Spaces:
Sleeping
Sleeping
| # ============================================================================== | |
| # 1. IMPORT NECESSARY LIBRARIES | |
| # ============================================================================== | |
| import joblib | |
| import numpy as np | |
| from datetime import date | |
| from typing import Annotated, Literal | |
| from fastapi import FastAPI, HTTPException, Form | |
| from fastapi.middleware.cors import CORSMiddleware | |
| from pydantic import BaseModel, Field | |
| # Import custom processing functions from the local 'function.py' module | |
| from function import prepare_data_pre, prepare_data_post, predict, explain | |
| # ============================================================================== | |
| # 2. FASTAPI APPLICATION INITIALIZATION | |
| # ============================================================================== | |
| app = FastAPI( | |
| title="Product Return Prediction API", | |
| version="1.0.1", | |
| description="API for predicting e-commerce product returns before and after delivery." | |
| ) | |
| # Configure CORS middleware to allow cross-origin requests | |
| app.add_middleware( | |
| CORSMiddleware, | |
| allow_origins=["*"], | |
| allow_credentials=True, | |
| allow_methods=["*"], | |
| allow_headers=["*"] | |
| ) | |
| # ============================================================================== | |
| # 3. PYDANTIC DATA MODELS (INPUT VALIDATION) | |
| # ============================================================================== | |
| class ConditionalInputPre(BaseModel): | |
| """Schema for pre-delivery return prediction input.""" | |
| customer_age: int = Field(default=20, ge=18, le=69, description="Customer age between 18 and 69") | |
| product_category: Literal['Groceries', 'Home & Living', 'Sports', 'Electronics', 'Beauty', 'Fashion'] = "Electronics" | |
| payment_method: Literal['Wallet', 'UPI', 'Credit Card', 'Debit Card', 'Cash on Delivery'] = "Wallet" | |
| order_value_usd: float = Field(default=20.00, ge=10.01, le=718.73, description="Order value in USD") | |
| order_date: date | |
| class ConditionalInputPost(BaseModel): | |
| """Schema for post-delivery return prediction input.""" | |
| customer_age: int = Field(default=20, ge=18, le=69, description="Customer age between 18 and 69") | |
| product_category: Literal['Groceries', 'Home & Living', 'Sports', 'Electronics', 'Beauty', 'Fashion'] = "Electronics" | |
| payment_method: Literal['Wallet', 'UPI', 'Credit Card', 'Debit Card', 'Cash on Delivery'] = "Wallet" | |
| order_value_usd: float = Field(default=20.00, ge=10.01, le=718.73, description="Order value in USD") | |
| delivery_time_days: int = Field(default=3, ge=1, le=14, description="Days taken for delivery") | |
| customer_rating: float = Field(default=4.0, ge=1.0, le=5.0, description="Customer rating from 1.0 to 5.0") | |
| order_date: date | |
| # ============================================================================== | |
| # 4. GLOBAL MODEL AND EXPLANATION DATA VARIABLES | |
| # ============================================================================== | |
| # Pre-delivery model state | |
| best_model_pre = None | |
| lime_training_data_pre = None | |
| # Post-delivery model state | |
| best_model_post = None | |
| lime_training_data_post = None | |
| # ============================================================================== | |
| # 5. MODEL LOADING FUNCTION | |
| # ============================================================================== | |
| def load_models() -> bool: | |
| """ | |
| Loads pre-trained machine learning models and LIME training data into global memory. | |
| Ensures models are only loaded if they haven't been initialized yet. | |
| """ | |
| global best_model_pre, lime_training_data_pre, best_model_post, lime_training_data_post | |
| all_requirements = [best_model_pre, lime_training_data_pre, best_model_post, lime_training_data_post] | |
| try: | |
| # Check if any model or data is missing from memory | |
