Added the CatBoost Model

Author: labrijisaad

conf/base/catalog.yml

@@ -132,7 +132,7 @@ train_test_split_visualization:
 # Trained xgboost model
 xgboost_model:
   type: pickle.PickleDataset
-  filepath: data/05_model_output/xgboost_model.pkl
+  filepath: data/05_model_output/xgboost/xgboost_model.pkl
   backend: pickle
   metadata:
     kedro-viz:
@@ -141,7 +141,7 @@ xgboost_model:
 # Trained Random Forest model
 random_forest_model:
   type: pickle.PickleDataset
-  filepath: data/05_model_output/random_forest_model.pkl
+  filepath: data/05_model_output/random_forest/random_forest_model.pkl
   backend: pickle
   metadata:
     kedro-viz:
@@ -150,7 +150,7 @@ random_forest_model:
 # Trained LightGBM model
 lightgbm_model:
   type: pickle.PickleDataset
-  filepath: data/05_model_output/lightgbm_model.pkl
+  filepath: data/05_model_output/lightgbm/lightgbm_model.pkl
   backend: pickle
   metadata:
     kedro-viz:
@@ -159,7 +159,7 @@ lightgbm_model:
 # xgboost feature importance plot
 xgboost_feature_importance_plot:
   type: matplotlib.MatplotlibWriter
-  filepath: data/04_reporting/xgboost_feature_importance_plot.png
+  filepath: data/04_reporting/xgboost/xgboost_feature_importance_plot.png
   save_args:
     format: png
   metadata:
@@ -169,7 +169,7 @@ xgboost_feature_importance_plot:
 # lightgbm feature importance plot
 lightgbm_feature_importance_plot:
   type: matplotlib.MatplotlibWriter
-  filepath: data/04_reporting/lightgbm_feature_importance_plot.png
+  filepath: data/04_reporting/lightgbm/lightgbm_feature_importance_plot.png
   save_args:
     format: png
   metadata:
@@ -179,7 +179,7 @@ lightgbm_feature_importance_plot:
 # Random Forest feature importance plot
 random_forest_feature_importance_plot:
   type: matplotlib.MatplotlibWriter
-  filepath: data/04_reporting/random_forest_feature_importance_plot.png
+  filepath: data/04_reporting/random_forest/random_forest_feature_importance_plot.png
   save_args:
     format: png
   metadata:
@@ -189,7 +189,7 @@ random_forest_feature_importance_plot:
 # real_data_and_xgboost_predictions_plot 
 real_data_and_xgboost_predictions_plot:
   type: matplotlib.MatplotlibWriter
-  filepath: data/04_reporting/real_data_and_xgboost_predictions_plot.png
+  filepath: data/04_reporting/xgboost/real_data_and_xgboost_predictions_plot.png
   save_args:
     format: png
   metadata:
@@ -199,7 +199,7 @@ real_data_and_xgboost_predictions_plot:
 # real_data_and_lightgbm_predictions_plot 
 real_data_and_lightgbm_predictions_plot:
   type: matplotlib.MatplotlibWriter
-  filepath: data/04_reporting/real_data_and_lightgbm_predictions_plot.png
+  filepath: data/04_reporting/lightgbm/real_data_and_lightgbm_predictions_plot.png
   save_args:
     format: png
   metadata:
@@ -209,7 +209,56 @@ real_data_and_lightgbm_predictions_plot:
 # real_data_and_rf_predictions_plot 
 real_data_and_rf_predictions_plot:
   type: matplotlib.MatplotlibWriter
-  filepath: data/04_reporting/real_data_and_rf_predictions_plot.png
+  filepath: data/04_reporting/random_forest/real_data_and_rf_predictions_plot.png
+  save_args:
+    format: png
+  metadata:
+    kedro-viz:
+      layer: reporting
+
+# Trained CatBoost model
+catboost_model:
+  type: pickle.PickleDataset
+  filepath: data/05_model_output/catboost/catboost_model.pkl
+  backend: pickle
+  metadata:
+    kedro-viz:
+      layer: model
+
+# CatBoost feature importance plot
+catboost_feature_importance_plot:
+  type: matplotlib.MatplotlibWriter
+  filepath: data/04_reporting/catboost/catboost_feature_importance_plot.png
+  save_args:
+    format: png
+  metadata:
+    kedro-viz:
+      layer: reporting
+
+# Real data and CatBoost predictions plot
+real_data_and_catboost_predictions_plot:
+  type: matplotlib.MatplotlibWriter
+  filepath: data/04_reporting/catboost/real_data_and_catboost_predictions_plot.png
+  save_args:
+    format: png
+  metadata:
+    kedro-viz:
+      layer: reporting
+
+# CatBoost SHAP summary plot for model explainability
+catboost_shap_summary_plot:
+  type: matplotlib.MatplotlibWriter
+  filepath: data/04_reporting/catboost/catboost_shap_summary_plot.png
+  save_args:
+    format: png
+  metadata:
+    kedro-viz:
+      layer: reporting
+
+# CatBoost Partial Dependence Plot
+catboost_partial_dependence_plot:
+  type: matplotlib.MatplotlibWriter
+  filepath: data/04_reporting/catboost/catboost_partial_dependence_plot.png
   save_args:
     format: png
   metadata:

