src/predictor.py

import pandas as pd
from fastai.tabular.all import *
from .mtga import mtga_id_to_card_name, card_name_to_mtga_id

class MTGPickPredictor:
    def __init__(self):
        self._models: Dict[str, any] = {}  # Cache for loaded models
        self._card_names: Dict[str, list] = {}  # Cache for card names per set
     
    def predict(self, set_name, input_data):
        if set_name not in self._models:
            self._load_model(set_name)

        df = self._json_to_df(set_name, input_data)        
        _, _, pred_probs = self._models[set_name].predict(df.iloc[0])

        topk_values, topk_indices = pred_probs.topk(3)
        
        result = ""
        output = []
        for _, (prob, idx) in enumerate(zip(topk_values, topk_indices)):
            card_name = self._card_names[set_name][idx]
            result = result + (f"{card_name}: {prob*100:.0f}%\n")
            output.append((card_name_to_mtga_id(card_name), card_name, prob))
            
        print(result)

        return output
    
    def _load_model(self, set_name: str):
        """Lazily loads a model for a specific set if not already loaded"""
        if set_name in self._models:
            return
        
        model_path = f"models/{set_name}_draft.pkl"
        model = load_learner(model_path)
        self._models[set_name] = model
        # Cache card names for this set
        self._card_names[set_name] = [
            col.replace('pack_card_', '') 
            for col in model.dls.train_ds.cont_names 
            if col.startswith('pack_card_')
        ]
            
    def _json_to_df(self, set_name, json_data):
        # Initialize all card columns with 0
        #all_cols = {f"pack_card_{card}": 0 for card in card_names} | {f"pool_{card}": 0 for card in card_names}
        # TODO: this is shortcut for testing
        all_cols = {col_name: 0 for col_name in self._models[set_name].dls.train_ds.cont_names}
        # Fill in pack cards
        for card_id in json_data['pack']:
            col_name = f"pack_card_{mtga_id_to_card_name(card_id)}"
            all_cols[col_name] += 1
            
        # Fill in pool cards    
        for card_id in json_data['pool']:
            col_name = f"pool_{mtga_id_to_card_name(card_id)}"
            all_cols[col_name] +=1
            
        return pd.DataFrame([all_cols]).astype(float)