services/market_services.py

import pandas as pd
import numpy as np
import joblib
import os
from fastapi import APIRouter, HTTPException
from pydantic import BaseModel
from typing import List, Dict, Optional

# --- Load Models & Data ---
MODELS_DIR = 'models'
models = {}

feature_names = [
    'dayofweek', 'dayofyear', 'month', 'year', 'quarter', 'weekofyear',
    'price_lag_7', 'price_lag_14', 'price_lag_30', 'rolling_mean_30', 'rolling_std_30'
]

# Ensure models dir exists
if os.path.exists(MODELS_DIR):
    for model_file in os.listdir(MODELS_DIR):
        if model_file.endswith('.pkl'):
            commodity_name = model_file.replace('.pkl', '').replace('_', '/')
            models[commodity_name] = joblib.load(os.path.join(MODELS_DIR, model_file))
            print(f"✅ Model loaded for: {commodity_name}")

try:
    # Ensure your CSV is accessible
    DF_FULL = pd.read_csv('final_output.csv', parse_dates=['created_at'], index_col='created_at')
    print("✅ Dataset loaded.")
except FileNotFoundError:
    print("❌ 'final_output.csv' not found. Predictions will fail.")
    DF_FULL = None 

# --- Helper Functions ---

# def _create_features(df):
#     df = df.copy()
#     df['dayofweek'] = df.index.dayofweek
#     df['dayofyear'] = df.index.dayofyear
#     df['month'] = df.index.month
#     df['year'] = df.index.year
#     df['quarter'] = df.index.quarter
#     df['weekofyear'] = df.index.isocalendar().week.astype(int)
#     # Lags and Rolling features
#     df['price_lag_7'] = df['modal_price'].shift(7)
#     df['price_lag_14'] = df['modal_price'].shift(14)
#     df['price_lag_30'] = df['modal_price'].shift(30)
#     df['rolling_mean_30'] = df['modal_price'].shift(1).rolling(window=30).mean()
#     df['rolling_std_30'] = df['modal_price'].shift(1).rolling(window=30).std()
#     return df.dropna()


def _create_features(df):
    df = df.copy()
    df['dayofweek'] = df.index.dayofweek
    df['dayofyear'] = df.index.dayofyear
    df['month'] = df.index.month
    df['year'] = df.index.year
    df['quarter'] = df.index.quarter
    df['weekofyear'] = df.index.isocalendar().week.astype(int)
    
    # Lags and Rolling features
    df['price_lag_7'] = df['modal_price'].shift(7)
    df['price_lag_14'] = df['modal_price'].shift(14)
    df['price_lag_30'] = df['modal_price'].shift(30)
    df['rolling_mean_30'] = df['modal_price'].shift(1).rolling(window=30).mean()
    df['rolling_std_30'] = df['modal_price'].shift(1).rolling(window=30).std()
    
    # REMOVE .dropna() here! We need the last row even if modal_price is NaN.
    return df

def get_market_prediction(model, df_full, commodity, last_known_date):
    """
    Iteratively predicts the next 180 days.
    """
    print(f"--- PREDICTION STARTED FOR {commodity} ---")
    
    # --- FIX START: DISABLE FEATURE NAME CHECK ---
    # This tells XGBoost: "Don't check column names, just trust the order of numbers I give you"
    try:
        booster = model.get_booster()
        booster.feature_names = None
    except Exception:
        pass # If it fails, we proceed anyway
    # --- FIX END ---

    df_commodity = df_full[df_full['commodity'] == commodity]
    df_daily = df_commodity.groupby(df_commodity.index).agg({'modal_price': 'mean'})
    
    future_dates = pd.date_range(start=last_known_date + pd.Timedelta(days=1), periods=180, freq='D')
    
    future_df = pd.DataFrame(index=future_dates)
    future_df['modal_price'] = np.nan
    
    df_extended = pd.concat([df_daily, future_df])
    
    for date in future_dates:
        subset = df_extended.loc[:date]
        if len(subset) < 30: continue 
        
        featured_subset = _create_features(subset)
        target_row = featured_subset.iloc[-1]
        
        if target_row[feature_names].isna().any():
            continue

        # 1. Extract values as a plain Numpy Array
        # Reshape to (1, 11) -> 1 row, 11 features
        input_values = target_row[feature_names].values.reshape(1, -1)
        
