app.py

import platform
import pickle
import os  # to read environment variables
from flask import Flask, jsonify, request
from flask_cors import CORS

# The deployment pickle references classes defined in sentiment_deploy_ensemble.py.
# Importing the module here makes those classes resolvable when pickle.load
# runs in this (separate) process. Ship sentiment_deploy_ensemble.py alongside app.py.
import sentiment_deploy_ensemble  # noqa: F401  (needed for unpickling)


# ////////////////////////////////////////////////////////////////////////
# TODO: Adjust the code part below
# ////////////////////////////////////////////////////////////////////////

GROUP_ID = 'Modelling-Giants' 
MODEL_FILE = 'route_c_ensemble_fp16.model'     # relative path to your model file (Space root)
MODEL_VERSION = 'v1.0-ensemble'


# Adjust the function below so that it calls your vectorizer and
# classifier functions packaged in the .model file.
def batch_predict(model, items):
    # Pull the batch of texts once (the ensemble classifier batches internally).
    texts = [item['text'] for item in items]

    # 'vectorizer' is a pass-through that applies the same light cleaning used
    # at training time; per-model tokenisation happens inside the classifier.
    X = model['vectorizer'].transform(texts)

    # 'classifier' returns one integer API label in {-1, 0, 1} per input,
    # in the same order as `texts`.
    labels = model['classifier'].predict(X)

    results = []
    for item, label in zip(items, labels):
        results.append({
            "id": item['id'],
            "label": int(label),
        })
    return results


# ////////////////////////////////////////////////////////////////////////
# You should not modify the code below.
# ////////////////////////////////////////////////////////////////////////

app = Flask(__name__)  # set up app
CORS(app)  # set up CORS policies

# load model file
with open(MODEL_FILE, 'rb') as file:
    model = pickle.load(file)

# define meta-data for API
meta_data = {
    "groupID": GROUP_ID,
    "modelFile": MODEL_FILE,
    "modelVersion": MODEL_VERSION,
    "pythonVersion": platform.python_version()
}


# api route
@app.route("/", methods=['GET', 'POST'])
def main():
    if request.method == 'POST':
        items = request.json['items']
        return jsonify({"items": batch_predict(model, items)})  # batch predictions
    else:
        return jsonify({"meta": meta_data})  # meta data


# start the api server when running the script
if __name__ == "__main__":
    # Look for a cloud-provided port variable (e.g., 7860 on Hugging Face),
    # but fallback to 8000 for local execution.
    port = int(os.environ.get("PORT", 8000))
    app.run(host="0.0.0.0", port=port, debug=True)