import os import re import json import joblib import numpy as np import pickle from crewai import Agent, Task, Crew, LLM from crewai.tools import tool from dataLoader import * # Load trained model, scaler, and feature columns model_filename = "modelUberML.pkl" with open(model_filename, 'rb') as file: model_data = pickle.load(file) ride_model = model_data['model'] scaler = model_data['scaler'] feature_columns = model_data['feature_columns'] # Define Tools for the Agents @tool("predict_ride") def predict_ride(rideID: str, return_json: bool = False): """Predict the success of an Uber ride using the trained ML model and save result to Firebase.""" ride_data_dict = pull_ride_data(rideID) if not ride_data_dict: error = f"Error: No ride data found for document ID '{rideID}'" if return_json: return {"status": "error", "message": error} return error # Define feature mapping (same as training) FEATURE_MAPPING = { "Avg CTAT": "Avg CTAT", "Avg VTAT": "Avg VTAT", "Booking Value": "Booking Value", "Customer Rating": "Customer Rating", "Driver Ratings": "Driver Ratings", "Ride Distance": "Ride Distance", "Booking Status_Cancelled by Driver": "Booking Status_Cancelled by Driver", "Booking Status_Incomplete": "Booking Status_Incomplete", "Booking Status_No Driver Found": "Booking Status_No Driver Found", "Payment Method_Credit Card": "Payment Method_Credit Card", "Payment Method_Debit Card": "Payment Method_Debit Card", "Payment Method_UPI": "Payment Method_UPI", "Payment Method_Uber Wallet": "Payment Method_Uber Wallet", "Vehicle Type_Bike": "Vehicle Type_Bike", "Vehicle Type_Go Mini": "Vehicle Type_Go Mini", "Vehicle Type_Go Sedan": "Vehicle Type_Go Sedan", "Vehicle Type_Premier Sedan": "Vehicle Type_Premier Sedan", "Vehicle Type_Uber XL": "Vehicle Type_Uber XL", "Vehicle Type_eBike": "Vehicle Type_eBike", "Pickup Location": "Pickup Location", "Drop Location": "Drop Location", } # Extract features input_features = {} for firestore_key, model_col in FEATURE_MAPPING.items(): val = ride_data_dict.get(firestore_key) if val is None: error = f"Error: Missing required feature '{firestore_key}'" if return_json: return {"status": "error", "message": error} return error input_features[model_col] = val # Create DataFrame df_sample = pd.DataFrame([input_features]) # One-hot encode categorical features categorical_cols = ['Pickup Location', 'Drop Location'] df_sample = pd.get_dummies(df_sample, columns=categorical_cols) # Ensure all training columns are present for col in feature_columns: if col not in df_sample.columns: df_sample[col] = 0 df_sample = df_sample[feature_columns] # Scale ride_scaled = scaler.transform(df_sample) # Predict prediction = ride_model.predict(ride_scaled)[0] probability = ride_model.predict_proba(ride_scaled)[0] confidence = float(max(probability)) # Save to Firebase success = push_prediction_result(rideID, prediction, confidence) # Prepare result result_dict = { "prediction": int(prediction), "confidence": confidence, "rideID": rideID, "saved_to_firebase": success } if return_json: return result_dict else: return json.dumps(result_dict, indent=2) @tool("format_prediction") def format_prediction(rideID: str, return_json: bool = False): """Fetch prediction + ride data and format into readable output""" pred_data = pull_prediction_data(rideID) ride_data = pull_ride_data(rideID) if not pred_data: error = f"No prediction found for rideID {rideID}" return {"status": "error", "message": f"No prediction found{rideID}"} if return_json else error if not ride_data: error = f"No ride data found for rideID {rideID}" return {"status": "error", "message": error} if return_json else error #Build the formatted result formatted = { "rideID": rideID, "prediction": pred_data.get("prediction", "N/A"), "confidence": pred_data.get("confidence", "N/A"), #Fields "bookingValue": ride_data.get("Booking Value"), "avgCTAT": ride_data.get("Avg CTAT"), "avgVTAT": ride_data.get("Avg VTAT"), "statusCompleted": ride_data.get("Booking Status_Completed"), } return formatted if return_json else json.dumps(formatted, indent=2) llm = LLM( model="huggingface/meta-llama/Llama-3.1-8B-Instruct" ) # Define Agents ride_success_predictor = Agent( role="Ride Success Calculator Agent", goal="Calculate the success or failure of a ride using ML model and LLM interpretation", backstory="You combine machine learning predictions with natural language explanations to provide personalized ride success assessments.", verbose=True, llm=llm, tools=[predict_ride] ) formatting_agent = Agent( role="Formatting Agent", goal="Retrieve prediction + ride data and format into readable output", backstory="You turn raw ML prediction results into clean, readable formatted summaries.", verbose=True, llm=llm, tools=[format_prediction], ) # Define Tasks predict_ride_success = Task( description=f"Calculate ride success/failure using the ML model based on ride features.", expected_output="Binary success/failure, confidence score, and brief explanation.", agent=ride_success_predictor, ) format_prediction_task = Task( description=f"Pull prediction and ride data for a given rideID, then format the results into a readable report", expected_output="A structured summary of ride performance and prediction results.", agent=formatting_agent, ) # Define Crew crew = Crew( agents=[ride_success_predictor, formatting_agent], tasks=[predict_ride_success, format_prediction_task], verbose=True ) def run_crew_for_ride(rideID: str): """Run the full crew pipeline for a given rideID and return the final output.""" # Inject rideID into tasks dynamically for task in crew.tasks: # Dynamically update task descriptions to include rideID if "predict_ride_success" in task.description or "Calculate ride success" in task.description: task.description = f"Predict ride success for rideID: {rideID}. Use the predict_ride tool with this rideID." if "format_prediction_task" in task.description or "Pull prediction and ride data" in task.description: task.description = f"Format the prediction and ride data for rideID: {rideID}. Use the format_prediction tool with this rideID." # Run crew result = crew.kickoff() return result.raw if hasattr(result, 'raw') else str(result) # Optional: Keep CLI for local testing if __name__ == "__main__": import sys if len(sys.argv) > 1: ride_id = sys.argv[1] print(f"Running crew for rideID: {ride_id}") output = run_crew_for_ride(ride_id) print("\n=== FINAL OUTPUT ===\n") print(output) else: print("Usage: python crew.py ")