogp_vsa10', 'fr_morpholine', 'fr_methoxy', 'fr_aniline', 'xp_4dv', 'fr_urea', 'c3sp2', 'fr_pyridine', 'hybratio', 'fr_thiazole', 'minabspartialcharge', 'sm', 'axp_0d', 'numaromaticheterocycles', 'nocount', 'xc_4d', 'peoe_vsa13', 'fr_amide', 'num_defined_stereocenters', 'amide_count', 'xc_6d', 'numrotatablebonds', 'c2sp2', 'fr_piperdine', 'numvalenceelectrons', 'c1sp1', 'fr_nitrile', 'fr_phenol', 'c4sp3', 'spe', 'numheteroatoms', 'estate_vsa11', 'sz', 'chi0', 'smr_vsa2', 'fr_ketone_topliss', 'slogp_vsa11', 'fr_benzene', 'fr_ndealkylation2', 'peoe_vsa5', 'fr_c_o', 'numsaturatedrings', 'exactmolwt', 'sare', 'numaliphaticrings', 'fr_al_oh_notert', 'fr_imine', 'frac_defined_stereo', 'numunspecifiedatomstereocenters', 'fr_ar_n', 'fr_bicyclic', 'fr_c_o_nocoo', 'numspiroatoms', 'fr_sulfone', 'fr_ndealkylation1'],
+    "target": "solubility",
+    "features": ['molwt', 'mollogp', 'molmr', 'heavyatomcount', 'numhacceptors', 'numhdonors', 'numheteroatoms', 'numrotatablebonds', 'numvalenceelectrons', 'numaromaticrings', 'numsaturatedrings', 'numaliphaticrings', 'ringcount', 'tpsa', 'labuteasa', 'balabanj', 'bertzct'],
     "compressed_features": [],
-    "train_all_data": True,
-    "hyperparameters": {'n_estimators': 500, 'max_depth': 6, 'learning_rate': 0.04},
+    "train_all_data": False,
+    "hyperparameters": {'training_config': {'max_epochs': 150}, 'model_config': {'layers': '128-64-32'}},
 }
 
