usage.md

Usage

In this visualizer you can train an MLP to fit the data points. Initially, you will have the predictions of an untrained model.

---

Dataset

Generate or upload a dataset to train on. If generating, enter a function and the visualizer will sample points uniformly in the domain based on the settings.

---

Model

This allows you to specify the MLP architecture. You can choose one of the presets or use a custom one by editing the text. Each layer must be in square brackets and the last layer must always be [output_units: 1]. Supported activation functions are: {relu, sigmoid, tanh, leaky_relu, elu, gelu, identity}.

---

Train

This allows you to train the MLP on the dataset. You can specify the optimizer options, and click Train Step to do make one gradient step. If you want to train faster, you can increase the Step increment to do multiple gradient steps each time you click Train Step.