import os
from pathlib import Path
import pandas as pd
from sklearn.preprocessing import MultiLabelBinarizer
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from utils.game import create_gameList
BASE_DIR = Path(__file__).resolve().parent.parent
DATA_DIR = BASE_DIR / "data"
CSV_PATH = DATA_DIR / "games_dataset.csv"
df = None
mlb_tags = None
mlb_genres = None
mlb_platforms = None
model = None
def setup_model():
global df, mlb_tags, mlb_genres, mlb_platforms, model
if not CSV_PATH.exists():
print("[INFO] games_dataset.csv not found. Creating dataset...")
create_gameList()
df = pd.read_csv(CSV_PATH, delimiter=";")
df["release_date"] = pd.to_datetime(df["release_date"], errors="coerce")
df["release_month"] = df["release_date"].dt.month.fillna(0).astype(int)
df["tags"] = df["tags"].fillna("").apply(lambda x: x.split(";") if x else [])
df["genres"] = df["genres"].fillna("").apply(lambda x: x.split(";") if x else [])
df["platforms"] = df["platforms"].fillna("").apply(lambda x: x.split(";") if x else [])
mlb_tags = MultiLabelBinarizer()
mlb_genres = MultiLabelBinarizer()
mlb_platforms = MultiLabelBinarizer()
tags_encoded = mlb_tags.fit_transform(df["tags"])
genres_encoded = mlb_genres.fit_transform(df["genres"])
platforms_encoded = mlb_platforms.fit_transform(df["platforms"])
X = np.hstack([
df[["price", "release_month"]].values,
tags_encoded,
genres_encoded,
platforms_encoded,
df[["ccu", "recommendations", "average_forever", "average_2weeks", "median_forever", "median_2weeks"]].fillna(0).values
])
y = df["score"].fillna(df["score"].mean())
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42
)
model = RandomForestRegressor(n_estimators=200, random_state=42)
model.fit(X_train, y_train)
def predict_score(price=None, release_month=None, tags=None, genres=None, platforms=None):
global df, mlb_tags, mlb_genres, mlb_platforms, model
if model is None:
raise RuntimeError("El modelo no está entrenado. Llama a setup_model() primero.")
if price is None:
price = df["price"].median()
if release_month is None:
release_month = 0
if tags is None:
tags = []
if genres is None:
genres = []
if platforms is None:
platforms = []
tags_vector = mlb_tags.transform([tags])
genres_vector = mlb_genres.transform([genres])
platforms_vector = mlb_platforms.transform([platforms])
extras = [0, 0, 0, 0, 0, 0]
X_new = np.hstack([[price, release_month], tags_vector[0], genres_vector[0], platforms_vector[0], extras])
return model.predict([X_new])[0]
