tf-time-series

Time series forecasting pipeline using TensorFlow Extended (TFX). Components of the pipeline are orchestrated using Apache Airflow. All components are visualized in the interactive notebook.

Setup and running

Requirements: python 3.7 and docker (for model serving).

To install all dependencies, download and preprocess dataset and prepare Airflow DAG run:

./setup.sh

Run Airflow web server with:

airflow webserver -p <port>

Parameter port is a port where you want to run Airflow.

Open another terminal window and run a scheduler with:

airflow scheduler

Open a browser at: http://localhost:<port> and trigger a DAG, or trigger it with:

airflow trigger_dag timeseries_forecasting

Model serving

When the pipeline is finished, the model is ready for serving. Set the AIRFLOW_HOME environment variable (the default home path is ~/airflow):

export AIRFLOW_HOME=/path/to/airflow

Pull the Docker image for serving:

docker pull tensorflow/serving

Run Docker container:

docker run -p 8501:8501 --mount type=bind,source=${AIRFLOW_HOME}/serving_model,target=/models/timeseries -e MODEL_NAME=timeseries -t tensorflow/serving

Test model serving with:

python src/test_serving.py