workbench 0.8.171__py3-none-any.whl → 0.8.172__py3-none-any.whl

workbench/api/feature_set.py

@@ -87,8 +87,8 @@ class FeatureSet(FeatureSetCore):
         model_import_str: str = None,
         custom_script: Union[str, Path] = None,
         custom_args: dict = None,
-        training_image: str = "xgb_training",
-        inference_image: str = "xgb_inference",
+        training_image: str = "training",
+        inference_image: str = "inference",
         inference_arch: str = "x86_64",
         **kwargs,
     ) -> Union[Model, None]:
@@ -105,8 +105,8 @@ class FeatureSet(FeatureSetCore):
             model_class (str, optional): Model class to use (e.g. "KMeans", "PyTorch", default: None)
             model_import_str (str, optional): The import for the model (e.g. "from sklearn.cluster import KMeans")
             custom_script (str, optional): The custom script to use for the model (default: None)
-            training_image (str, optional): The training image to use (default: "xgb_training")
-            inference_image (str, optional): The inference image to use (default: "xgb_inference")
+            training_image (str, optional): The training image to use (default: "training")
+            inference_image (str, optional): The inference image to use (default: "inference")
             inference_arch (str, optional): The architecture to use for inference (default: "x86_64")
             kwargs (dict, optional): Additional keyword arguments to pass to the model
 

workbench/core/artifacts/artifact.py

@@ -236,6 +236,12 @@ class Artifact(ABC):
             This functionality will work for FeatureSets, Models, and Endpoints
             but not for DataSources. The DataSource class overrides this method.
         """
+
+        # Check for ReadOnly Role
+        if self.aws_account_clamp.read_only_role:
+            self.log.info("Cannot add metadata with a ReadOnly Role...")
+            return
+
         # Sanity check
         aws_arn = self.arn()
         if aws_arn is None:
@@ -444,10 +450,12 @@ class Artifact(ABC):
 
 if __name__ == "__main__":
     """Exercise the Artifact Class"""
-    from workbench.api.data_source import DataSource
-    from workbench.api.feature_set import FeatureSet
+    from workbench.api import DataSource, FeatureSet, Endpoint
+
+    # Grab an Endpoint (which is a subclass of Artifact)
+    end = Endpoint("wine-classification")
 
-    # Create a DataSource (which is a subclass of Artifact)
+    # Grab a DataSource (which is a subclass of Artifact)
     data_source = DataSource("test_data")
 
     # Just some random tests

workbench/core/artifacts/model_core.py

@@ -37,16 +37,45 @@ class ModelType(Enum):
     UNKNOWN = "unknown"
 
 
+# Deprecated Images
+"""
+        # US East 1 images
+        "py312-general-ml-training"
+        ("us-east-1", "training", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-sklearn-xgb-training:0.1"
+        ),
+        ("us-east-1", "inference", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1"
+        ),
+
+        # US West 2 images
+        ("us-west-2", "training", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-sklearn-xgb-training:0.1"
+        ),
+        ("us-west-2", "inference", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1"
+        ),
+
+        # ARM64 images
+        ("us-east-1", "inference", "0.1", "arm64"): (
+            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1-arm64"
+        ),
+        ("us-west-2", "inference", "0.1", "arm64"): (
+            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1-arm64"
+        ),
+"""
+
+
 class ModelImages:
     """Class for retrieving workbench inference images"""
 
