Introduction

TF Watcher is a simple to use Python package and web app which allows you to easily monitor your model training or testing process on mobile devices. We built this to specially support easily monitoring training or testing in Google Colab, AzureML and Kaggle though this can pretty much be used on any machine or remote server.

See also

Checkout this quickstart example which you can run directly on Google Colab to get started with using this package: Quickstart Example

To make this super easy to use and easily merge in with your development workflow we make use of TensorFlow’s Callbacks which allow us to easily call our code at certain points during model training. This package then accumalates the training data and sends it to Firebase Realtime database allowing you to easily monitor and share live logs from anywhere through the web app.

Almost all the callbacks we made with this package are easily usable by simply specifying them as a callback while training or testing your model (see the documentation for more details):

import tfwatcher

monitor_callback = tfwatcher.callbacks.EpochEnd()
model.fit(..., callbacks=[monitor_callback])

You can also use this projeect in your custom training loops and also works in the same way in a non-eager TensorFlow graph (@tf.function).