 
-def compute_confidence(
-        df: pd.DataFrame,
-        median_interval_width: float,
-        lower_q: str = "q_10",
-        upper_q: str = "q_90",
-        alpha: float = 1.0,
-        beta: float = 1.0,
-) -> pd.DataFrame:
-    """
-    Compute confidence scores (0.0 to 1.0) based on prediction interval width
-    and distance from median using exponential decay.
-
-    Args:
-        df: DataFrame with 'prediction', 'q_50', and quantile columns
-        median_interval_width: Pre-computed median interval width from training data
-        lower_q: Lower quantile column name (default: 'q_10')
-        upper_q: Upper quantile column name (default: 'q_90')
-        alpha: Weight for interval width term (default: 1.0)
-        beta: Weight for distance from median term (default: 1.0)
-
-    Returns:
-        DataFrame with added 'confidence' column
-    """
-    # Interval width
-    interval_width = (df[upper_q] - df[lower_q]).abs()
-
-    # Distance from median, normalized by interval width
-    distance_from_median = (df['prediction'] - df['q_50']).abs()
-    normalized_distance = distance_from_median / (interval_width + 1e-6)
-
-    # Cap the distance penalty at 1.0
-    normalized_distance = np.minimum(normalized_distance, 1.0)
-
-    # Confidence using exponential decay
-    interval_term = interval_width / median_interval_width
-    df['confidence'] = np.exp(-(alpha * interval_term + beta * normalized_distance))
-
-    return df
-
-
-# Function to check if dataframe is empty
-def check_dataframe(df: pd.DataFrame, df_name: str) -> None:
-    """
-    Check if the provided dataframe is empty and raise an exception if it is.
-
-    Args:
-        df (pd.DataFrame): DataFrame to check
-        df_name (str): Name of the DataFrame
-    """
-    if df.empty:
-        msg = f"*** The training data {df_name} has 0 rows! ***STOPPING***"
-        print(msg)
-        raise ValueError(msg)
-
-
-def match_features_case_insensitive(df: pd.DataFrame, model_features: list) -> pd.DataFrame:
-    """
-    Matches and renames DataFrame columns to match model feature names (case-insensitive).
-    Prioritizes exact matches, then case-insensitive matches.
-
-    Raises ValueError if any model features cannot be matched.
-    """
-    df_columns_lower = {col.lower(): col for col in df.columns}
-    rename_dict = {}
-    missing = []
-    for feature in model_features:
-        if feature in df.columns:
-            continue  # Exact match
-        elif feature.lower() in df_columns_lower:
-            rename_dict[df_columns_lower[feature.lower()]] = feature
-        else:
-            missing.append(feature)
-
-    if missing:
-        raise ValueError(f"Features not found: {missing}")
-
-    # Rename the DataFrame columns to match the model features
-    return df.rename(columns=rename_dict)
-
-
-def convert_categorical_types(df: pd.DataFrame, features: list, category_mappings={}) -> tuple:
-    """
-    Converts appropriate columns to categorical type with consistent mappings.
-
-    Args:
-        df (pd.DataFrame): The DataFrame to process.
-        features (list): List of feature names to consider for conversion.
-        category_mappings (dict, optional): Existing category mappings. If empty dict, we're in
-                                            training mode. If populated, we're in inference mode.
-
-    Returns:
-        tuple: (processed DataFrame, category mappings dictionary)
-    """
-    # Training mode
-    if category_mappings == {}:
-        for col in df.select_dtypes(include=["object", "string"]):
-            if col in features and df[col].nunique() < 20:
-                print(f"Training mode: Converting {col} to category")
-                df[col] = df[col].astype("category")
-                category_mappings[col] = df[col].cat.categories.tolist()  # Store category mappings
-
-    # Inference mode
-    else:
-        for col, categories in category_mappings.items():
-            if col in df.columns:
-                print(f"Inference mode: Applying categorical mapping for {col}")
-                df[col] = pd.Categorical(df[col], categories=categories)  # Apply consistent categorical mapping
-
-    return df, category_mappings
-
-
-def decompress_features(
-        df: pd.DataFrame, features: List[str], compressed_features: List[str]
-) -> Tuple[pd.DataFrame, List[str]]:
-    """Prepare features for the model by decompressing bitstring features
-
-    Args:
-        df (pd.DataFrame): The features DataFrame
-        features (List[str]): Full list of feature names
-        compressed_features (List[str]): List of feature names to decompress (bitstrings)
-
-    Returns:
-        pd.DataFrame: DataFrame with the decompressed features
-        List[str]: Updated list of feature names after decompression
-
-    Raises:
-        ValueError: If any missing values are found in the specified features
-    """
-
-    # Check for any missing values in the required features
-    missing_counts = df[features].isna().sum()
-    if missing_counts.any():
-        missing_features = missing_counts[missing_counts > 0]
-        print(
-            f"WARNING: Found missing values in features: {missing_features.to_dict()}. "
-            "WARNING: You might want to remove/replace all NaN values before processing."
-        )
-
-    # Decompress the specified compressed features
-    decompressed_features = features.copy()
-    for feature in compressed_features:
-        if (feature not in df.columns) or (feature not in features):
-            print(f"Feature '{feature}' not in the features list, skipping decompression.")
-            continue
-
-        # Remove the feature from the list of features to avoid duplication
-        decompressed_features.remove(feature)
-
-        # Handle all compressed features as bitstrings
-        bit_matrix = np.array([list(bitstring) for bitstring in df[feature]], dtype=np.uint8)
-        prefix = feature[:3]
-
-        # Create all new columns at once - avoids fragmentation
-        new_col_names = [f"{prefix}_{i}" for i in range(bit_matrix.shape[1])]
-        new_df = pd.DataFrame(bit_matrix, columns=new_col_names, index=df.index)
-
-        # Add to features list
-        decompressed_features.extend(new_col_names)
-
-        # Drop original column and concatenate new ones
-        df = df.drop(columns=[feature])
-        df = pd.concat([df, new_df], axis=1)
-
-    return df, decompressed_features
-
-
 if __name__ == "__main__":
     # Template Parameters
     target = TEMPLATE_PARAMS["target"]
@@ -200,7 +47,7 @@ if __name__ == "__main__":
     orig_features = features.copy()
     compressed_features = TEMPLATE_PARAMS["compressed_features"]
     train_all_data = TEMPLATE_PARAMS["train_all_data"]
-    hyperparameters = TEMPLATE_PARAMS["hyperparameters"]
+    hyperparameters = TEMPLATE_PARAMS["hyperparameters"] or {}
     validation_split = 0.2
 
