https://github.com/ndrean/linear_regression_nx_axon

Linear Regressions with Nx, matrices, gradient descent, Neural Network with Axon
https://github.com/ndrean/linear_regression_nx_axon

axon elixir nx

Last synced: 6 months ago
JSON representation

Linear Regressions with Nx, matrices, gradient descent, Neural Network with Axon

• Host: GitHub
• URL: https://github.com/ndrean/linear_regression_nx_axon
• Owner: ndrean
• Created: 2023-01-11T09:44:45.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2023-11-15T09:26:16.000Z (almost 2 years ago)
• Last Synced: 2025-02-06T06:46:16.132Z (8 months ago)
• Topics: axon, elixir, nx
• Homepage:
• Size: 31.3 KB
• Stars: 3
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# linear_regression_nx_axon

Goal: learning a bit of `Nx`, numerical functions with `defn` and `Axon`, and how to use `Livebook` and `Vegalite` by playing with linear regression.

I take a "straightforward" model made of points in the form `(i, i+:rand.uniform)` in a range `1..50`. 

I want to find the coefficients of the best linear fit to this set.

I start by calculating the exact solution with `Nx` matrices, in other words by playing with tensors, `transpose` and `LinAlg.inverse`.

Then I use statistical formulas to run this calculations.

Then approaching the solution with a gradient descent. This time, I use the `.grad` macro. 

Some graphics are added to understand the importance of coefficients.

Finally, a first encounter with `Axon` by using the most basic Neural Network. The activation is `:linear` and the loss is `: mean_squared_error`. It is important to use `compile: EXLA`.