Neural Net Playground

This is a small project designed to play around with some functionality of Xethub for visualizing and tracking changing in models. It will focus on building up examples from Chapter 10 of Hands-on Machine Learning with Scikit-Learn, Keras & Tensorflow.

Netron and Pull Requests

The code in src/fashion_minst.py was created primarily to test how the pull request workflow functioned with the ability to display model architectures with Netron.

After training the Keras model on the Fashion MNIST dataset and committing those changes, I created a branch, added a layer, and generated a pull request. On the files page of that pull request in the panel models/fashion_mnist_model.keras you can see the before and after rendered in the browser.

Custom Views and Learning Curves

Similarly, I wanted to experience how custom views could be leveraged to see a model's predictive evolution over time. Using the Decision Tree/Feature Importance Custom Views Example, I built a flow that saved the learning curves after training another model in src/california_regression.py on the California Housing price regression dataset.

On each commit to the main branch, a XetHub Action kicks off, invoking the run.sh script, which in turn runs src/california_regression.py and commits the images generated by the script to a custom-views branch. As the model changes with each commit, you can see the learning curves change as well by going to the branch and navigating through the folders (which are named using the commit hash).