handler.py

import tensorflow as tf
from typing import Dict, Any
from predictor import FightPredictor
from config import MODEL_PATH
from image_service import get_fighter_image_url
import os

class CustomLSTM(tf.keras.layers.LSTM):
    """Custom LSTM layer that removes time_major from kwargs"""
    @classmethod
    def from_config(cls, config):
        # Remove time_major if present in config
        config.pop('time_major', None)
        return super().from_config(config)


class EndpointHandler:
    
    def __init__(self, model_path: str):
        # Register the custom layer
        tf.keras.utils.get_custom_objects()['LSTM'] = CustomLSTM
        
        full_path = os.path.join(model_path, MODEL_PATH)
        # Load model using path from config
        self.model = tf.keras.models.load_model(str(full_path))
        
        # Initialize predictor
        self.predictor = FightPredictor(self.model)

    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
        """
        Custom inference handler for UFC predictions.

        Args:
            data: A dictionary containing the input data. 
                  Expected keys: "fighter1" and "fighter2", or "random_matchup".

        Returns:
            A dictionary with prediction results and fighter images.
        """
        inputs = data.get("inputs", {})
        ping = inputs.get("ping")
        random_matchup = inputs.get("random_matchup")

        if ping:
            return {"pong": True}
        
        if not random_matchup:
            fighter1_name = inputs.get("fighter1")
            fighter2_name = inputs.get("fighter2")
            if not fighter1_name or not fighter2_name:
                return {"error": "Both 'fighter1' and 'fighter2' must be provided when not using random_matchup"}
        else:
            fighter1_name = None
            fighter2_name = None

        # Get prediction using FightPredictor
        result = self.predictor.get_prediction(fighter1_name, fighter2_name, verbose=True)
        
        if result is None:
            return {"error": "Prediction failed"}
            
        f1_data, f2_data, details = result
        
        # Get fighter images
        f1_image = get_fighter_image_url(f1_data['name'])
        f2_image = get_fighter_image_url(f2_data['name'])

        return {
            "fighter1": {
                "name": f1_data['name'],
                "image_url": f1_image,
                "probability": f1_data['prob'],
                "details": {
                    "form_score": float(f1_data['form_score']),
                    "total_fights": f1_data['total_fights']
                }
            },
            "fighter2": {
                "name": f2_data['name'],
                "image_url": f2_image,
                "probability": f2_data['prob'],
                "details": {
                    "form_score": float(f2_data['form_score']),
                    "total_fights": f2_data['total_fights']
                }
            },
            "details": {
                "age_difference": float(details['age_difference'])
            }
        }