Identity Net

What is Identity Net?

The name for this project comes from the identity matrix. The network is trained to return the same number as its input (between 0 and 1). Developing deep neural networks is tough, and the point of this project is to demonstrate how to train a model and deploy it to an existing react app. It's contrived, but it's also self-validating.

The model is built and trained with pytorch and saved as an onnx model. I like the flexibility I have in building networks in pytorch and the community around it is 👍. Onnx models are very light weight (30kb), and are mostly straightforward to setup. I say mostly because it's still new and the community (and code examples are few and far between).

How is this project organized?

There are two parts to this project that make up an example process of designing a model, training a model, and embedding it in a react app.

1. Generate_Model.ipynb is a Jupyter Notebook that steps through the process of configuring the network, building the dataset, training, and saving. An online version of this notebook is available here. https://notebooks.azure.com/larryschirmer/projects/identitynet

2. src/MainPage/MainPage.js is the component that uses the saved model. The component uses a slider as input and passes the value to src/MainPage/useONNX.js which generates the value and passes it back. An online version of this app is available here https://codesandbox.io/s/k0yy89n57o