TF-ranking example

This repository contains an example TFTransform/TFRanking pipeline that :

• Serializes generated data in the ELWC format and writes the result in TFrecords
• Trains a TFRanking model with the Keras API
• Saves it for later use with a signature function that takes raw example as input for predictions

Usage

To be able to run the code, you should activate a Python Virtual Environment and install the packages in requirements.txt

vitrualenv --python=python3.5.7 venv
source venv/bin/activate

pip install -r requirements.txt

Running the code

• To serialize data, use:

rm -rf ./output/*
python transform_pipeline.py

It will generate a TFRecords file and the TFTransform metadata in the ./output directory.

• To train and save the model, use:

python train_keras_model.py