     # Script arguments for input/output directories
@@ -253,8 +100,8 @@ if __name__ == "__main__":
     print(f"VALIDATION: {df_val.shape}")
 
     # Extract sample weights if present
-    if 'sample_weight' in df_train.columns:
-        sample_weights = df_train['sample_weight']
+    if "sample_weight" in df_train.columns:
+        sample_weights = df_train["sample_weight"]
         print(f"Using sample weights: min={sample_weights.min():.2f}, max={sample_weights.max():.2f}, mean={sample_weights.mean():.2f}")
     else:
         sample_weights = None
@@ -266,7 +113,9 @@ if __name__ == "__main__":
     y_train = df_train[target]
     y_validate = df_val[target]
 
+    # ==========================================
     # Train XGBoost for point predictions
+    # ==========================================
     print("\nTraining XGBoost for point predictions...")
     print(f"  Hyperparameters: {hyperparameters}")
     xgb_model = XGBRegressor(enable_categorical=True, **hyperparameters)
@@ -274,136 +123,27 @@ if __name__ == "__main__":
 
     # Evaluate XGBoost performance
     y_pred_xgb = xgb_model.predict(X_validate)
-    xgb_rmse = root_mean_squared_error(y_validate, y_pred_xgb)
-    xgb_mae = mean_absolute_error(y_validate, y_pred_xgb)
-    xgb_r2 = r2_score(y_validate, y_pred_xgb)
-
-    xgb_medae = median_absolute_error(y_validate, y_pred_xgb)
-    xgb_spearman = spearmanr(y_validate, y_pred_xgb).correlation
+    xgb_metrics = compute_regression_metrics(y_validate, y_pred_xgb)
 
