backend/logic/deterministic.py

import copy
from .ml_model import ml_engine
import json

KEY_MAP = {
    "gender": "Gender",
    "married": "Married",
    "dependents": "Dependents",
    "education": "Education",
    "self_employed": "Self_Employed",
    "applicant_income": "ApplicantIncome",
    "coapplicant_income": "CoapplicantIncome",
    "loan_amount": "LoanAmount",
    "loan_amount_term": "Loan_Amount_Term",
    "credit_history": "Credit_History",
    "property_area": "Property_Area"
}

def map_keys(data: dict) -> dict:
    return {KEY_MAP.get(k, k): v for k, v in data.items()}

def prepare_input(base_data: dict) -> dict:
    # Compute derived features for the ML Engine
    total_income = base_data["ApplicantIncome"] + base_data["CoapplicantIncome"]
    
    # Avoid division by zero
    term = base_data["Loan_Amount_Term"] if base_data["Loan_Amount_Term"] > 0 else 360
    
    emi = (base_data["LoanAmount"] * 1000) / term
    balance_income = total_income - emi

    base_data["Total_Income"] = total_income
    base_data["EMI"] = emi
    base_data["Balance_Income"] = balance_income
    return base_data

def process_pipeline(input_data: dict) -> dict:
    """Evaluates the loan application and runs sensitivity analysis if rejected."""
    
    # 0. Map snake_case API keys to PascalCase Model keys
    mapped_input = map_keys(input_data)
    
    # 1. Base Evaluation
    processed_input = prepare_input(copy.deepcopy(mapped_input))
    
    result, confidence = ml_engine.predict(processed_input)
    
    dti_ratio = 0.0
    if processed_input["Total_Income"] > 0:
        dti_ratio = processed_input["EMI"] / processed_input["Total_Income"]
    dti_ratio_pct = min(100.0, dti_ratio * 100)

    response = {
        "prediction": result,
        "confidence": confidence,
        "dti_ratio": dti_ratio_pct,
        "optimized_suggestion": "Your profile meets all current thresholds. No optimization necessary.",
        "feature_importances": ml_engine.get_feature_importances(),
        "benchmarks": ml_engine.get_benchmarks()
    }

    # 2. What-If Optimization (If Rejected)
    if result == "N":
        # Simulate lower loan amount
        sim_data_loan = copy.deepcopy(mapped_input)
        loan_decrease_needed = 0
        loan_approved = False
        
        while sim_data_loan["LoanAmount"] > 10 and not loan_approved:
            sim_data_loan["LoanAmount"] -= 10 # Decrease by 10k
            loan_decrease_needed += 10
            
            p_input = prepare_input(copy.deepcopy(sim_data_loan))
            sim_res, sim_conf = ml_engine.predict(p_input)
            if sim_res == "Y":
                loan_approved = True
                break
                
        # Simulate higher coapplicant income
        sim_data_inc = copy.deepcopy(mapped_input)
        inc_increase_needed = 0
        inc_approved = False
        
        while inc_increase_needed < 20000 and not inc_approved:
            sim_data_inc["CoapplicantIncome"] += 1000 # Increase by 1k
            inc_increase_needed += 1000
            
            p_input_inc = prepare_input(copy.deepcopy(sim_data_inc))
            sim_res_inc, sim_conf_inc = ml_engine.predict(p_input_inc)
            if sim_res_inc == "Y":
                inc_approved = True
                break

        # Generate rule-based optimization text (Determnistic)
        suggestion = "Unfortunately, we couldn't find a minor adjustment to approve your loan. Consider improving your CIBIL score."
        
        if loan_approved and inc_approved:
            suggestion = f"Reducing your loan request by ₹{loan_decrease_needed * 1000} OR adding a co-applicant income of ~₹{inc_increase_needed} would likely result in approval."
        elif loan_approved:
            suggestion = f"Reducing your loan request by ₹{loan_decrease_needed * 1000} would likely result in an approval."
        elif inc_approved:
            suggestion = f"Adding a co-applicant income of ~₹{inc_increase_needed} would likely result in an approval."
            
        if mapped_input["Credit_History"] == 0:
             suggestion = "Your Credit History is 0. This is the primary blocking factor. Improving your credit standing is required before other adjustments will work."

        response["optimized_suggestion"] = suggestion

    return response