| if any(req is None for req in all_requirements): | |
| print("🔄 [SYSTEM] Loading models and training data into memory...") | |
| # --------------------------------------------------------- | |
| # PRE-DELIVERY MODELS & DATA | |
| # --------------------------------------------------------- | |
| if best_model_pre is None: | |
| print("📦 [LOAD] Loading Pre-Delivery Model...") | |
| best_model_pre = joblib.load("product-return-model/pre/best_model.joblib") | |
| if lime_training_data_pre is None: | |
| print("📊 [LOAD] Loading Pre-Delivery LIME Data...") | |
| lime_training_data_pre = np.load("product-return-model/pre/lime_training_data.npy") | |
| # --------------------------------------------------------- | |
| # POST-DELIVERY MODELS & DATA | |
| # --------------------------------------------------------- | |
| if best_model_post is None: | |
| print("📦 [LOAD] Loading Post-Delivery Model...") | |
| best_model_post = joblib.load("product-return-model/post/best_model.joblib") | |
| if lime_training_data_post is None: | |
| print("📊 [LOAD] Loading Post-Delivery LIME Data...") | |
| lime_training_data_post = np.load("product-return-model/post/lime_training_data.npy") | |
| print("✅ [SYSTEM] All models and data loaded successfully.") | |
| else: | |
| print("⏭️ [SYSTEM] Models are already loaded in memory. Skipping load operation.") | |
| return True | |
| # --------------------------------------------------------- | |
| # EXCEPTION HANDLING & ERROR ROUTING | |
| # --------------------------------------------------------- | |
| except Exception as e: | |
| error_type = type(e).__name__ | |
| error_msg = str(e).lower() | |
| error_raw = str(e) | |
| if error_type == "FileNotFoundError" or "no such file" in error_msg: | |
| raise HTTPException(status_code=500, detail=f"🚨 [FILE ERROR] {error_type}: Model or data file is missing. Ensure the paths are correct. Details: {error_raw}") | |
| elif error_type == "ValueError" or "unpickling" in error_msg: | |
| raise HTTPException(status_code=500, detail=f"🚨 [LOAD ERROR] {error_type}: Failed to load file. Corrupted joblib/npy. Details: {error_raw}") | |
| else: | |
| raise HTTPException(status_code=500, detail=f"🚨 [SYSTEM ERROR] {error_type}: Unexpected error loading models. Details: {error_raw}") | |
| # ============================================================================== | |
| # 6. ROOT ENDPOINT (API METADATA & DOCUMENTATION) | |
| # ============================================================================== | |
| def home() -> dict: | |
| """ | |
| Root endpoint: Provides server health status, API metadata, and detailed usage documentation. | |
| Serves as a friendly landing page for developers integrating this Product Return Prediction API. | |
| """ | |
| print("🌐 [API] Root endpoint accessed. Serving metadata and documentation.") | |
| return { | |
| "status": "✅ Online", | |
| "service": "Product Return Prediction & LIME Explanation API (Pre & Post Delivery)", | |
| "version": "1.0.1", | |
| "live_urls": { | |
| "base_url": "https://silvio0-product-return-api.hf.space", | |
| "documentation": "https://silvio0-product-return-api.hf.space/docs", | |
| "pre_delivery_prediction": "https://silvio0-product-return-api.hf.space/predict/pre", | |
| "post_delivery_prediction": "https://silvio0-product-return-api.hf.space/predict/post", | |
| "pre_delivery_explanation": "https://silvio0-product-return-api.hf.space/explain/pre", | |
| "post_delivery_explanation": "https://silvio0-product-return-api.hf.space/explain/post" | |
| }, | |
| "usage_guide": { | |
| "endpoints": { | |
| "/predict/pre": "POST method - Predicts return probability BEFORE delivery based on order details.", | |
| "/predict/post": "POST method - Predicts return probability AFTER delivery including shipping time and rating.", | |
| "/explain/pre": "POST method - Generates an interactive LIME HTML explanation for pre-delivery factors.", | |
| "/explain/post": "POST method - Generates an interactive LIME HTML explanation for post-delivery factors." | |
