app.py

from lib import github
from fastai.tabular.all import *
from flask import Flask, request, jsonify, abort, render_template
from werkzeug.exceptions import HTTPException
import os
from dotenv import load_dotenv

load_dotenv()

MODEL_PATH = './model/model.pkl'

VIEWER_LOGIN = os.environ.get("VIEWER_LOGIN")
if VIEWER_LOGIN is None:
    print("VIEWER_LOGIN not found")
    exit(0)

GITHUB_API_TOKEN = os.environ.get("GITHUB_API_TOKEN")
if GITHUB_API_TOKEN is None:
    print("GITHUB_API_TOKEN not found")
    exit(0)

# TODO Share with notebook
def _preprocess(df):
    # Fill missing values with zeroes
    df = df.fillna(value=0)
    # Clip outliers in scoring inputs to provide a better range to the regression
    df['viewer_events_count'].clip(upper=3, inplace=True)
    df['viewer_comments_count'].clip(upper=3, inplace=True)
    df['viewer_comments_body_count'].clip(upper=1000, inplace=True)
    # Inconsistencies between githubarchive and Github API, cap to avoid noise
    df['viewer_repo_issues_opened_count'].clip(upper=5, inplace=True)
    df['viewer_repo_pull_requests_opened_count'].clip(upper=5, inplace=True)
    df['creator_repo_issues_opened_count'].clip(upper=5, inplace=True)
    df['creator_repo_pull_requests_opened_count'].clip(upper=5, inplace=True)
    return df

app = Flask(__name__, static_url_path='/static')

@app.route('/')
def index():
    return render_template('index.html')


@app.route('/predict', methods=['POST'])
def predict():
    data = request.get_json()

    # Retrieve params from Github
    # Let Flask handle error handling on params
    try:
        params = github.fetch(
            url=data['url'],
            viewer_login=VIEWER_LOGIN,
            github_api_token=GITHUB_API_TOKEN
        )
    except ValueError:
        abort(400, description='Not a valid URL')
    except:
        abort(500, description='Could not retrieve data from Github')

    # Create DataFrame
    test_df = pd.DataFrame.from_dict([params])

    # Reapply preprocessing
    test_df = _preprocess(test_df)

    # Scores were normalised to a 0-1 range,
    # so predictions in this range express level of interest
    learn = load_learner(MODEL_PATH)
    row, pred, _ = learn.predict(test_df.iloc[0])

    return jsonify({'pred': pred.item()})


@app.errorhandler(HTTPException)
def handle_exception(e):
    """Return JSON instead of HTML for HTTP errors."""
    # start with the correct headers and status code from the error
    response = e.get_response()
    # replace the body with JSON
    response.data = json.dumps({
        "code": e.code,
        "name": e.name,
        "description": e.description,
    })
    response.content_type = "application/json"
    return response

if __name__ == '__main__':
    app.run(debug=False, port=os.getenv('PORT', 5000))