Flask with ML spam prediction

This is a Flask App with ML Naive Bayes model for check message on the Spam statement, Ukrainian words oriented.

Also we train and use here Count Vectorizer(so 2 different models will be used for prediction).

All functions for prepearing text in ml.preprocess_text.py:

1. Text Preprocessing (removing emails, mentions, urls)
2. Replace symbols (like like ₽, @, 0, €) to letters
3. Tokenizer
4. StopWords
5. Lemmatizer
6. Stemmer

Installation

For project startup uses docker-compose.yml. In case with prod you can use docker-compose.prod.yml.

You can also use python venv for development. All dependent python packages in requirements.txt

Configuration

Docker-compose use configurations from .env.dev or .env.prod files.

For Flask config uses services/web/project/config.py.

Auth

You must add Bearer Token(API_KEY) to each reqest

App routes

• /healthy (check app)
• /static/path:filename (get static file)
• /upload (upload file to media folder)
• /web/spam-predict (simple UI form for checking Spam)
• /api/spam-predict (api request for checking Spam)
• /api/set-models (api request for setting Naive Bayes and Count Vectorizer ml models)
• /api/train-data-upload (api request for upload new train file for ML)
• /api/train-models (train new Naive Bayes and Count Vectorizer models)
• /api/translit-convertor (cyrillic to rs convertor)
• /api/spec-symb (replacing special characters with letters)
• /api/test-compare-models (for testing/compare old and new models)

Usage Example

For example we will check spam staitment uses json request

POST http://host:5000/api/spam-predict
BODY
{
    "message": "Я спав а він стріляв",
    "rate": 0.8
}

the response will be like

{
    "result": {
        "deleteRank": 0.8,
        "isSpam": false,
        "spamRate": 0.2528499848085302,
        "tensorToken": "я спат а стріл"
    },
    "time": 6.2989999999999995
}