     image_uris = {
         # US East 1 images
-        ("us-east-1", "xgb_training", "0.1", "x86_64"): (
-            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-sklearn-xgb-training:0.1"
+        ("us-east-1", "training", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-general-ml-training:0.1"
         ),
-        ("us-east-1", "xgb_inference", "0.1", "x86_64"): (
-            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1"
+        ("us-east-1", "inference", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-general-ml-inference:0.1"
         ),
         ("us-east-1", "pytorch_training", "0.1", "x86_64"): (
             "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-pytorch-training:0.1"
@@ -55,11 +84,11 @@ class ModelImages:
             "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-pytorch-inference:0.1"
         ),
         # US West 2 images
-        ("us-west-2", "xgb_training", "0.1", "x86_64"): (
-            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-sklearn-xgb-training:0.1"
+        ("us-west-2", "training", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-general-ml-training:0.1"
         ),
-        ("us-west-2", "xgb_inference", "0.1", "x86_64"): (
-            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1"
+        ("us-west-2", "inference", "0.1", "x86_64"): (
+            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-general-ml-inference:0.1"
         ),
         ("us-west-2", "pytorch_training", "0.1", "x86_64"): (
             "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-pytorch-training:0.1"
@@ -68,12 +97,6 @@ class ModelImages:
             "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-pytorch-inference:0.1"
         ),
         # ARM64 images
-        ("us-east-1", "xgb_inference", "0.1", "arm64"): (
-            "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1-arm64"
-        ),
-        ("us-west-2", "xgb_inference", "0.1", "arm64"): (
-            "507740646243.dkr.ecr.us-west-2.amazonaws.com/aws-ml-images/py312-sklearn-xgb-inference:0.1-arm64"
-        ),
         # Meta Endpoint inference images
         ("us-east-1", "meta-endpoint", "0.1", "x86_64"): (
             "507740646243.dkr.ecr.us-east-1.amazonaws.com/aws-ml-images/py312-meta-endpoint:0.1"

workbench/core/cloud_platform/aws/aws_account_clamp.py

@@ -54,7 +54,10 @@ class AWSAccountClamp:
 
         # Check our Assume Role
         self.log.info("Checking Workbench Assumed Role...")
-        self.aws_session.assumed_role_info()
+        role_info = self.aws_session.assumed_role_info()
+
+        # Check if the Role is a 'ReadOnly' role
+        self.read_only_role = "readonly" in role_info["AssumedRoleArn"].lower()
 
         # Check our Workbench API Key and Load the License
         self.log.info("Checking Workbench API License...")

workbench/core/transforms/features_to_model/features_to_model.py

@@ -37,8 +37,8 @@ class FeaturesToModel(Transform):
         model_import_str=None,
         custom_script=None,
         custom_args=None,
-        training_image="xgb_training",
-        inference_image="xgb_inference",
+        training_image="training",
+        inference_image="inference",
         inference_arch="x86_64",
     ):
         """FeaturesToModel Initialization
@@ -50,8 +50,8 @@ class FeaturesToModel(Transform):
             model_import_str (str, optional): The import string for the model (default None)
             custom_script (str, optional): Custom script to use for the model (default None)
             custom_args (dict, optional): Custom arguments to pass to custom model scripts (default None)
-            training_image (str, optional): Training image (default "xgb_training")
-            inference_image (str, optional): Inference image (default "xgb_inference")
+            training_image (str, optional): Training image (default "training")
+            inference_image (str, optional): Inference image (default "inference")
             inference_arch (str, optional): Inference architecture (default "x86_64")
         """
 

workbench/model_scripts/custom_models/uq_models/generated_model_script.py

@@ -1,6 +1,7 @@
-# Model: NGBoost Regressor with Distribution output
-from ngboost import NGBRegressor
-from xgboost import XGBRegressor  # Base Estimator
+# Model: XGBoost for point predictions + LightGBM with MAPIE for conformalized intervals
+from mapie.regression import ConformalizedQuantileRegressor
+from lightgbm import LGBMRegressor
+from xgboost import XGBRegressor
 from sklearn.model_selection import train_test_split
 