conf/base/parameters_catboost_pipeline.yml

@@ -0,0 +1,35 @@
+# Parameters specific to the CatBoost model
+catboost_training_pipeline.catboost_model_params:
+  target: total_consumption
+  threshold: 2010-05-17
+  iterations: 1000
+  depth: 3
+  learning_rate: 0.01
+  loss_function: 'RMSE'
+  eval_metric: 'RMSE'
+  verbose_eval: 100
+  random_state: 42
+  data_types:
+    boolean_columns: 
+      - 'is_holiday'
+      - 'conditions_clear'
+      - 'conditions_overcast'
+      - 'conditions_partiallycloudy'
+      - 'conditions_rain'
+      - 'conditions_rainovercast'
+      - 'conditions_rainpartiallycloudy'
+      - 'conditions_snowovercast'
+      - 'conditions_snowpartiallycloudy'
+      - 'conditions_snowrain'
+      - 'conditions_snowrainovercast'
+      - 'conditions_snowrainpartiallycloudy'
+
+# Features for Partial Dependence Plot (PDP)
+catboost_training_pipeline.catboost_pdp_features:
+  - 'temp_lag_1'
+  - 'dayofweek'
+  - 'total_consumption_lag_1'
+  - 'tempmax_lag_1'
+  - 'feelslike_lag_1'
+  - 'tempmin_lag_1'
+  - 'tempmin_rolling_mean_3'

src/energy_forcasting_model/pipelines/catboost_pipeline/__init__.py

@@ -0,0 +1,10 @@
+"""
+This is a boilerplate pipeline 'catboost_pipeline'
+generated using Kedro 0.19.10
+"""
+
+from .pipeline import create_pipeline
+
+__all__ = ["create_pipeline"]
+
+__version__ = "0.1"

src/energy_forcasting_model/pipelines/catboost_pipeline/nodes.py

@@ -0,0 +1,129 @@
+import logging
+import matplotlib.pyplot as plt
+from catboost import CatBoostRegressor
+from shap import TreeExplainer, summary_plot
+from sklearn.inspection import partial_dependence, PartialDependenceDisplay
+
+def train_catboost_model(X_train, y_train, params):
+    """
+    Trains a CatBoost regression model using the given training data and parameters.
+    
+    Args:
+        X_train (DataFrame): Training features.
+        y_train (Series/DataFrame): Training target.
+        params (dict): Dictionary containing CatBoost parameters.
+    
+    Returns:
+        CatBoostRegressor: Trained CatBoost model.
+    """
+    # Initialize logger
+    logger = logging.getLogger(__name__)
+    logger.info("Starting CatBoost model training...")
+
+    # Log model parameters for reproducibility/debugging
+    logger.info(
+        f"CatBoost parameters: iterations={params['iterations']}, "
+        f"depth={params['depth']}, learning_rate={params['learning_rate']}, "
+        f"loss_function={params['loss_function']}, eval_metric={params['eval_metric']}, "
+        f"verbose_eval={params.get('verbose_eval', True)}"
+    )
+
+    # Convert specified columns to boolean based on provided data types
+    boolean_columns = params.get("data_types", {}).get("boolean_columns", [])
+    for col in boolean_columns:
+        if col in X_train.columns:
+            logger.debug(f"Converting column {col} to boolean.")
+            X_train[col] = X_train[col].astype("bool")
+
+    # Drop rows with null values from X_train and align y_train
+    X_train_clean = X_train.dropna()
+    y_train_clean = y_train.loc[X_train_clean.index]
+
+    # Log how many rows were dropped
+    dropped_rows = X_train.shape[0] - X_train_clean.shape[0]
+    logger.info(f"Dropped {dropped_rows} rows with null values from training data.")
+
+    # Instantiate CatBoostRegressor with the given parameters
+    cat_model = CatBoostRegressor(
+        allow_writing_files=False,
+        iterations=params["iterations"],
+        depth=params["depth"],
+        learning_rate=params["learning_rate"],
+        loss_function=params["loss_function"],
+        eval_metric=params["eval_metric"],
+        random_seed=params.get("random_state", 42),
+        verbose=params.get("verbose_eval", True)
+    )
+
+    # Train the model with an evaluation set for monitoring
+    logger.info("Training the CatBoost model...")
+    cat_model.fit(
+        X_train_clean,
+        y_train_clean,
+        eval_set=[(X_train_clean, y_train_clean)],
+        verbose=params.get("verbose_eval", True)
+    )
+
+    # Log the completion of the training process
+    logger.info("CatBoost model training completed successfully.")
+
+    return cat_model
+
+
+def explain_catboost_model(model, X_train):
+    """
+    Generates a SHAP summary plot for the given CatBoost model.
+
+    Args:
+        model: Trained CatBoost model.
+        X_train: Training feature set.
+
+    Returns:
+        A matplotlib figure object with the SHAP summary plot.
+    """
+    logger = logging.getLogger(__name__)
+    logger.info("Computing SHAP values for CatBoost model...")
+
+    # Create a SHAP explainer and compute SHAP values
+    explainer = TreeExplainer(model)
+    shap_values = explainer.shap_values(X_train)
+
+    # Create a SHAP summary plot
+    plt.figure()
+    summary_plot(shap_values, X_train, show=False)
+    fig = plt.gcf()  # Get current figure
+    plt.close(fig)
+    
+    logger.info("SHAP summary plot created successfully.")
+    return fig
+
+
+def plot_partial_dependence_catboost(model, X_train, features):
+    """
+    Generates a partial dependence plot for specified features using the trained CatBoost model.
+    
+    Args:
+        model: Trained CatBoost model.
+        X_train (DataFrame): Training features.
+        features (list): List of feature names or indices for which to compute partial dependence.
+    
+    Returns:
+        matplotlib.figure.Figure: The partial dependence plot figure.
+    """
+    logger = logging.getLogger(__name__)
+    logger.info("Creating partial dependence plot for CatBoost model...")
+    
+    # Create the plot using scikit-learn's PartialDependenceDisplay
+    fig, ax = plt.subplots(figsize=(12, 8))
+    display = PartialDependenceDisplay.from_estimator(
+        model,
+        X_train,
+        features=features,
+        ax=ax
+    )
+    ax.set_title("Partial Dependence Plot")
+    plt.tight_layout()
+    plt.close(fig)
+    
+    logger.info("Partial dependence plot created successfully.")
+    return fig

