Interactive machine learning demonstrations. A clickable graph allow you to input data. Use the controls buttons to begin a gradient descent algorithm (TODO: more fun things!). Visualie the current model fit alongside the best analytic fit, with stats such as model cost history and a graph displaying the cost for nearby parameter values.
See it live at ozydingo.github.io/descend
Here's a screenshot:
- Linear Regression (done)
- Logistic regression & classification (todo)
- Autoregressive model (todo)
- Basic neural net (todo)
- Adjustible learning rate parameter.
- Add non-linear terms (orders, sin).
- Add trace to costscape to show history of model param estimates