from flask import Flask, render_template, request
import joblib
import pandas as pd
import google.generativeai as genai
from openai import OpenAI
import os
import time
from dotenv import load_dotenv

# Load environment variables from .env file
load_dotenv()

app = Flask(__name__)

# Load the trained Random Forest models
rf_ferti_name = joblib.load('rf_ferti_name.pkl')
rf_ferti_value = joblib.load('rf_ferti_value.pkl')

# Manually define the encodings based on the provided dictionaries
soil_type_encodings = {'Black': 0, 'Clayey': 1, 'Loamy': 2, 'Red': 3, 'Sandy': 4}
crop_type_encodings = {'Barley': 0, 'Cotton': 1, 'Ground Nuts': 2, 'Maize': 3, 'Millets': 4,
                       'Oil seeds': 5, 'Other Variety': 6, 'Paddy': 7, 'Pulses': 8, 'Sugarcane': 9,
                       'Tobacco': 10, 'Wheat': 11}
fertilizer_name_encodings = {'10-26-26': 0, '14-35-14': 1, '15-15-15': 2, '17-17-17': 3, '20-20': 4,
                             '20-20-20': 5, '28-28': 6, 'Ammonium sulfate': 7, 'Biofertilizer (e.g., Rhizobium)': 8,
                             'Calcium nitrate': 9, 'DAP': 10, 'Ferrous sulfate': 11, 'Magnesium sulfate': 12,
                             'Potassium chloride/Muriate of potash (MOP)': 13, 'Potassium sulfate/Sulfate of potash (SOP)': 14,
                             'Rock phosphate (RP)': 15, 'Single superphosphate (SSP)': 16, 'Triple superphosphate (TSP)': 17,
                             'Urea': 18, 'Zinc sulfate': 19}

# --- ENHANCED LLM CONFIGURATION ---
GEMINI_API_KEY = os.getenv('GEMINI_API_KEY')
NVIDIA_API_KEY = os.getenv('NVIDIA_API_KEY')

if GEMINI_API_KEY:
    genai.configure(api_key=GEMINI_API_KEY)

# Model configurations with retry settings
GEMINI_MODELS = [
    {"name": "gemini-2.0-flash-exp", "max_retries": 2, "timeout": 30, "description": "Latest experimental"},
    {"name": "gemini-1.5-pro-latest", "max_retries": 2, "timeout": 45, "description": "Most capable"},
    {"name": "gemini-1.5-flash", "max_retries": 3, "timeout": 20, "description": "Fast and reliable"},
    {"name": "gemini-1.5-flash-8b", "max_retries": 3, "timeout": 15, "description": "Lightweight"},
]

NVIDIA_MODELS = [
    {"name": "meta/llama-3.2-90b-vision-instruct", "max_retries": 2, "timeout": 40, "description": "High capability"},
    {"name": "meta/llama-3.2-11b-vision-instruct", "max_retries": 2, "timeout": 30, "description": "Balanced"},
]


def retry_with_backoff(func, max_retries=3, initial_delay=1):
    """Retry a function with exponential backoff."""
    for attempt in range(max_retries):
        try:
            return func()
        except Exception as e:
            if attempt == max_retries - 1:
                raise
            delay = initial_delay * (2 ** attempt)
            print(f"  >> Retry {attempt + 1}/{max_retries} after {delay}s (Error: {type(e).__name__})")
            time.sleep(delay)


def generate_with_gemini(prompt, model_config):
    """Generate text using a specific Gemini model with retry logic."""
    model_name = model_config["name"]
    max_retries = model_config.get("max_retries", 2)
    
    def _attempt():
        print(f"  >> Attempting Gemini: {model_name}")
        model = genai.GenerativeModel(model_name)
        response = model.generate_content(prompt)
        
        if not response or not response.text:
            raise ValueError("Empty response from model")
        
        return response.text
    
    try:
        return retry_with_backoff(_attempt, max_retries=max_retries)
    except Exception as e:
        print(f"  >> FAILED {model_name}: {type(e).__name__}")
        return None


def generate_with_nvidia(prompt, model_config):
    """Generate text using NVIDIA API with retry logic."""
    if not NVIDIA_API_KEY:
        return None
    
    model_name = model_config["name"]
    max_retries = model_config.get("max_retries", 2)
    
    def _attempt():
        print(f"  >> Attempting NVIDIA: {model_name}")
        client = OpenAI(
            base_url="https://integrate.api.nvidia.com/v1",
            api_key=NVIDIA_API_KEY
        )
        
        completion = client.chat.completions.create(
            model=model_name,
            messages=[{"role": "user", "content": prompt}],
            max_tokens=500,
            temperature=0.7
        )
        
        response_text = completion.choices[0].message.content
        if not response_text:
            raise ValueError("Empty response from NVIDIA")
        