        # 2. Predict using the Numpy Array directly (Faster & Safer now)
        # Since we disabled feature_names check, this will now work.
        prediction = model.predict(input_values)[0]
        
        df_extended.loc[date, 'modal_price'] = prediction
    
    daily_forecast_df = df_extended.loc[future_dates].copy()
    daily_forecast_df.rename(columns={'modal_price': 'forecast'}, inplace=True)
    
    print("--- PREDICTION SUCCESS ---")
    return daily_forecast_df



# import pandas as pd
# import numpy as np
# import joblib
# import os
# from fastapi import APIRouter, HTTPException
# from typing import List, Dict, Optional

# # --- Load Models & Data ---
# MODELS_DIR = 'models'
# models = {}

# feature_names = [
#     'dayofweek', 'dayofyear', 'month', 'year', 'quarter', 'weekofyear',
#     'price_lag_7', 'price_lag_14', 'price_lag_30', 'rolling_mean_30', 'rolling_std_30'
# ]

# # Ensure models dir exists
# if os.path.exists(MODELS_DIR):
#     for model_file in os.listdir(MODELS_DIR):
#         if model_file.endswith('.pkl'):
#             commodity_name = model_file.replace('.pkl', '').replace('_', '/')
#             models[commodity_name] = joblib.load(os.path.join(MODELS_DIR, model_file))
#             print(f"✅ Model loaded for: {commodity_name}")

# try:
#     # Ensure your CSV is accessible
#     DF_FULL = pd.read_csv('final_output.csv', parse_dates=['created_at'], index_col='created_at')
#     print("✅ Dataset loaded.")
# except FileNotFoundError:
#     print("❌ 'final_output.csv' not found. Predictions will fail.")
#     DF_FULL = None 

# # --- Helper Functions ---

# def _create_features(df):
#     df = df.copy()
#     df['dayofweek'] = df.index.dayofweek
#     df['dayofyear'] = df.index.dayofyear
#     df['month'] = df.index.month
#     df['year'] = df.index.year
#     df['quarter'] = df.index.quarter
#     df['weekofyear'] = df.index.isocalendar().week.astype(int)
    
#     # Lags and Rolling features
#     df['price_lag_7'] = df['modal_price'].shift(7)
#     df['price_lag_14'] = df['modal_price'].shift(14)
#     df['price_lag_30'] = df['modal_price'].shift(30)
#     df['rolling_mean_30'] = df['modal_price'].shift(1).rolling(window=30).mean()
#     df['rolling_std_30'] = df['modal_price'].shift(1).rolling(window=30).std()
    
#     # CRITICAL: Do NOT dropna() here, or we lose the row we are trying to predict
#     return df

# def get_market_prediction(model, df_full, commodity, last_known_date):
#     """
#     Iteratively predicts the next 180 days.
#     Returns a DataFrame so the route can iterate over it.
#     """
#     print(f"--- PREDICTION STARTED FOR {commodity} ---")
    
#     # --- FIX 1: DISABLE FEATURE NAME CHECK ---
#     # This prevents the "X does not have valid feature names" error
#     try:
#         booster = model.get_booster()
#         booster.feature_names = None
#     except Exception:
#         pass 
#     # -----------------------------------------

#     df_commodity = df_full[df_full['commodity'] == commodity]
#     df_daily = df_commodity.groupby(df_commodity.index).agg({'modal_price': 'mean'})
    
#     future_dates = pd.date_range(start=last_known_date + pd.Timedelta(days=1), periods=180, freq='D')
    
#     future_df = pd.DataFrame(index=future_dates)
#     future_df['modal_price'] = np.nan
    
#     df_extended = pd.concat([df_daily, future_df])
    
#     for date in future_dates:
#         subset = df_extended.loc[:date]
#         if len(subset) < 30: continue 
        
#         featured_subset = _create_features(subset)
#         target_row = featured_subset.iloc[-1]
        
#         # Only check if features are NaN (modal_price is allowed to be NaN)
#         if target_row[feature_names].isna().any():
#             continue

#         # --- FIX 2: Pass Pure Values ---
#         # Reshape to (1, 11) -> 1 row, 11 features
#         input_values = target_row[feature_names].values.reshape(1, -1)
        
#         # Predict using the values (bypassing column name check)
#         prediction = model.predict(input_values)[0]
        
#         df_extended.loc[date, 'modal_price'] = prediction
    
#     # Extract just the future part
#     daily_forecast_df = df_extended.loc[future_dates].copy()
#     daily_forecast_df.rename(columns={'modal_price': 'forecast'}, inplace=True)
    
#     print("--- PREDICTION SUCCESS ---")
    
#     # Returns DataFrame (Matches your route code)
#     return daily_forecast_df