Usage – External Tools

This page will give an overview over some of the external tools we use in our simulation framework, and how to use them.

• TensorBoard
• Docker

TensorBoard

Find relevant documentation for the TensorBoard tool here.

The TensorBoard is an external tool from the RL library TensorFlow. In our framework, we use it to visualize various metrics collected during training and monitoring sessions, such as total profits achieved by the different vendors or prices set for the various price channels.

During the simulation, the TensorBoard will automatically record these metrics and save them in the results/runs folder of the user's datapath. At any point during or after the simulation, the TensorBoard can be started using the following command:

tensorboard serve --logdir path/to/results/runs

The path specified can be changed to just include one subfolder of the runs folder to track just one of the experiments.

Note: TensorBoard might not work with Safari, but Chrome does the job.

Docker

Find general documentation for docker here and the documentation of the docker SDK for Python here.

We use docker to encapsulate individual simulations in a container, so that we can more easily run multiple simulations in parallel. Docker is utilized automatically when working with our user interface, but can also be set up for local use.

The API between the docker daemon and the recommerce framework was written in Python using the Docker SDK for Python.