Last active
September 24, 2022 13:22
-
-
Save vankesteren/4f1060943ddc1369e50b8b9ec3285968 to your computer and use it in GitHub Desktop.
Gradient descent for linear regression using Zygote AutoDiff
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| # Gradient descent with autodiff for linear regression | |
| using Zygote | |
| # Data | |
| X = randn(1000, 10) | |
| b = (1:10) | |
| y = X * b + randn(1000) | |
| # MSE for linear model | |
| function mse(bhat) | |
| res = y - X * bhat | |
| res'res / length(res) | |
| end | |
| # Gradient w.r.t. parameters | |
| grad(bhat) = gradient(mse, bhat)[1] | |
| # Initialize params & control | |
| bhat = zeros(10) | |
| step = 1.0e-3 | |
| maxit = 1_000_000 | |
| tol = eps(1.0) | |
| # gradient descent | |
| for i = 1:maxit | |
| global bhat | |
| Δ = step .* grad(bhat) | |
| if (Δ'Δ < tol) | |
| break | |
| end | |
| bhat -= Δ | |
| print("loss ", i, " ", mse(bhat), "\n") | |
| end |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment