I second Joe's sentiment! This looks real good (well commented code and lots of references/links to articles/code). Thanks for sharing.
On Wed, Jul 25, 2018 at 11:59 AM, Joe Bogner <[email protected]> wrote: > This looks really good and will be a great help to those who want to get a > better understanding of the algorithms. Thank you for sharing > > On Wed, Jul 25, 2018, 10:00 AM 'Jon Hough' via Programming < > [email protected]> wrote: > > > I added my **Work In Progress** machine learning library to github: > > https://github.com/jonghough/jlearn > > > > The library is mostly for didactic purposes, self learning etc. but > > It may be of use, or of interest, to anyone experimenting with machine > > learning using J. > > > > Bare in mind, I am absolutely not an expert at machine learning, so there > > could > > be a lot of egregious errors. > > > > I managed to write a convnet which can get very high accuracy with the > > MNIST dataset, which is a pretty simple "image" library. > > Trying with CIFAR-10 dataset, I still max out at around 50%-60% accuracy, > > and am struggling to get any better (struggling to find time to improve, > > since a single epoch through the 50,000 training images takes a couple of > > hours+). > > > > > > Thanks, > > Jon > > ---------------------------------------------------------------------- > > For information about J forums see http://www.jsoftware.com/forums.htm > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm > ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
