import os
import json
import random

# Tentukan path file JSON secara global agar bisa dibaca dan ditulis ulang
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
JSON_PATH = os.path.join(BASE_DIR, "trained_mdp_policy.json")

def load_policy():
    with open(JSON_PATH, "r") as file:
        return json.load(file)

def save_policy(policy_data):
    with open(JSON_PATH, "w") as file:
        json.dump(policy_data, file, indent=2)

# Load policy ke dalam memori aplikasi
policy = load_policy()

def generate_state_id(stress: str, prod_hours: float, sleep: float, screen: float, physical: float) -> str:
    if prod_hours < 35.0: state_prod = "LOW_PROD"
    elif 35.0 <= prod_hours <= 45.0: state_prod = "NORMAL_PROD"
    else: state_prod = "HIGH_PROD"

    if sleep < 5.8: state_sleep = "LACK_SLEEP"
    elif 5.8 <= sleep <= 7.5: state_sleep = "ENOUGH_SLEEP"
    else: state_sleep = "GOOD_SLEEP"

    if screen < 3.0: state_screen = "LOW_SCREEN"
    elif 3.0 <= screen <= 5.0: state_screen = "MEDIUM_SCREEN"
    else: state_screen = "HIGH_SCREEN"

    if physical < 2.0: state_phys = "LOW_ACT"
    elif 2.0 <= physical <= 4.0: state_phys = "MEDIUM_ACT"
    else: state_phys = "HIGH_ACT"

    return f"STRESS_{stress.upper()}_{state_prod}_{state_sleep}_{state_screen}_{state_phys}"

def get_stress_management_recommendation(stress: str, prod_hours: float, sleep: float, screen: float, physical: float):
    global policy
    user_state = generate_state_id(stress, prod_hours, sleep, screen, physical)

    # Logika Cell 4 dari kode Syifa
    if user_state in policy and len(policy[user_state]) > 0:
        actions_dict = policy[user_state]
        sorted_actions = sorted(actions_dict.items(), key=lambda x: x[1], reverse=True)
        top_5_actions = sorted_actions[:5]
    else:
        # Fallback jika state tidak ditemukan
        top_5_actions = [('travelling', 10.0), ('watching sports', 9.0), ('social media engagement', 8.0), ('pilates', 7.0), ('meditation', 6.0)]

    action_names = [item[0] for item in top_5_actions]
    action_scores = [item[1] for item in top_5_actions]

    total_score = sum(action_scores)
    probabilities = [score / total_score for score in action_scores]

    # Ambil 1 rekomendasi utama secara acak berdasarkan bobot probabilitas
    chosen_recommendation = random.choices(action_names, weights=probabilities, k=1)[0]

    # Format output top 5 agar rapi di API
    formatted_top_5 = [{"activity": name, "score": round(score, 4)} for name, score in top_5_actions]

    return {
        "state_id": user_state,
        "chosen_recommendation": chosen_recommendation,
        "top_5_candidates": formatted_top_5
    }

def process_user_feedback(stress: str, prod_hours: float, sleep: float, screen: float, physical: float, action_given: str, feedback_status: str):
    global policy
    user_state = generate_state_id(stress, prod_hours, sleep, screen, physical)

    # Reload data terbaru dari file sebelum memodifikasi (mencegah desinkronisasi)
    policy = load_policy()

    if user_state in policy and action_given in policy[user_state]:
        current_score = policy[user_state][action_given]

        # Logika Cell 6 penyesuaian skor berdasarkan feedback status
        if feedback_status.lower() == "negative":
            new_score = current_score * 0.70
        else:
            new_score = current_score * 1.10

        policy[user_state][action_given] = new_score
        
        # Simpan kembali perubahan ke file JSON
        save_policy(policy)
        return {"status": "success", "message": f"Skor untuk '{action_given}' pada state '{user_state}' berhasil diperbarui menjadi {round(new_score, 4)}."}
    
    return {"status": "ignored", "message": "State atau tindakan tidak ditemukan dalam policy untuk diperbarui."}