train.py

import pandas as pd
import pickle
import os
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import train_test_split
import xgboost as xgb

DATA_PATH = "synthetic_transactions_samples_5000.csv"
TEXT_COLUMN = "Sanction_Context"
LABEL_COLUMNS = [
    "Red_Flag_Reason",
    "Maker_Action",
    "Escalation_Level",
    "Risk_Category",
    "Risk_Drivers",
    "Investigation_Outcome"
]

TFIDF_MAX_FEATURES = 5000
NGRAM_RANGE = (1, 2)
USE_STOPWORDS = True
RANDOM_STATE = 42
TEST_SIZE = 0.2

OUTPUT_DIR = "./"
MODEL_SAVE_PATH = os.path.join(OUTPUT_DIR, "xgb_models.pkl")
LABEL_ENCODERS_PATH = os.path.join(OUTPUT_DIR, "label_encoders.pkl")
TFIDF_VECTORIZER_PATH = os.path.join(OUTPUT_DIR, "tfidf_vectorizer.pkl")

def load_data(path):
    df = pd.read_csv(path)
    df.dropna(subset=[TEXT_COLUMN] + LABEL_COLUMNS, inplace=True)
    return df

def save_pickle(obj, path):
    with open(path, "wb") as f:
        pickle.dump(obj, f)

def train():
    print("📥 Loading data...")
    df = load_data(DATA_PATH)
    X = df[TEXT_COLUMN]

    print("🔤 Fitting TF-IDF...")
    stop_words = 'english' if USE_STOPWORDS else None
    tfidf = TfidfVectorizer(
        max_features=TFIDF_MAX_FEATURES,
        ngram_range=NGRAM_RANGE,
        stop_words=stop_words
    )
    X_tfidf = tfidf.fit_transform(X)
    save_pickle(tfidf, TFIDF_VECTORIZER_PATH)

    models = {}
    label_encoders = {}

    for label in LABEL_COLUMNS:
        print(f"🔁 Processing label: {label}")
        le = LabelEncoder()
        y = le.fit_transform(df[label])

        X_train, _, y_train, _ = train_test_split(
            X_tfidf, y, test_size=TEST_SIZE, random_state=RANDOM_STATE
        )

        model = xgb.XGBClassifier(
            use_label_encoder=False,
            eval_metric="mlogloss",
            random_state=RANDOM_STATE
        )
        model.fit(X_train, y_train)

        models[label] = model
        label_encoders[label] = le

    save_pickle(models, MODEL_SAVE_PATH)
    save_pickle(label_encoders, LABEL_ENCODERS_PATH)
    print("✅ Training complete.")

if __name__ == "__main__":
    train()