For serving a text classifier in TensorFlow 2.X, check out my tensorflow-serving-2.0 example.

Toxic Comment Classifier

Restructured and simplified my solution to Kaggle's Toxic Comment Classification Challenge for deployment.

See Accompanying Medium Article for background and more detail.

Usage Requirements

Install all requirements packages using pip install -r requirements.txt

This repo uses the Kaggle API to fetch the data sets. The Kaggle API requires a bit more work setting up:

1. Create a Kaggle account and accept the competition rules
2. Generate an API token from kaggle.com/USERNAME/account (it will prompt you to download a kaggle.json file which contains the credentials)
3. Set Kaggle credentials as environment variables:

export KAGGLE_USERNAME = [kaggle username]
export KAGGLE_KEY = [generated key]

Serve the Model

Run python profanity_detector/main.py which fetches the data, trains the model and exports a SavedModel file into the /model path. The SavedModel subdirectory can be used to serve the model using TensorFlow Serving.

Start the Server

Instructions paraphrased from the TensorFlow Serving repo.

1. Install Docker
2. Fetch the latest version of TensorFlow Serving Docker docker pull tensorflow/serving:latest
3. Specify the directory for export. In the root directory of this repo, run ModelPath="$(pwd)/model"

To start the server:

docker run -t --rm -p 8501:8501 \
    -v "$ModelPath/:/models/toxic_comment_classifier" \
    -e MODEL_NAME=toxic_comment_classifier \
    tensorflow/serving

Sample Curl Command

curl -d '{"signature_name": "predict","inputs":{"input": "raw text goes here"}}' \
  -X POST http://localhost:8501/v1/models/toxic_comment_classifier:predict

Please note limitations of this server:

• preprocessing steps applied to training data set is not applied to input (to be addressed later)