     print(f"\nXGBoost Point Prediction Performance:")
-    print(f"rmse: {xgb_rmse:.3f}")
-    print(f"mae: {xgb_mae:.3f}")
-    print(f"medae: {xgb_medae:.3f}")
-    print(f"r2: {xgb_r2:.3f}")
-    print(f"spearmanr: {xgb_spearman:.3f}")
-
-    # Define confidence levels we want to model
-    confidence_levels = [0.50, 0.68, 0.80, 0.90, 0.95]  # 50%, 68%, 80%, 90%, 95% confidence intervals
-
-    # Store MAPIE models for each confidence level
-    mapie_models = {}
-
-    # Train models for each confidence level
-    for confidence_level in confidence_levels:
-        alpha = 1 - confidence_level
-        lower_q = alpha / 2
-        upper_q = 1 - alpha / 2
-
-        print(f"\nTraining quantile models for {confidence_level * 100:.0f}% confidence interval...")
-        print(f"  Quantiles: {lower_q:.3f}, {upper_q:.3f}, 0.500")
-
-        # Train three models for this confidence level
-        quantile_estimators = []
-        for q in [lower_q, upper_q, 0.5]:
-            print(f"    Training model for quantile {q:.3f}...")
-            est = LGBMRegressor(
-                objective="quantile",
-                alpha=q,
-                n_estimators=1000,
-                max_depth=6,
-                learning_rate=0.01,
-                num_leaves=31,
-                min_child_samples=20,
-                subsample=0.8,
-                colsample_bytree=0.8,
-                random_state=42,
-                verbose=-1,
-                force_col_wise=True,
-            )
-            est.fit(X_train, y_train)
-            quantile_estimators.append(est)
-
-        # Create MAPIE CQR model for this confidence level
-        print(f"  Setting up MAPIE CQR for {confidence_level * 100:.0f}% confidence...")
-        mapie_model = ConformalizedQuantileRegressor(
-            quantile_estimators, confidence_level=confidence_level, prefit=True
-        )
-
-        # Conformalize the model
-        print(f"  Conformalizing with validation data...")
-        mapie_model.conformalize(X_validate, y_validate)
-
-        # Store the model
-        mapie_models[f"mapie_{confidence_level:.2f}"] = mapie_model
-
-        # Validate coverage for this confidence level
-        y_pred, y_pis = mapie_model.predict_interval(X_validate)
-        coverage = np.mean((y_validate >= y_pis[:, 0, 0]) & (y_validate <= y_pis[:, 1, 0]))
-        print(f"  Coverage: Target={confidence_level * 100:.0f}%, Empirical={coverage * 100:.1f}%")
-
-    support = len(df_val)
+    print_regression_metrics(xgb_metrics)
+
+    # ==========================================
+    # Train UQ models using the harness
+    # ==========================================
+    uq_models, uq_metadata = train_uq_models(X_train, y_train, X_validate, y_validate)
+
     print(f"\nOverall Model Performance Summary:")
-    print(f"rmse: {xgb_rmse:.3f}")
-    print(f"mae: {xgb_mae:.3f}")
-    print(f"medae: {xgb_medae:.3f}")
-    print(f"r2: {xgb_r2:.3f}")
-    print(f"spearmanr: {xgb_spearman:.3f}")
-    print(f"support: {support}")
-
-    # Analyze interval widths across confidence levels
-    print(f"\nInterval Width Analysis:")
-    for conf_level in confidence_levels:
-        model = mapie_models[f"mapie_{conf_level:.2f}"]
-        _, y_pis = model.predict_interval(X_validate)
-        widths = y_pis[:, 1, 0] - y_pis[:, 0, 0]
-        print(f"  {conf_level * 100:.0f}% CI: Mean width={np.mean(widths):.3f}, Std={np.std(widths):.3f}")
-
-    # Compute normalization statistics for confidence calculation
-    print(f"\nComputing normalization statistics for confidence scores...")
-
-    # Add predictions directly to validation dataframe
-    df_val["prediction"] = xgb_model.predict(X_validate)
-
-    # Add all quantile predictions
-    for conf_level in confidence_levels:
-        model_name = f"mapie_{conf_level:.2f}"
-        model = mapie_models[model_name]
-        y_pred, y_pis = model.predict_interval(X_validate)
-
-        if conf_level == 0.50:
-            df_val["q_25"] = y_pis[:, 0, 0]
-            df_val["q_75"] = y_pis[:, 1, 0]
-            # y_pred is the median prediction
-            df_val["q_50"] = y_pred
-        elif conf_level == 0.68:
-            df_val["q_16"] = y_pis[:, 0, 0]
-            df_val["q_84"] = y_pis[:, 1, 0]
-        elif conf_level == 0.80:
-            df_val["q_10"] = y_pis[:, 0, 0]
-            df_val["q_90"] = y_pis[:, 1, 0]
-        elif conf_level == 0.90:
-            df_val["q_05"] = y_pis[:, 0, 0]
-            df_val["q_95"] = y_pis[:, 1, 0]
-        elif conf_level == 0.95:
-            df_val["q_025"] = y_pis[:, 0, 0]
-            df_val["q_975"] = y_pis[:, 1, 0]
-
-    # Compute normalization stats using q_10 and q_90 (default range)
-    interval_width = (df_val["q_90"] - df_val["q_10"]).abs()
-    median_interval_width = float(interval_width.median())
-    print(f"  Median interval width (q_10-q_90): {median_interval_width:.6f}")
-
-    # Save median interval width for confidence calculation
-    with open(os.path.join(args.model_dir, "median_interval_width.json"), "w") as fp:
-        json.dump(median_interval_width, fp)
+    print_regression_metrics(xgb_metrics)
 
+    # ==========================================
+    # Save all models
+    # ==========================================
     # Save the trained XGBoost model
     joblib.dump(xgb_model, os.path.join(args.model_dir, "xgb_model.joblib"))
 
-    # Save all MAPIE models
-    for model_name, model in mapie_models.items():
-        joblib.dump(model, os.path.join(args.model_dir, f"{model_name}.joblib"))
+    # Save UQ models using the harness
+    save_uq_models(uq_models, uq_metadata, args.model_dir)
 
     # Save the feature list
     with open(os.path.join(args.model_dir, "feature_columns.json"), "w") as fp:
@@ -416,14 +156,14 @@ if __name__ == "__main__":
 