 # Model Performance Scores
@@ -19,19 +20,12 @@ import numpy as np
 import pandas as pd
 from typing import List, Tuple
 
-# Local Imports
-from proximity import Proximity
-
-
-
 # Template Placeholders
 TEMPLATE_PARAMS = {
-    "id_column": "udm_mol_bat_id",
-    "target": "udm_asy_res_intrinsic_clearance_ul_per_min_per_mg_protein",
-    "features": ['bcut2d_logplow', 'numradicalelectrons', 'smr_vsa5', 'fr_lactam', 'fr_morpholine', 'fr_aldehyde', 'slogp_vsa1', 'fr_amidine', 'bpol', 'fr_ester', 'fr_azo', 'kappa3', 'peoe_vsa5', 'fr_ketone_topliss', 'vsa_estate9', 'estate_vsa9', 'bcut2d_mrhi', 'fr_ndealkylation1', 'numrotatablebonds', 'minestateindex', 'fr_quatn', 'peoe_vsa3', 'fr_epoxide', 'fr_aniline', 'minpartialcharge', 'fr_nitroso', 'fpdensitymorgan2', 'fr_oxime', 'fr_sulfone', 'smr_vsa1', 'kappa1', 'fr_pyridine', 'numaromaticrings', 'vsa_estate6', 'molmr', 'estate_vsa1', 'fr_dihydropyridine', 'vsa_estate10', 'fr_alkyl_halide', 'chi2n', 'fr_thiocyan', 'fpdensitymorgan1', 'fr_unbrch_alkane', 'slogp_vsa9', 'chi4n', 'fr_nitro_arom', 'fr_al_oh', 'fr_furan', 'fr_c_s', 'peoe_vsa8', 'peoe_vsa14', 'numheteroatoms', 'fr_ndealkylation2', 'maxabspartialcharge', 'vsa_estate2', 'peoe_vsa7', 'apol', 'numhacceptors', 'fr_tetrazole', 'vsa_estate1', 'peoe_vsa9', 'naromatom', 'bcut2d_chghi', 'fr_sh', 'fr_halogen', 'slogp_vsa4', 'fr_benzodiazepine', 'molwt', 'fr_isocyan', 'fr_prisulfonamd', 'maxabsestateindex', 'minabsestateindex', 'peoe_vsa11', 'slogp_vsa12', 'estate_vsa5', 'numaliphaticcarbocycles', 'bcut2d_mwlow', 'slogp_vsa7', 'fr_allylic_oxid', 'fr_methoxy', 'fr_nh0', 'fr_coo2', 'fr_phenol', 'nacid', 'nbase', 'chi3v', 'fr_ar_nh', 'fr_nitrile', 'fr_imidazole', 'fr_urea', 'bcut2d_mrlow', 'chi1', 'smr_vsa6', 'fr_aryl_methyl', 'narombond', 'fr_alkyl_carbamate', 'fr_piperzine', 'exactmolwt', 'qed', 'chi0n', 'fr_sulfonamd', 'fr_thiazole', 'numvalenceelectrons', 'fr_phos_acid', 'peoe_vsa12', 'fr_nh1', 'fr_hdrzine', 'fr_c_o_nocoo', 'fr_lactone', 'estate_vsa6', 'bcut2d_logphi', 'vsa_estate7', 'peoe_vsa13', 'numsaturatedcarbocycles', 'fr_nitro', 'fr_phenol_noorthohbond', 'rotratio', 'fr_barbitur', 'fr_isothiocyan', 'balabanj', 'fr_arn', 'fr_imine', 'maxpartialcharge', 'fr_sulfide', 'slogp_vsa11', 'fr_hoccn', 'fr_n_o', 'peoe_vsa1', 'slogp_vsa6', 'heavyatommolwt', 'fractioncsp3', 'estate_vsa8', 'peoe_vsa10', 'numaliphaticrings', 'fr_thiophene', 'maxestateindex', 'smr_vsa10', 'labuteasa', 'smr_vsa2', 'fpdensitymorgan3', 'smr_vsa9', 'slogp_vsa10', 'numaromaticheterocycles', 'fr_nh2', 'fr_diazo', 'chi3n', 'fr_ar_coo', 'slogp_vsa5', 'fr_bicyclic', 'fr_amide', 'estate_vsa10', 'fr_guanido', 'chi1n', 'numsaturatedrings', 'fr_piperdine', 'fr_term_acetylene', 'estate_vsa4', 'slogp_vsa3', 'fr_coo', 'fr_ether', 'estate_vsa7', 'bcut2d_chglo', 'fr_oxazole', 'peoe_vsa6', 'hallkieralpha', 'peoe_vsa2', 'chi2v', 'nocount', 'vsa_estate5', 'fr_nhpyrrole', 'fr_al_coo', 'bertzct', 'estate_vsa11', 'minabspartialcharge', 'slogp_vsa8', 'fr_imide', 'kappa2', 'numaliphaticheterocycles', 'numsaturatedheterocycles', 'fr_hdrzone', 'smr_vsa4', 'fr_ar_n', 'nrot', 'smr_vsa8', 'slogp_vsa2', 'chi4v', 'fr_phos_ester', 'fr_para_hydroxylation', 'smr_vsa3', 'nhohcount', 'estate_vsa2', 'mollogp', 'tpsa', 'fr_azide', 'peoe_vsa4', 'numhdonors', 'fr_al_oh_notert', 'fr_c_o', 'chi0', 'fr_nitro_arom_nonortho', 'vsa_estate3', 'fr_benzene', 'fr_ketone', 'vsa_estate8', 'smr_vsa7', 'fr_ar_oh', 'fr_priamide', 'ringcount', 'estate_vsa3', 'numaromaticcarbocycles', 'bcut2d_mwhi', 'chi1v', 'heavyatomcount', 'vsa_estate4', 'chi0v'],