| }, | |
| "payload_structure_pre_delivery": { | |
| "customer_age": "integer (Required) - Range: 18 to 69.", | |
| "product_category": "string (Required) - 'Groceries', 'Home & Living', 'Sports', 'Electronics', 'Beauty', or 'Fashion'.", | |
| "payment_method": "string (Required) - 'Wallet', 'UPI', 'Credit Card', 'Debit Card', or 'Cash on Delivery'.", | |
| "order_value_usd": "float (Required) - Range: 10.01 to 718.73.", | |
| "order_date": "string (Required) - Format: 'YYYY-MM-DD'." | |
| }, | |
| "payload_structure_post_delivery": { | |
| "customer_age": "integer (Required) - Range: 18 to 69.", | |
| "product_category": "string (Required) - 'Groceries', 'Home & Living', 'Sports', 'Electronics', 'Beauty', or 'Fashion'.", | |
| "payment_method": "string (Required) - 'Wallet', 'UPI', 'Credit Card', 'Debit Card', or 'Cash on Delivery'.", | |
| "order_value_usd": "float (Required) - Range: 10.01 to 718.73.", | |
| "delivery_time_days": "integer (Required) - Range: 1 to 14.", | |
| "customer_rating": "float (Required) - Range: 1.0 to 5.0.", | |
| "order_date": "string (Required) - Format: 'YYYY-MM-DD'." | |
| }, | |
| "payload_example_pre": { | |
| "customer_age": 25, | |
| "product_category": "Electronics", | |
| "payment_method": "Credit Card", | |
| "order_value_usd": 150.50, | |
| "order_date": "2026-04-10" | |
| }, | |
| "payload_example_post": { | |
| "customer_age": 28, | |
| "product_category": "Fashion", | |
| "payment_method": "Wallet", | |
| "order_value_usd": 45.99, | |
| "delivery_time_days": 3, | |
| "customer_rating": 4.5, | |
| "order_date": "2026-04-12" | |
| } | |
| }, | |
| "author": "Silvio Christian Joe" | |
| } | |
| # ============================================================================== | |
| # 7. PRE-DELIVERY PREDICTION ENDPOINT | |
| # ============================================================================== | |
| def predict_pre(input: Annotated[ConditionalInputPre, Form()]) -> dict: | |
| """ | |
| API Endpoint to predict the likelihood of a product return BEFORE delivery. | |
| Accepts customer and order details via form data. | |
| """ | |
| global best_model_pre | |
| print(f"\n📥 [API REQUEST] Received request at '/predict/pre' for Order Value: ${input.order_value_usd}") | |
| # Ensure models are loaded into memory before processing | |
| if not load_models(): | |
| print("❌ [API ERROR] Failed to load models into memory.") | |
| raise HTTPException(status_code=500, detail="🚨 [SYSTEM ERROR] Critical failure: Machine Learning models could not be loaded.") | |
| try: | |
| print("⚙️ [PROCESSING] Preparing pre-delivery data features...") | |
| # Transform raw input into a model-ready DataFrame using custom function | |
| df_testing = prepare_data_pre( | |
| customer_age = input.customer_age, | |
| product_category = input.product_category, | |
| payment_method = input.payment_method, | |
| order_value_usd = input.order_value_usd, | |
| order_date = input.order_date | |
| ) | |
| print("🧠 [PREDICTING] Running inference through the Pre-Delivery Model...") | |
| # Generate predictions using the loaded model | |
| prediction, returned_proba, prediction_conf = predict(best_model_pre, df_testing) | |
| print(f"✅ [SUCCESS] Prediction generated: {prediction} (Confidence: {prediction_conf})") | |
| # Return the structured JSON response | |
| return { | |
| "prediction": prediction, | |
| "returned_proba": returned_proba, | |
| "prediction_conf": prediction_conf | |
| } | |
| # --------------------------------------------------------- | |
| # EXCEPTION HANDLING & ERROR ROUTING (API LEVEL) | |
| # --------------------------------------------------------- | |
| except Exception as e: | |
| error_type = type(e).__name__ | |
| error_msg = str(e).lower() | |
| error_raw = str(e) | |
| print("\n" + "="*70) | |
| print("💥 [CRITICAL FAILURE] API Request aborted during prediction!") | |
| print("-" * 70) | |
| # 1. Handling Missing Columns/Features during data preparation | |
