import re, os
import pandas as pd
import joblib, json
import numpy as np
from pathlib import Path
from sentence_transformers import SentenceTransformer
from kaggle.api.kaggle_api_extended import KaggleApi
os.environ["KAGGLE_USERNAME"] = os.getenv("KAGGLE_USERNAME", "")
os.environ["KAGGLE_KEY"] = os.getenv("KAGGLE_KEY", "")

# Verify if variables exist
if not os.environ["KAGGLE_USERNAME"] or not os.environ["KAGGLE_KEY"]:
    raise ValueError("KAGGLE_USERNAME or KAGGLE_KEY not set in environment!")

api = KaggleApi()
api.authenticate()

MODEL_PATH = "model.pkl"
if not os.path.exists(MODEL_PATH):
    api.dataset_download_file("myfaizal/model", file_name="model.pkl", path=".", unzip=True)


bert_model = SentenceTransformer('all-MiniLM-L6-v2')

def CarPricePrediction(car_names, car_prices, car_sold):
    # inisialisasi
    p_scaler = joblib.load('modeling/results/price_norm.pkl')
    eop_scaler = joblib.load('modeling/results/eop_norm.pkl')
    model = joblib.load(MODEL_PATH)

    pattern = re.compile(r"^Rp\.?\s?\d{1,3}(\.\d{3})+$")
    name_pattern = re.compile(r"^(19[8-9][0-9]|20[0-2][0-9]|2025)\s+\w+(\s+\w+)+$", re.IGNORECASE)

    names = car_names.strip()
    prices = car_prices.strip()
    car_sold = int(car_sold)

    if not pattern.match(prices):
        return 'Harga harus berformat rupiah misal: Rp. 2.000.000 atau Rp 2.000.000'
    
    if not name_pattern.match(names):
        return 'Nama mobil harus berformat: <tahun> Mistubishi Pajero'
    
    test = {
        'car_names': [names],
        'car_prices': [prices]
    }

    test_df = pd.DataFrame(test)

    # feature engineering
    test_df['year'] = test_df['car_names'].str.extract(r'(\b\d{4}\b)').astype(int)

    if car_sold < test_df['year'].tolist()[0]:
        return "Tidak bisa tahun penjualan kurang dari tahun pembelian"

    test_df['car_names_clean'] = test_df['car_names'].str.replace(r'^\d{4}\s+', '', regex=True).str.split(r' - ').str[0].str.strip()
    test_df['estimated_original_price'] = test_df['car_prices'].str.replace(r'\D', '', regex=True).apply(lambda x: int(x) if x else None)
    test_df['ages'] = test_df['year'].apply(lambda x: car_sold - int(x))

    test_df['eop_norm'] = eop_scaler.transform(test_df[['estimated_original_price']])

    car_names_embeddings = bert_model.encode(test_df['car_names_clean'].tolist(), show_progress_bar=True)
    car_names_df = pd.DataFrame(car_names_embeddings, columns=[f'bert_{i}' for i in range(car_names_embeddings.shape[1])])

    test_input = pd.concat([car_names_df, test_df[['eop_norm', 'ages']].reset_index(drop=True)], axis=1)
    test_result = model.predict(test_input)
    test_result = p_scaler.inverse_transform(np.array(test_result).reshape(-1, 1))
    test_result = test_result.flatten()

    test_result_str =  [f"Rp. {format(round(n), ',').replace(',', '.')}" for n in test_result]

    test_df['deprecate_percentage'] = ((test_df['estimated_original_price'] - round(test_result[0])) / test_df['estimated_original_price'])

    return f"""
        Informasi penjualan:
            - {car_names}
            - Harga pembelian: {car_prices}
            - Harga penjualan: {test_result_str[0]} 
            - Tahun penjualan: {car_sold} 
            - Depresiasi sebesar: {round(test_df['deprecate_percentage'].tolist()[0]*100)}%
    """