+    "target": "udm_asy_res_value",
+    "features": ['bcut2d_logplow', 'numradicalelectrons', 'smr_vsa5', 'fr_lactam', 'fr_morpholine', 'fr_aldehyde', 'slogp_vsa1', 'fr_amidine', 'bpol', 'fr_ester', 'fr_azo', 'kappa3', 'peoe_vsa5', 'fr_ketone_topliss', 'vsa_estate9', 'estate_vsa9', 'bcut2d_mrhi', 'fr_ndealkylation1', 'numrotatablebonds', 'minestateindex', 'fr_quatn', 'peoe_vsa3', 'fr_epoxide', 'fr_aniline', 'minpartialcharge', 'fr_nitroso', 'fpdensitymorgan2', 'fr_oxime', 'fr_sulfone', 'smr_vsa1', 'kappa1', 'fr_pyridine', 'numaromaticrings', 'vsa_estate6', 'molmr', 'estate_vsa1', 'fr_dihydropyridine', 'vsa_estate10', 'fr_alkyl_halide', 'chi2n', 'fr_thiocyan', 'fpdensitymorgan1', 'fr_unbrch_alkane', 'slogp_vsa9', 'chi4n', 'fr_nitro_arom', 'fr_al_oh', 'fr_furan', 'fr_c_s', 'peoe_vsa8', 'peoe_vsa14', 'numheteroatoms', 'fr_ndealkylation2', 'maxabspartialcharge', 'vsa_estate2', 'peoe_vsa7', 'apol', 'numhacceptors', 'fr_tetrazole', 'vsa_estate1', 'peoe_vsa9', 'naromatom', 'bcut2d_chghi', 'fr_sh', 'fr_halogen', 'slogp_vsa4', 'fr_benzodiazepine', 'molwt', 'fr_isocyan', 'fr_prisulfonamd', 'maxabsestateindex', 'minabsestateindex', 'peoe_vsa11', 'slogp_vsa12', 'estate_vsa5', 'numaliphaticcarbocycles', 'bcut2d_mwlow', 'slogp_vsa7', 'fr_allylic_oxid', 'fr_methoxy', 'fr_nh0', 'fr_coo2', 'fr_phenol', 'nacid', 'nbase', 'chi3v', 'fr_ar_nh', 'fr_nitrile', 'fr_imidazole', 'fr_urea', 'bcut2d_mrlow', 'chi1', 'smr_vsa6', 'fr_aryl_methyl', 'narombond', 'fr_alkyl_carbamate', 'fr_piperzine', 'exactmolwt', 'qed', 'chi0n', 'fr_sulfonamd', 'fr_thiazole', 'numvalenceelectrons', 'fr_phos_acid', 'peoe_vsa12', 'fr_nh1', 'fr_hdrzine', 'fr_c_o_nocoo', 'fr_lactone', 'estate_vsa6', 'bcut2d_logphi', 'vsa_estate7', 'peoe_vsa13', 'numsaturatedcarbocycles', 'fr_nitro', 'fr_phenol_noorthohbond', 'rotratio', 'fr_barbitur', 'fr_isothiocyan', 'balabanj', 'fr_arn', 'fr_imine', 'maxpartialcharge', 'fr_sulfide', 'slogp_vsa11', 'fr_hoccn', 'fr_n_o', 'peoe_vsa1', 'slogp_vsa6', 'heavyatommolwt', 'fractioncsp3', 'estate_vsa8', 'peoe_vsa10', 'numaliphaticrings', 'fr_thiophene', 'maxestateindex', 'smr_vsa10', 'labuteasa', 'smr_vsa2', 'fpdensitymorgan3', 'smr_vsa9', 'slogp_vsa10', 'numaromaticheterocycles', 'fr_nh2', 'fr_diazo', 'chi3n', 'fr_ar_coo', 'slogp_vsa5', 'fr_bicyclic', 'fr_amide', 'estate_vsa10', 'fr_guanido', 'chi1n', 'numsaturatedrings', 'fr_piperdine', 'fr_term_acetylene', 'estate_vsa4', 'slogp_vsa3', 'fr_coo', 'fr_ether', 'estate_vsa7', 'bcut2d_chglo', 'fr_oxazole', 'peoe_vsa6', 'hallkieralpha', 'peoe_vsa2', 'chi2v', 'nocount', 'vsa_estate5', 'fr_nhpyrrole', 'fr_al_coo', 'bertzct', 'estate_vsa11', 'minabspartialcharge', 'slogp_vsa8', 'fr_imide', 'kappa2', 'numaliphaticheterocycles', 'numsaturatedheterocycles', 'fr_hdrzone', 'smr_vsa4', 'fr_ar_n', 'nrot', 'smr_vsa8', 'slogp_vsa2', 'chi4v', 'fr_phos_ester', 'fr_para_hydroxylation', 'smr_vsa3', 'nhohcount', 'estate_vsa2', 'mollogp', 'tpsa', 'fr_azide', 'peoe_vsa4', 'numhdonors', 'fr_al_oh_notert', 'fr_c_o', 'chi0', 'fr_nitro_arom_nonortho', 'vsa_estate3', 'fr_benzene', 'fr_ketone', 'vsa_estate8', 'smr_vsa7', 'fr_ar_oh', 'fr_priamide', 'ringcount', 'estate_vsa3', 'numaromaticcarbocycles', 'bcut2d_mwhi', 'chi1v', 'heavyatomcount', 'vsa_estate4', 'chi0v', 'chiral_centers', 'r_cnt', 's_cnt', 'db_stereo', 'e_cnt', 'z_cnt', 'chiral_fp', 'db_fp'],
     "compressed_features": [],
-    "train_all_data": False,
-    "track_columns": ['udm_asy_res_intrinsic_clearance_ul_per_min_per_mg_protein']
+    "train_all_data": True
 }
 