| if error_type == "KeyError" or "key" in error_msg: | |
| print(f"🚨 [DATA ERROR] {error_type}: Missing required feature/column. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=400, detail=f"[DATA ERROR] {error_type}: A required data field is missing from the processing pipeline. Details: {error_raw}") | |
| # 2. Handling Data Type Mismatches (e.g., trying to process string as float) | |
| elif error_type == "TypeError" or "type" in error_msg: | |
| print(f"🚨 [DATA ERROR] {error_type}: Incompatible data type encountered. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=400, detail=f"[DATA ERROR] {error_type}: Incorrect data type passed to the processing function. Details: {error_raw}") | |
| # 3. Handling Invalid Values (e.g., feature shape mismatch with ML model) | |
| elif error_type == "ValueError" or "value" in error_msg or "shape" in error_msg: | |
| print(f"🚨 [MODEL ERROR] {error_type}: Dimension mismatch or invalid values for model inference. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=422, detail=f"[MODEL ERROR] {error_type}: The input data shape or values do not match the model's expectations. Details: {error_raw}") | |
| # 4. Handling Corrupted Model Objects (e.g., missing .predict() method) | |
| elif error_type == "AttributeError" or "attribute" in error_msg: | |
| print(f"🚨 [SYSTEM ERROR] {error_type}: Model object is corrupted or missing methods. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[SYSTEM ERROR] {error_type}: Internal model architecture error. Details: {error_raw}") | |
| # 5. Handling Unfitted Models (Scikit-Learn specific error) | |
| elif error_type == "NotFittedError" or "fitted" in error_msg: | |
| print(f"🚨 [MODEL ERROR] {error_type}: Attempting to predict using an untrained model. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[MODEL ERROR] {error_type}: The loaded machine learning model is not trained. Details: {error_raw}") | |
| # 6. Fallback for any other unknown errors | |
| else: | |
| print(f"🚨 [UNKNOWN ERROR] {error_type}: Unexpected failure during execution. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[UNKNOWN ERROR] {error_type}: An unexpected server error occurred during prediction. Details: {error_raw}") | |
| # ============================================================================== | |
| # 8. POST-DELIVERY PREDICTION ENDPOINT | |
| # ============================================================================== | |
| def predict_post(input: Annotated[ConditionalInputPost, Form()]) -> dict: | |
| """ | |
| API Endpoint to predict the likelihood of a product return AFTER delivery. | |
| Accepts customer, order, and post-delivery details via form data. | |
| """ | |
| global best_model_post | |
| print(f"\n📥 [API REQUEST] Received request at '/predict/post' for Order Value: ${input.order_value_usd}") | |
| # Ensure models are loaded into memory before processing | |
| if not load_models(): | |
| print("❌ [API ERROR] Failed to load models into memory.") | |
| raise HTTPException(status_code=500, detail="🚨 [SYSTEM ERROR] Critical failure: Machine Learning models could not be loaded.") | |
| try: | |
| print("⚙️ [PROCESSING] Preparing post-delivery data features...") | |
| # Transform raw input into a model-ready DataFrame using custom function | |
| df_testing = prepare_data_post( | |
| customer_age = input.customer_age, | |
| product_category = input.product_category, | |
| payment_method = input.payment_method, | |
| order_value_usd = input.order_value_usd, | |
| delivery_time_days = input.delivery_time_days, | |
| customer_rating = input.customer_rating, | |
| order_date = input.order_date | |
| ) | |
| print("🧠 [PREDICTING] Running inference through the Post-Delivery Model...") | |
| # Generate predictions using the loaded model | |
| prediction, returned_proba, prediction_conf = predict(best_model_post, df_testing) | |
| print(f"✅ [SUCCESS] Prediction generated: {prediction} (Confidence: {prediction_conf})") | |
| # Return the structured JSON response | |
