app.py

from flask import Flask, request, jsonify
import joblib
import numpy as np
import pandas as pd
import os

# Create a Flask application instance
app = Flask(__name__)

# Define the path to the serialized model file
model_filename = 'best_model.joblib'
model_path = os.path.join(os.path.dirname(__file__), model_filename)


# Load the serialized model
try:
    loaded_model = joblib.load(model_path)
    print(f"Model loaded successfully from '{model_path}'")
except Exception as e:
    print(f"Error loading model: {e}")
    loaded_model = None # Set model to None if loading fails

# Define the prediction endpoint
@app.route('/predict', methods=['POST'])
def predict():
    if loaded_model is None:
        return jsonify({'error': 'Model not loaded'}), 500

    try:
        # Get the data from the request
        data = request.get_json()

        # Convert the input data to a pandas DataFrame
        # Assuming the input data is a list of dictionaries,
        # where each dictionary represents a data point with features.
        # The order of features should match the training data.
        input_df = pd.DataFrame(data)

        # NOTE: The preprocessor object is not available here.
        # In a real deployment, you would also serialize and load the preprocessor
        # or recreate it with the exact same steps and parameters.
        # For this example, we'll assume the input is already preprocessed or
        # we skip preprocessing for simplicity (not recommended for production).
        # If preprocessing is needed, you would do:
        # input_processed = loaded_preprocessor.transform(input_df)
        # For this example, let's assume the input data is already in the expected
        # format for the loaded model (which expects processed features).
        # In a real scenario, ensure the input data format matches the training data.
        # Assuming the input data is already preprocessed (e.g., one-hot encoded and scaled)
        # and is in the form of a list of lists or a numpy array that can be
        # converted to a format compatible with the loaded model's expected input shape.
        # For simplicity in this example, we will assume the input JSON
        # directly corresponds to the processed features expected by the model.
        # In a real-world scenario, you would need to implement the preprocessing steps here
        # using the serialized preprocessor or by recreating it.

        # Convert input_df to numpy array or appropriate format if needed by the model
        # Based on the training code, the model was trained on a processed numpy array
        # We will assume the input data JSON is structured to represent this processed array.
        # If the input JSON is raw data, you'll need the preprocessor here.
        # For now, let's assume the input JSON data can be directly passed to predict
        # if it's structured correctly as a list of lists or similar.
        # A safer approach would be to expect raw data and use a loaded preprocessor.

        # For now, let's assume the input data JSON is a list of dictionaries,
        # and we convert it to a DataFrame and then to a numpy array for prediction
        # if the model expects a numpy array.
        # However, since the model was trained on X_processed which was a sparse matrix initially
        # from the ColumnTransformer, direct conversion to a numpy array might lose sparsity
        # or cause issues if the original preprocessor's output format is critical.
        # A robust solution requires serializing and loading the preprocessor as well.

        # Given the context of previous cells, X_processed was likely a numpy array after transformation.
        # Let's assume the input JSON data can be directly used to create a numpy array
        # with the correct number of features (34 in this case, based on X_processed shape).

        # Example assuming input data is a list of lists matching the processed features shape
        input_data_processed = np.array(data['features'])


        # Make predictions
        predictions = loaded_model.predict(input_data_processed)

        # Convert predictions to a list
        predictions_list = predictions.tolist()

        # Return the predictions as a JSON response
        return jsonify({'predictions': predictions_list})

    except Exception as e:
        return jsonify({'error': str(e)}), 400

# To run the Flask application (for local testing)
if __name__ == '__main__':
    # For local testing, you can run:
    # Ensure the model file is in the same directory or adjust the model_path
    # app.run(debug=True)
    pass