Elixir 103: Kino, Nx, Explorer, ML, Regression, and VegaLite

This episode of ElixirZone demonstrates:

1) Kino: Interactive widgets for Livebook (https://github.com/livebook-dev/kino)
2) Nx: Multi-dimensional arrays (tensors) and numerical definitions (https://github.com/elixir-nx/nx)
3) Explorer: DataFrames for Elixir (https://github.com/elixir-nx/explorer)
4) VegaLite: Elixir bindings to Vega-Lite (https://github.com/livebook-dev/vega_...)
5) Vega-Lite: A concise, declarative JSON syntax to create an expressive range of visualizations for data analysis and presentation (https://vega.github.io/vega-lite/)
6) An Nx-based machine learning model extending Sean Moriarity's ML-driven univariate regression example (https://dockyard.com/blog/2021/04/08/...) to an ML-driven multivariate regression
7) A regression model calculated using Nx.LinAlg (https://github.com/elixir-nx/nx/blob/...)

You can view and run all .livemd files presented in this episode. They’re available at https://github.com/JamesLavin/ElixirZ...

Code from all ElixirZone episodes can be found at https://github.com/JamesLavin/ElixirZone

0:00 Opening
0:55 Support for Ukraine and explanation that I added ML & regression models to this episode that I didn't mention in my original intro, recorded 5 days earlier
4:18 Resource links
7:39 Overview of libraries we'll use today
17:19 Why you might NOT want to use Elixir for data analysis (yet): R, Python, and Julia
21:55 Other cool Elixir data analysis libraries we won't demo today
29:49 Nx
36:11 Kino.Input
37:17 Kino.ETS
40:19 Kino.Ecto
47:00 Kino.JS
48:25 Explorer
55:23 VegaLite
1:04:00 Kino.VegaLite.push/2
1:16:49 Running Nx requires Rust
1:19:05 Exploring Explorer LiveBook
1:26:27 Using ML and Nx to iteratively approximate the best solution to a multivariate regression model (also: for comprehensions with :reduce option)
1:47:29 Nx.Defn.grad
2:00:51 Using Nx and Nx.LinAlg to directly calculate the exact solution to a linear multivariate regression model
2:12:13 Plotting math functions using Kino.VegaLite.push/2 (to simulate a stream of incoming data)
2:21:02 Putting everything together and using our new tools to explore, analyze, and visualize a real dataset (on NHL games). Along the way, we'll clean, transform, and merge/join the data.
2:57:07 Closing statement on Ukraine