| return { | |
| "prediction": prediction, | |
| "returned_proba": returned_proba, | |
| "prediction_conf": prediction_conf | |
| } | |
| # --------------------------------------------------------- | |
| # EXCEPTION HANDLING & ERROR ROUTING (API LEVEL) | |
| # --------------------------------------------------------- | |
| except Exception as e: | |
| error_type = type(e).__name__ | |
| error_msg = str(e).lower() | |
| error_raw = str(e) | |
| print("\n" + "="*70) | |
| print("💥 [CRITICAL FAILURE] API Request aborted during prediction!") | |
| print("-" * 70) | |
| # 1. Handling Missing Columns/Features during data preparation | |
| if error_type == "KeyError" or "key" in error_msg: | |
| print(f"🚨 [DATA ERROR] {error_type}: Missing required feature/column. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=400, detail=f"[DATA ERROR] {error_type}: A required data field is missing from the processing pipeline. Details: {error_raw}") | |
| # 2. Handling Data Type Mismatches (e.g., trying to process string as float) | |
| elif error_type == "TypeError" or "type" in error_msg: | |
| print(f"🚨 [DATA ERROR] {error_type}: Incompatible data type encountered. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=400, detail=f"[DATA ERROR] {error_type}: Incorrect data type passed to the processing function. Details: {error_raw}") | |
| # 3. Handling Invalid Values (e.g., feature shape mismatch with ML model) | |
| elif error_type == "ValueError" or "value" in error_msg or "shape" in error_msg: | |
| print(f"🚨 [MODEL ERROR] {error_type}: Dimension mismatch or invalid values for model inference. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=422, detail=f"[MODEL ERROR] {error_type}: The input data shape or values do not match the model's expectations. Details: {error_raw}") | |
| # 4. Handling Corrupted Model Objects (e.g., missing .predict() method) | |
| elif error_type == "AttributeError" or "attribute" in error_msg: | |
| print(f"🚨 [SYSTEM ERROR] {error_type}: Model object is corrupted or missing methods. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[SYSTEM ERROR] {error_type}: Internal model architecture error. Details: {error_raw}") | |
| # 5. Handling Unfitted Models (Scikit-Learn specific error) | |
| elif error_type == "NotFittedError" or "fitted" in error_msg: | |
| print(f"🚨 [MODEL ERROR] {error_type}: Attempting to predict using an untrained model. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[MODEL ERROR] {error_type}: The loaded machine learning model is not trained. Details: {error_raw}") | |
| # 6. Fallback for any other unknown errors | |
| else: | |
| print(f"🚨 [UNKNOWN ERROR] {error_type}: Unexpected failure during execution. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[UNKNOWN ERROR] {error_type}: An unexpected server error occurred during prediction. Details: {error_raw}") | |
| # ============================================================================== | |
| # 9. PRE-DELIVERY EXPLANATION ENDPOINT (LIME) | |
| # ============================================================================== | |
| def explain_pre(input: Annotated[ConditionalInputPre, Form()]) -> dict: | |
| """ | |
| API Endpoint to generate a LIME explanation for a PRE-delivery prediction. | |
| Returns an HTML string detailing feature contributions to the model's decision. | |
| """ | |
| global best_model_pre, lime_training_data_pre | |
| print(f"\n📥 [API REQUEST] Received request at '/explain/pre' for Order Value: ${input.order_value_usd}") | |
| # Ensure models and LIME background data are loaded into memory | |
| if not load_models(): | |
| print("❌ [API ERROR] Failed to load models or LIME training data into memory.") | |
| raise HTTPException(status_code=500, detail="🚨 [SYSTEM ERROR] Critical failure: Machine Learning assets could not be loaded.") | |
| try: | |
| print("⚙️ [PROCESSING] Preparing pre-delivery data features for explanation...") | |
| # Transform raw input into a model-ready DataFrame | |