        return response_text
    
    try:
        return retry_with_backoff(_attempt, max_retries=max_retries)
    except Exception as e:
        print(f"  >> FAILED NVIDIA {model_name}: {type(e).__name__}")
        return None


def generate_ai_suggestions(pred_fertilizer_name):
    """Generate AI suggestions with enhanced fallback system."""
    print("\n" + "=" * 60)
    print(f"🌱 GENERATING AI SUGGESTIONS FOR: {pred_fertilizer_name}")
    print("=" * 60)
    
    prompt = (
        f"For {pred_fertilizer_name} fertilizer, generate 3-4 Short Informative sentences each on a new line. Content should not be very big max to max 4 sentence thats all okay"
        f"Text should be justified and should not contain any special characters."
    )
    
    response_text = None
    used_model = "None"
    
    # PHASE 1: Try Gemini models
    if GEMINI_API_KEY:
        print("\n--- PHASE 1: Trying Gemini Models ---")
        for idx, model_config in enumerate(GEMINI_MODELS, 1):
            print(f"[{idx}/{len(GEMINI_MODELS)}] Testing {model_config['name']}...")
            response_text = generate_with_gemini(prompt, model_config)
            
            if response_text:
                used_model = f"Gemini-{model_config['name']}"
                print(f"  ✓ SUCCESS with {used_model}")
                break
    
    # PHASE 2: Try NVIDIA models (fallback)
    if not response_text and NVIDIA_API_KEY:
        print("\n--- PHASE 2: Trying NVIDIA Models (Fallback) ---")
        for idx, model_config in enumerate(NVIDIA_MODELS, 1):
            print(f"[{idx}/{len(NVIDIA_MODELS)}] Testing {model_config['name']}...")
            response_text = generate_with_nvidia(prompt, model_config)
            
            if response_text:
                used_model = f"NVIDIA-{model_config['name']}"
                print(f"  ✓ SUCCESS with {used_model}")
                break
    
    # PHASE 3: Final fallback
    if not response_text:
        print("\n❌ All LLM providers failed. Using fallback text.")
        response_text = (
            f"{pred_fertilizer_name} is a commonly used fertilizer in agriculture. "
            f"It provides essential nutrients to crops. "
            f"Follow recommended dosage for best results. "
            f"Consult local agricultural experts for specific guidance."
        )
        used_model = "Fallback"
    
    print(f"\n✅ Generated using: {used_model}")
    print("=" * 60 + "\n")
    
    return response_text


@app.route('/', methods=['GET', 'POST'])
def index():
    if request.method == 'POST':
        # Retrieve form data
        temperature = float(request.form['temperature'])
        humidity = float(request.form['humidity'])
        moisture = float(request.form['moisture'])
        soil_type = request.form['soil_type']
        crop_type = request.form['crop_type']
        nitrogen = float(request.form['nitrogen'])
        potassium = float(request.form['potassium'])
        phosphorous = float(request.form['phosphorous'])

        # Encode categorical data
        soil_type_encoded = soil_type_encodings.get(soil_type, -1)
        crop_type_encoded = crop_type_encodings.get(crop_type, -1)

        # Create a DataFrame for the input
        user_input = pd.DataFrame({
            'Temperature': [temperature],
            'Humidity': [humidity],
            'Moisture': [moisture],
            'Nitrogen': [nitrogen],
            'Potassium': [potassium],
            'Phosphorous': [phosphorous],
            'Soil Type': [soil_type_encoded],
            'Crop Type': [crop_type_encoded]
        })

        # Predict Fertilizer Name
        pred_fertilizer_name = rf_ferti_name.predict(user_input)[0]
        pred_fertilizer_name = [name for name, value in fertilizer_name_encodings.items() if value == pred_fertilizer_name][0]

        # Predict Fertilizer Quantity
        pred_fertilizer_qty = rf_ferti_value.predict(user_input)[0]
        
        # Generate AI suggestions with fallback system
        pred_info = generate_ai_suggestions(pred_fertilizer_name)

        return render_template('index.html', prediction=True, fertilizer_name=pred_fertilizer_name,
                               fertilizer_qty=pred_fertilizer_qty, optimal_usage=pred_fertilizer_qty, pred_info=pred_info)
    return render_template('index.html', prediction=False)


if __name__ == '__main__':
    print("\n" + "=" * 60)
    print("🚀 Starting Fertilizer Recommendation App")
    print("=" * 60)
    print(f"📊 Configuration:")
    print(f"  - Gemini API: {'✓ Configured' if GEMINI_API_KEY else '✗ Not configured'}")
    print(f"  - NVIDIA API: {'✓ Configured' if NVIDIA_API_KEY else '✗ Not configured'}")
    print(f"  - Gemini Models: {len(GEMINI_MODELS)}")
    print(f"  - NVIDIA Models: {len(NVIDIA_MODELS)}")
    print("=" * 60 + "\n")
    app.run(port=7860, host='0.0.0.0')