 
@@ -106,8 +100,10 @@ def convert_categorical_types(df: pd.DataFrame, features: list, category_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 XGBoost model
+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
@@ -132,7 +128,7 @@ def decompress_features(df: pd.DataFrame, features: List[str], compressed_featur
         )
 
     # Decompress the specified compressed features
-    decompressed_features = 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.")
@@ -161,13 +157,11 @@ def decompress_features(df: pd.DataFrame, features: List[str], compressed_featur
 
 if __name__ == "__main__":
     # Template Parameters
-    id_column = TEMPLATE_PARAMS["id_column"]
     target = TEMPLATE_PARAMS["target"]
     features = TEMPLATE_PARAMS["features"]
     orig_features = features.copy()
     compressed_features = TEMPLATE_PARAMS["compressed_features"]
     train_all_data = TEMPLATE_PARAMS["train_all_data"]
-    track_columns = TEMPLATE_PARAMS["track_columns"]  # Can be None
     validation_split = 0.2
 
     # Script arguments for input/output directories
@@ -225,71 +219,175 @@ if __name__ == "__main__":
     print(f"FIT/TRAIN: {df_train.shape}")
     print(f"VALIDATION: {df_val.shape}")
 
-    # We're using XGBoost for point predictions and NGBoost for uncertainty quantification
-    xgb_model = XGBRegressor()
-    ngb_model = NGBRegressor()
-
     # Prepare features and targets for training
     X_train = df_train[features]
     X_validate = df_val[features]
     y_train = df_train[target]
     y_validate = df_val[target]
 