src/energy_forcasting_model/pipelines/catboost_pipeline/pipeline.py

@@ -0,0 +1,85 @@
+from kedro.pipeline import Pipeline, node, pipeline
+
+from .nodes import (
+    train_catboost_model,
+    explain_catboost_model,
+    plot_partial_dependence_catboost
+)
+
+from ..random_forest_pipeline.nodes import (
+    plot_real_data_and_predictions_with_train,
+    plot_feature_importance,
+    generate_predictions,
+)
+
+def create_pipeline(**kwargs) -> Pipeline:
+    return pipeline(
+        [
+            node(  # Node 1: Train CatBoost Model
+                func=train_catboost_model,
+                inputs=[
+                    "X_train",
+                    "y_train",
+                    "params:catboost_model_params",
+                ],
+                outputs="catboost_model",
+                name="train_catboost_model_node",
+                tags=["model_training", "catboost"],
+            ),
+            node(  # Node 2: Plot Feature Importance
+                func=plot_feature_importance,
+                inputs=[
+                    "catboost_model",
+                    "X_train",
+                ],
+                outputs="catboost_feature_importance_plot",
+                name="plot_feature_importance_node",
+                tags=["feature_importance", "visualization", "catboost", "model_training"],
+            ),
+            node(  # Node 3: Generate Predictions
+                func=generate_predictions,
+                inputs=[
+                    "X_test",
+                    "catboost_model",
+                ],
+                outputs="catboost_model_predictions",
+                name="generate_predictions_node",
+                tags=["predictions", "catboost", "model_training"],
+            ),
+            node(  # Node 4: Plot Real Data and Predictions
+                func=plot_real_data_and_predictions_with_train,
+                inputs=[
+                    "y_train",
+                    "y_test",
+                    "catboost_model_predictions",
+                ],
+                outputs="real_data_and_catboost_predictions_plot",
+                name="plot_real_data_and_predictions_node",
+                tags=["data_visualization", "catboost", "model_training"],
+            ),
+            node(  # Node 5: Generate SHAP Summary Plot for Explainability
+                func=explain_catboost_model,
+                inputs=["catboost_model", "X_train"],
+                outputs="catboost_shap_summary_plot",
+                name="explain_catboost_model_node",
+                tags=["explainability", "catboost", "model_training"],
+            ),
+            node(  # Node 7: Plot Partial Dependence
+                func=plot_partial_dependence_catboost,
+                inputs=["catboost_model", "X_train", "params:catboost_pdp_features"],
+                outputs="catboost_partial_dependence_plot",
+                name="plot_partial_dependence_node",
+                tags=["partial_dependence", "catboost", "model_training"],
+            ),
+        ],
+        tags="model_training",
+        namespace="catboost_training_pipeline",
+        inputs=["X_train", "y_train", "X_test", "y_test"],
+        outputs=[
+            "catboost_model",
+            "catboost_feature_importance_plot",
+            "real_data_and_catboost_predictions_plot",
+            "catboost_shap_summary_plot",
+            "catboost_partial_dependence_plot",
+        ],
+    )