predict.py

import pandas as pd
from config import MODEL_PATH, TFIDF_PATH, TEXT_COLUMN, LABEL_COLUMNS
from utils import load_model_and_vectorizer
from schemas import TransactionData

def predict_labels(input_record: TransactionData):
    # Load the model and vectorizer
    model, vectorizer = load_model_and_vectorizer(MODEL_PATH, TFIDF_PATH)

    # Convert input Pydantic model to DataFrame
    input_data = pd.DataFrame([input_record.dict()])

    # Prepare text input by selecting relevant fields
    sanction_context = input_data[TEXT_COLUMN].iloc[0]

    # Vectorize the input text
    X_vec = vectorizer.transform([sanction_context])

    # Predict labels
    y_pred = model.predict(X_vec)

    # Format predictions
    predictions = {
        LABEL_COLUMNS[i]: y_pred[0][i] for i in range(len(LABEL_COLUMNS))
    }

    return predictions