-    # Train both models using the training data
+    # Train XGBoost for point predictions
+    print("\nTraining XGBoost for point predictions...")
+    xgb_model = XGBRegressor(
+        n_estimators=1000,
+        max_depth=6,
+        learning_rate=0.01,
+        subsample=0.8,
+        colsample_bytree=0.8,
+        random_state=42,
+        verbosity=0
+    )
     xgb_model.fit(X_train, y_train)
-    ngb_model.fit(X_train, y_train, X_val=X_validate, Y_val=y_validate)
-
-    # Make Predictions on the Validation Set
-    print(f"Making Predictions on Validation Set...")
-    preds = xgb_model.predict(X_validate)
-
-    # Calculate various model performance metrics (regression)
-    rmse = root_mean_squared_error(y_validate, preds)
-    mae = mean_absolute_error(y_validate, preds)
-    r2 = r2_score(y_validate, preds)
-    print(f"RMSE: {rmse:.3f}")
-    print(f"MAE: {mae:.3f}")
-    print(f"R2: {r2:.3f}")
+
+    # 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)
+
+    print(f"\nXGBoost Point Prediction Performance:")
+    print(f"RMSE: {xgb_rmse:.3f}")
+    print(f"MAE: {xgb_mae:.3f}")
+    print(f"R2: {xgb_r2:.3f}")
+
+    # Define confidence levels we want to model
+    confidence_levels = [0.50, 0.80, 0.90, 0.95]  # 50%, 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}%")
+
+    print(f"\nOverall Model Performance Summary:")
+    print(f"XGBoost RMSE: {xgb_rmse:.3f}")
+    print(f"XGBoost MAE: {xgb_mae:.3f}")
+    print(f"XGBoost R2: {xgb_r2:.3f}")
     print(f"NumRows: {len(df_val)}")
 
+    # 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}")
+
     # Save the trained XGBoost model
     xgb_model.save_model(os.path.join(args.model_dir, "xgb_model.json"))
 
-    # Save the trained NGBoost model
-    joblib.dump(ngb_model, os.path.join(args.model_dir, "ngb_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 the features (this will validate input during predictions)
+    # Save the feature list
     with open(os.path.join(args.model_dir, "feature_columns.json"), "w") as fp:
-        json.dump(orig_features, fp)  # We save the original features, not the decompressed ones
-
-    # Now the Proximity model
-    model = Proximity(df_train, id_column, features, target, track_columns=track_columns)
+        json.dump(features, fp)
+
+    # Save category mappings if any
+    if category_mappings:
+        with open(os.path.join(args.model_dir, "category_mappings.json"), "w") as fp:
+            json.dump(category_mappings, fp)
+
+    # Save model configuration
+    model_config = {
+        "model_type": "XGBoost_MAPIE_CQR_LightGBM",
+        "confidence_levels": 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),
+            "n_validation": len(df_val)
+        }
+    }
+    with open(os.path.join(args.model_dir, "model_config.json"), "w") as fp:
+        json.dump(model_config, fp, indent=2)
 
-    # Now serialize the model
-    model.serialize(args.model_dir)
+    print(f"\nModel training complete!")
+    print(f"Saved 1 XGBoost model and {len(mapie_models)} MAPIE models to {args.model_dir}")
 
 
 #
 # Inference Section
 #
 def model_fn(model_dir) -> dict:
-    """Load and return XGBoost, NGBoost, and Prox Model from model directory."""
+    """Load XGBoost and all MAPIE models from the specified directory."""
+
+    # Load model configuration to know which models to load
+    with open(os.path.join(model_dir, "model_config.json")) as fp:
+        config = json.load(fp)
 
     # Load XGBoost regressor
     xgb_path = os.path.join(model_dir, "xgb_model.json")
     xgb_model = XGBRegressor(enable_categorical=True)
     xgb_model.load_model(xgb_path)
 
-    # Load NGBoost regressor
-    ngb_model = joblib.load(os.path.join(model_dir, "ngb_model.joblib"))
+    # 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"))
 
-    # Deserialize the proximity model
-    prox_model = Proximity.deserialize(model_dir)
+    # Load category mappings if they exist
+    category_mappings = {}
+    category_path = os.path.join(model_dir, "category_mappings.json")
+    if os.path.exists(category_path):
+        with open(category_path) as fp:
+            category_mappings = json.load(fp)
 
     return {
-        "xgboost": xgb_model,
-        "ngboost": ngb_model,
-        "proximity": prox_model
+        "xgb_model": xgb_model,
+        "mapie_models": mapie_models,
+        "confidence_levels": config["confidence_levels"],
+        "category_mappings": category_mappings
     }
 
 
@@ -305,7 +403,7 @@ def input_fn(input_data, content_type):
     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))  # Assumes JSON array of records
+        return pd.DataFrame(json.loads(input_data))
     else:
         raise ValueError(f"{content_type} not supported!")
 