| df_testing = prepare_data_pre( | |
| customer_age = input.customer_age, | |
| product_category = input.product_category, | |
| payment_method = input.payment_method, | |
| order_value_usd = input.order_value_usd, | |
| order_date = input.order_date | |
| ) | |
| print("🔍 [EXPLAINING] Generating LIME explanation. This may take a moment...") | |
| # Generate the explanation HTML using the local explain function | |
| explanation = explain(best_model_pre, lime_training_data_pre, df_testing) | |
| print("✅ [SUCCESS] LIME Explanation HTML generated successfully.") | |
| # Return the generated HTML string wrapped in a JSON response | |
| return { | |
| "explanation_html": explanation | |
| } | |
| # --------------------------------------------------------- | |
| # EXCEPTION HANDLING & ERROR ROUTING (API LEVEL) | |
| # --------------------------------------------------------- | |
| except Exception as e: | |
| error_type = type(e).__name__ | |
| error_msg = str(e).lower() | |
| error_raw = str(e) | |
| print("\n" + "="*70) | |
| print("💥 [CRITICAL FAILURE] API Request aborted during LIME explanation!") | |
| print("-" * 70) | |
| # 1. Handling Missing Columns/Features | |
| if error_type == "KeyError" or "key" in error_msg: | |
| print(f"🚨 [DATA ERROR] {error_type}: Missing required feature/column. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=400, detail=f"[DATA ERROR] {error_type}: A required data field is missing from the processing pipeline. Details: {error_raw}") | |
| # 2. Handling Data Type Mismatches | |
| elif error_type == "TypeError" or "type" in error_msg: | |
| print(f"🚨 [DATA ERROR] {error_type}: Incompatible data type encountered. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=400, detail=f"[DATA ERROR] {error_type}: Incorrect data type passed to the processing function. Details: {error_raw}") | |
| # 3. Handling LIME Data Mismatch (Crucial for Explainable AI) | |
| elif error_type == "ValueError" or "value" in error_msg or "shape" in error_msg: | |
| print(f"🚨 [LIME ERROR] {error_type}: Background training data mismatch. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=422, detail=f"[LIME ERROR] {error_type}: The input data shape does not match the LIME background dataset. Details: {error_raw}") | |
| # 4. Handling Corrupted Model/Explainer Objects | |
| elif error_type == "AttributeError" or "attribute" in error_msg: | |
| print(f"🚨 [SYSTEM ERROR] {error_type}: Model or Explainer object is corrupted. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[SYSTEM ERROR] {error_type}: Internal model architecture error during explanation generation. Details: {error_raw}") | |
| # 5. Handling Unfitted Models | |
| elif error_type == "NotFittedError" or "fitted" in error_msg: | |
| print(f"🚨 [MODEL ERROR] {error_type}: Attempting to explain an untrained model. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[MODEL ERROR] {error_type}: The loaded machine learning model is not trained. Details: {error_raw}") | |
| # 6. Fallback for any other unknown errors | |
| else: | |
| print(f"🚨 [UNKNOWN ERROR] {error_type}: Unexpected failure during execution. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[UNKNOWN ERROR] {error_type}: An unexpected server error occurred during LIME generation. Details: {error_raw}") | |
| # ============================================================================== | |
| # 10. POST-DELIVERY EXPLANATION ENDPOINT (LIME) | |
| # ============================================================================== | |
| def explain_post(input: Annotated[ConditionalInputPost, Form()]) -> dict: | |
| """ | |
| API Endpoint to generate a LIME explanation for a POST-delivery prediction. | |
| Returns an HTML string detailing feature contributions to the model's decision, | |
| including post-delivery metrics like delivery time and customer rating. | |
| """ | |
| global best_model_post, lime_training_data_post | |
| print(f"\n📥 [API REQUEST] Received request at '/explain/post' for Order Value: ${input.order_value_usd}") | |
