MSDA-denosing
The code for Spearmints experiments described in http://fastml.com/very-fast-denoising-autoencoder-with-a-robot-arm/.
spearmint - optimize hyperparams (layers and noise) for a random forest (or ridge regression)
spearmint_variable_noise - optimize noise separately for each of 10 layers
denoised_rf.r - train a random forest on denoised data
f_rmse.r - RMSE function
rf.r - train a random forest on original data
The original experiments described in the article used only the last layer from MSDA. Updated run_denoise.m to use all layers, see user comments at the end of the post.