@@ -313,23 +411,26 @@ def input_fn(input_data, content_type):
 def output_fn(output_df, accept_type):
     """Supports both CSV and JSON output formats."""
     if "text/csv" in accept_type:
-        csv_output = output_df.fillna("N/A").to_csv(index=False)  # CSV with N/A for missing values
+        # 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"  # JSON array of records (NaNs -> null)
+        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 with our XGB Quantile Regression Model
+    """Make predictions using XGBoost for point estimates and MAPIE for conformalized intervals
 
     Args:
         df (pd.DataFrame): The input DataFrame
-        models (dict): The dictionary of models to use for predictions
+        models (dict): Dictionary containing XGBoost and MAPIE models
 
     Returns:
-        pd.DataFrame: The DataFrame with the predictions added
+        pd.DataFrame: DataFrame with XGBoost predictions and conformalized intervals
     """
 
     # Grab our feature columns (from training)
@@ -340,44 +441,62 @@ def predict_fn(df, models) -> pd.DataFrame:
     # Match features in a case-insensitive manner
     matched_df = match_features_case_insensitive(df, model_features)
 
-    # Use XGBoost for point predictions
-    df["prediction"] = models["xgboost"].predict(matched_df[model_features])
-
-    # NGBoost predict returns distribution objects
-    y_dists = models["ngboost"].pred_dist(matched_df[model_features])
-
-    # Extract parameters from distribution
-    dist_params = y_dists.params
-
-    # Extract mean and std from distribution parameters
-    df["prediction_uq"] = dist_params['loc']  # mean
-    df["prediction_std"] = dist_params['scale']  # standard deviation
-
-    # Add 95% prediction intervals using ppf (percent point function)
-    # Note: Our hybrid model uses XGB point prediction and NGBoost UQ
-    #  so we need to adjust the bounds to include the point prediction
-    df["q_025"] = np.minimum(y_dists.ppf(0.025), df["prediction"])
-    df["q_975"] = np.maximum(y_dists.ppf(0.975), df["prediction"])
-
-    # Add 90% prediction intervals
-    df["q_05"] = y_dists.ppf(0.05)  # 5th percentile
-    df["q_95"] = y_dists.ppf(0.95)  # 95th percentile
-
-    # Add 80% prediction intervals
-    df["q_10"] = y_dists.ppf(0.10)  # 10th percentile
-    df["q_90"] = y_dists.ppf(0.90)  # 90th percentile
+    # Apply categorical mappings if they exist
+    if models.get("category_mappings"):
+        matched_df, _ = convert_categorical_types(
+            matched_df,
+            model_features,
+            models["category_mappings"]
+        )
 
-    # Add 50% prediction intervals
-    df["q_25"] = y_dists.ppf(0.25)  # 25th percentile
-    df["q_75"] = y_dists.ppf(0.75)  # 75th percentile
+    # Get features for prediction
+    X = matched_df[model_features]
+
+    # 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]
+        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]
+
+    # Add median (q_50) from XGBoost prediction
+    df["q_50"] = df["prediction"]
+
+    # Calculate uncertainty metrics based on 95% interval
+    interval_width = df["q_975"] - df["q_025"]
+    df["prediction_std"] = interval_width / 3.92
 
     # Reorder the quantile columns for easier reading
     quantile_cols = ["q_025", "q_05", "q_10", "q_25", "q_75", "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]
 
-    # Compute Nearest neighbors with Proximity model
-    models["proximity"].neighbors(df)
+    # Uncertainty score
+    df["uncertainty_score"] = interval_width / (np.abs(df["prediction"]) + 1e-6)
+
+    # Confidence bands
+    df["confidence_band"] = pd.cut(
+        df["uncertainty_score"],
+        bins=[0, 0.5, 1.0, 2.0, np.inf],
+        labels=["high", "medium", "low", "very_low"]
+    )
 
-    # Return the modified DataFrame
     return df