| # Ensure models and LIME background data are loaded into memory | |
| if not load_models(): | |
| print("❌ [API ERROR] Failed to load Post-Delivery models or LIME training data into memory.") | |
| raise HTTPException(status_code=500, detail="🚨 [SYSTEM ERROR] Critical failure: Machine Learning assets could not be loaded.") | |
| try: | |
| print("⚙️ [PROCESSING] Preparing post-delivery data features for explanation...") | |
| # Transform raw input into a model-ready DataFrame | |
| df_testing = prepare_data_post( | |
| customer_age = input.customer_age, | |
| product_category = input.product_category, | |
| payment_method = input.payment_method, | |
| order_value_usd = input.order_value_usd, | |
| delivery_time_days = input.delivery_time_days, | |
| customer_rating = input.customer_rating, | |
| order_date = input.order_date | |
| ) | |
| print("🔍 [EXPLAINING] Generating Post-Delivery LIME explanation. This may take a moment...") | |
| # Generate the explanation HTML using the local explain function | |
| explanation = explain(best_model_post, lime_training_data_post, df_testing) | |
| print("✅ [SUCCESS] Post-Delivery LIME Explanation HTML generated successfully.") | |
| # Return the generated HTML string wrapped in a JSON response | |
| return { | |
| "explanation_html": explanation | |
| } | |
| # --------------------------------------------------------- | |
| # EXCEPTION HANDLING & ERROR ROUTING (API LEVEL) | |
| # --------------------------------------------------------- | |
| except Exception as e: | |
| error_type = type(e).__name__ | |
| error_msg = str(e).lower() | |
| error_raw = str(e) | |
| print("\n" + "="*70) | |
| print("💥 [CRITICAL FAILURE] API Request aborted during LIME explanation!") | |
| print("-" * 70) | |
| # 1. Handling Missing Columns/Features | |
| if error_type == "KeyError" or "key" in error_msg: | |
| print(f"🚨 [DATA ERROR] {error_type}: Missing required feature/column. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=400, detail=f"[DATA ERROR] {error_type}: A required data field is missing from the processing pipeline. Details: {error_raw}") | |
| # 2. Handling Data Type Mismatches | |
| elif error_type == "TypeError" or "type" in error_msg: | |
| print(f"🚨 [DATA ERROR] {error_type}: Incompatible data type encountered. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=400, detail=f"[DATA ERROR] {error_type}: Incorrect data type passed to the processing function. Details: {error_raw}") | |
| # 3. Handling LIME Data Mismatch (Crucial for Explainable AI) | |
| elif error_type == "ValueError" or "value" in error_msg or "shape" in error_msg: | |
| print(f"🚨 [LIME ERROR] {error_type}: Background training data mismatch. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=422, detail=f"[LIME ERROR] {error_type}: The input data shape does not match the LIME background dataset. Details: {error_raw}") | |
| # 4. Handling Corrupted Model/Explainer Objects | |
| elif error_type == "AttributeError" or "attribute" in error_msg: | |
| print(f"🚨 [SYSTEM ERROR] {error_type}: Model or Explainer object is corrupted. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[SYSTEM ERROR] {error_type}: Internal model architecture error during explanation generation. Details: {error_raw}") | |
| # 5. Handling Unfitted Models | |
| elif error_type == "NotFittedError" or "fitted" in error_msg: | |
| print(f"🚨 [MODEL ERROR] {error_type}: Attempting to explain an untrained model. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[MODEL ERROR] {error_type}: The loaded machine learning model is not trained. Details: {error_raw}") | |
| # 6. Fallback for any other unknown errors | |
| else: | |
| print(f"🚨 [UNKNOWN ERROR] {error_type}: Unexpected failure during execution. Details: {error_raw}") | |
| print("="*70 + "\n") | |
| raise HTTPException(status_code=500, detail=f"[UNKNOWN ERROR] {error_type}: An unexpected server error occurred during LIME generation. Details: {error_raw}") |