train.py

import pandas as pd
from sklearn.model_selection import train_test_split
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.linear_model import LogisticRegression
from sklearn.multioutput import MultiOutputClassifier
from sklearn.metrics import accuracy_score
import joblib
import os
from typing import Dict, Any

from config import DATA_PATH, MODEL_PATH, TFIDF_PATH, MODEL_SAVE_DIR

def train_model() -> Dict[str, Any]:
    try:
        # Ensure the model save directory exists
        os.makedirs(MODEL_SAVE_DIR, exist_ok=True)

        # Load data
        df = pd.read_csv(DATA_PATH)

        # Features and labels
        X = df["Sanction_Context"]
        y = df[["Maker_Action", "Escalation_Level", "Risk_Category", "Risk_Drivers", "Red_Flag_Reason", "Investigation_Outcome"]]

        # Train-test split for evaluation
        X_train, X_test, y_train, y_test = train_test_split(
            X, y, test_size=0.2, random_state=42, stratify=y["Maker_Action"]
        )

        # TF-IDF vectorization
        vectorizer = TfidfVectorizer(max_features=10000, stop_words='english')  # Added max_features and stop_words
        X_train_vec = vectorizer.fit_transform(X_train)
        X_test_vec = vectorizer.transform(X_test)

        # Multi-output Logistic Regression model
        model = MultiOutputClassifier(LogisticRegression(max_iter=1000))
        model.fit(X_train_vec, y_train)

        # Predict on test set
        y_pred = model.predict(X_test_vec)

        # Calculate accuracy per label
        accuracy = {}
        for i, col in enumerate(y.columns):
            accuracy[col] = round(accuracy_score(y_test[col], y_pred[:, i]), 4)

        # Save model and vectorizer
        joblib.dump(model, MODEL_PATH)
        joblib.dump(vectorizer, TFIDF_PATH)

        return {
            "message": f"Model trained and saved to '{MODEL_SAVE_DIR}'",
            "accuracy": accuracy
        }

    except Exception as e:
        return {
            "message": "Training failed",
            "error": str(e)
        }