ModelViz

A teaching aid to understand neural network training.

---

Elevator Pitch

Epochs? Units? Batch Size? Activation?
There are a plethora of parameters that play with when training an neural network. Students have a hard time understanding the effect of each one. That's why I created ModelViz. ModelViz makes it simple to visually see the effects that these parameters have. Students can customize a neural network, change the training parameters, and see the neural network learn over time. With ModelViz, students can understand how each value can affect the performance of a neural network.

Features

1. Selectable Training Data
   • Train a neural network on different mathematical functions
     • y = sin(x)
     • y = x ^ 2
     • y = 2 ^ x
2. Model Creation
   • Create a model with unlimited layers.
   • Choose the neuron count.
   • Select an activation type.
3. Modifiable Training Parameters
   • Epochs
   • Batch Size
   • Domain of Function
   • Point Count
   • Snapshot Count
4. Interactive Visualization
   • View a graph of your model's predictions over time
   • 
5. KNN Viewer
   • For fun, view a scatter plot of a KNN attempting to model a function

Tech Stack

• Framework: SvelteKit
• Scripts: TypeScript
• Styling: Vanilla CSS
• Deployment: Cloudflare