backend/app/ai/meta_classifier.py

import numpy as np
import joblib
from sklearn.linear_model import LogisticRegression
from app.models.feedback_model import Feedback
from app.db.database import SessionLocal
from pathlib import Path
from app.ai.model_loader import model_loader
from app.core.config import settings

def get_model_path():
    return Path(settings.MODEL_DIR) / model_loader.get_latest_model_version()

def retrain_from_feedback():
    db = SessionLocal()
    
    data = db.query(Feedback).all()
    
    if len(data) < 10:
        print("Not enough feedback to retrain")
        db.close()
        return

    X = []
    y = []
    
    batch = data[:10]
    
    for row in batch:
        X.append([row.freq_score, row.cnn_score])
        y.append(1 if row.label == "ai" else 0)
    
    X = np.array(X)
    y = np.array(y)
    
    model = LogisticRegression()
    model.fit(X, y)
    
    joblib.dump(model, get_model_path())
    
    print("Model retrained on real feedback data")
    
    for row in batch:
        db.delete(row)
    db.commit()
    db.close()

def train_meta_model():
    X = np.array([
        [0.6, 0.5],
        [0.7, 0.6],
        [0.8, 0.7],
        [0.9, 0.8],
        [1.0, 0.9],
        [1.1, 1.0],
        [1.2, 1.1],
        [1.3, 1.2],
        [1.4, 1.3],
        [1.5, 1.4],
    ])
    
    y = np.array([
        0,0,0,0,0,
        1,1,1,1,1
    ])
    
    model = LogisticRegression()
    model.fit(X, y)
    
    joblib.dump(model, get_model_path())
    
def load_model():
    model_path = get_model_path()
    if not model_path.exists():
        train_meta_model()
    return joblib.load(model_path)

def predict_ai(freq_score: float, cnn_score: float):
    model = load_model()
    
    X = np.array([[freq_score, cnn_score]])
    
    prob = model.predict_proba(X)[0][1]
    
    if prob > 0.75:
        label = "Likely AI Generated"
    elif prob > 0.45:
        label = "Suspicious"
    else:
        label = "Likely Real"

    return label, float(prob)