     # Save model configuration
     model_config = {
-        "model_type": "XGBoost_MAPIE_CQR_LightGBM",
-        "confidence_levels": confidence_levels,
+        "model_type": "XGBoost_MAPIE_UQ",
+        "confidence_levels": uq_metadata["confidence_levels"],
         "n_features": len(features),
         "target": target,
         "validation_metrics": {
-            "xgb_rmse": float(xgb_rmse),
-            "xgb_mae": float(xgb_mae),
-            "xgb_r2": float(xgb_r2),
+            "xgb_rmse": float(xgb_metrics["rmse"]),
+            "xgb_mae": float(xgb_metrics["mae"]),
+            "xgb_r2": float(xgb_metrics["r2"]),
             "n_validation": len(df_val),
         },
     }
@@ -431,16 +171,16 @@ if __name__ == "__main__":
         json.dump(model_config, fp, indent=2)
 
     print(f"\nModel training complete!")
-    print(f"Saved 1 XGBoost model and {len(mapie_models)} MAPIE models to {args.model_dir}")
+    print(f"Saved XGBoost model and {len(uq_models)} UQ models to {args.model_dir}")
 
 
 #
 # Inference Section
 #
 def model_fn(model_dir) -> dict:
-    """Load XGBoost and all MAPIE models from the specified directory."""
+    """Load XGBoost and all UQ models from the specified directory."""
 
-    # Load model configuration to know which models to load
+    # Load model configuration
     with open(os.path.join(model_dir, "model_config.json")) as fp:
         config = json.load(fp)
 
@@ -448,11 +188,8 @@ def model_fn(model_dir) -> dict:
     xgb_path = os.path.join(model_dir, "xgb_model.joblib")
     xgb_model = joblib.load(xgb_path)
 
-    # Load all MAPIE models
-    mapie_models = {}
-    for conf_level in config["confidence_levels"]:
-        model_name = f"mapie_{conf_level:.2f}"
-        mapie_models[model_name] = joblib.load(os.path.join(model_dir, f"{model_name}.joblib"))
+    # Load UQ models using the harness
+    uq_models, uq_metadata = load_uq_models(model_dir)
 
     # Load category mappings if they exist
     category_mappings = {}
@@ -461,68 +198,24 @@ def model_fn(model_dir) -> dict:
         with open(category_path) as fp:
             category_mappings = json.load(fp)
 
-    # Load median interval width for confidence calculation
-    median_interval_width = None
-    median_width_path = os.path.join(model_dir, "median_interval_width.json")
-    if os.path.exists(median_width_path):
-        with open(median_width_path) as fp:
-            median_interval_width = json.load(fp)
-
     return {
         "xgb_model": xgb_model,
-        "mapie_models": mapie_models,
-        "confidence_levels": config["confidence_levels"],
+        "uq_models": uq_models,
+        "uq_metadata": uq_metadata,
         "category_mappings": category_mappings,
-        "median_interval_width": median_interval_width,
     }
 
 
-def input_fn(input_data, content_type):
-    """Parse input data and return a DataFrame."""
-    if not input_data:
-        raise ValueError("Empty input data is not supported!")
-
-    # Decode bytes to string if necessary
-    if isinstance(input_data, bytes):
-        input_data = input_data.decode("utf-8")
-
-    if "text/csv" in content_type:
-        return pd.read_csv(StringIO(input_data))
-    elif "application/json" in content_type:
-        return pd.DataFrame(json.loads(input_data))
-    else:
-        raise ValueError(f"{content_type} not supported!")
-
-
-def output_fn(output_df, accept_type):
-    """Supports both CSV and JSON output formats."""
-    if "text/csv" in accept_type:
-        # Convert categorical columns to string to avoid fillna issues
-        for col in output_df.select_dtypes(include=["category"]).columns:
-            output_df[col] = output_df[col].astype(str)
-        csv_output = output_df.fillna("N/A").to_csv(index=False)
-        return csv_output, "text/csv"
-    elif "application/json" in accept_type:
-        return output_df.to_json(orient="records"), "application/json"
-    else:
-        raise RuntimeError(f"{accept_type} accept type is not supported by this script.")
-
-
 def predict_fn(df, models) -> pd.DataFrame:
-    """Make predictions using XGBoost for point estimates and MAPIE for conformalized intervals
+    """Make predictions using XGBoost for point estimates and UQ harness for intervals.
 
