This is an R implementation of Adam Coates' paper on unsupervised learning [1]. The name Aha is derived from Adam-Honglak-Andrew, the initial of the three authors, signifying sort of an epiphany that a single layer network could learn as well as a deep neural net, achieving state-of-the-art result on certain datasets.
To install directly from github, open a terminal, type R, then
devtools::install_github('htso/AHA')
You need the following packages. To install from a terminal, type
install.packages("Rcpp", "LiblineaR", "pixmap", "R.matlab", "R.utils")
In their paper, the algorithm is tested on three datasets : Cifar10, NORB, and STL. I provide the functions to download these datasets directly from the web. But beware that it may take a long time to download certain files, e.g. CIFAR10.
I include a toy dataset, which is small subset of CIFAR10. To load it, just type
data(tinyCifar10)
I provide a demo in the /demo subfolder. To run it,
setwd(system.file(package="AHA"))
demo(runscript)
It may take a long time to finish a run, even on the toy dataset.
Tested it on Linux (ubuntu 14.04) and Windows 7, should work on OS X.
[1] Coates, Adam, Andrew Y. Ng, and Honglak Lee. "An analysis of single-layer networks in unsupervised feature learning." International conference on artificial intelligence and statistics. 2011.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.