     Args:
         df (pd.DataFrame): The input DataFrame
-        models (dict): Dictionary containing XGBoost and MAPIE models
+        models (dict): Dictionary containing XGBoost and UQ models
 
     Returns:
-        pd.DataFrame: DataFrame with XGBoost predictions and conformalized intervals
+        pd.DataFrame: DataFrame with predictions and conformalized intervals
     """
-
-    # Flag for outlier stretch adjustment for the prediction intervals
-    # if the predicted values are outside the intervals
-    outlier_stretch = False
-
     # Grab our feature columns (from training)
     model_dir = os.environ.get("SM_MODEL_DIR", "/opt/ml/model")
     with open(os.path.join(model_dir, "feature_columns.json")) as fp:
@@ -541,65 +234,15 @@ def predict_fn(df, models) -> pd.DataFrame:
     # Get XGBoost point predictions
     df["prediction"] = models["xgb_model"].predict(X)
 
-    # Get predictions from each MAPIE model for conformalized intervals
-    for conf_level in models["confidence_levels"]:
-        model_name = f"mapie_{conf_level:.2f}"
-        model = models["mapie_models"][model_name]
-
-        # Get conformalized predictions
-        y_pred, y_pis = model.predict_interval(X)
-
-        # Map confidence levels to quantile names
-        if conf_level == 0.50:  # 50% CI
-            df["q_25"] = y_pis[:, 0, 0]
-            df["q_75"] = y_pis[:, 1, 0]
-            # y_pred is the median prediction
-            df["q_50"] = y_pred
-        elif conf_level == 0.68:  # 68% CI
-            df["q_16"] = y_pis[:, 0, 0]
-            df["q_84"] = y_pis[:, 1, 0]
-        elif conf_level == 0.80:  # 80% CI
-            df["q_10"] = y_pis[:, 0, 0]
-            df["q_90"] = y_pis[:, 1, 0]
-        elif conf_level == 0.90:  # 90% CI
-            df["q_05"] = y_pis[:, 0, 0]
-            df["q_95"] = y_pis[:, 1, 0]
-        elif conf_level == 0.95:  # 95% CI
-            df["q_025"] = y_pis[:, 0, 0]
-            df["q_975"] = y_pis[:, 1, 0]
-
-    # Calculate a pseudo-standard deviation from the 68% interval width
-    df["prediction_std"] = (df["q_84"] - df["q_16"]).abs() / 2.0
-
-    # Reorder the quantile columns for easier reading
-    quantile_cols = ["q_025", "q_05", "q_10", "q_16", "q_25", "q_50", "q_75", "q_84", "q_90", "q_95", "q_975"]
-    other_cols = [col for col in df.columns if col not in quantile_cols]
-    df = df[other_cols + quantile_cols]
-
-    # Adjust the outer quantiles to ensure they encompass the prediction
-    if outlier_stretch:
-        # Lower intervals adjustments
-        df["q_025"] = np.minimum(df["q_025"], df["prediction"])
-        df["q_05"] = np.minimum(df["q_05"], df["prediction"])
-        df["q_10"] = np.minimum(df["q_10"], df["prediction"])
-        df["q_16"] = np.minimum(df["q_16"], df["prediction"])
-        df["q_25"] = np.minimum(df["q_25"], df["prediction"])
-
-        # Upper intervals adjustments
-        df["q_75"] = np.maximum(df["q_75"], df["prediction"])
-        df["q_84"] = np.maximum(df["q_84"], df["prediction"])
-        df["q_90"] = np.maximum(df["q_90"], df["prediction"])
-        df["q_95"] = np.maximum(df["q_95"], df["prediction"])
-        df["q_975"] = np.maximum(df["q_975"], df["prediction"])
-
-    # Compute confidence scores using pre-computed normalization stats
+    # Get prediction intervals using UQ harness
+    df = predict_intervals(df, X, models["uq_models"], models["uq_metadata"])
+
+    # Compute confidence scores
     df = compute_confidence(
         df,
+        median_interval_width=models["uq_metadata"]["median_interval_width"],
         lower_q="q_10",
         upper_q="q_90",
-        alpha=1.0,
-        beta=1.0,
-        median_interval_width=models["median_interval_width"],
     )
 
     return df