Monthly Archives: August 2016

Hallucinating words

Posted on August 26, 2016 by underwhelmingforce

After a few failed experiments (namely, trying to train char-rnn on the grant award database to discover that the file has been corrupted with non-unicode characters which caused loss to explode and training on the list is kernel source, which … Continue reading →

Posted in Uncategorized | Leave a comment

The adventures of Robo-Chef

Posted on August 16, 2016 by underwhelmingforce

Since my previous post, I’ve gotten the model better.  How much better? About 22%! That’s pretty exciting. Here, exclusive, and brand new, are some highlights and some dumps from this 404-star recipe bot. At temperauture=0.1, we get the least creative, … Continue reading →

Posted in char-rnn | Tagged | Leave a comment

Char-rnn retro-post

Posted on August 16, 2016 by underwhelmingforce

As with my previous post, this post is intended to serve as more permanent storage for my previous posts regarding the topic at hand– in this case, Andrej Karpathy’s infamous Char-RNN. First introduced to the world in this incredible post … Continue reading →

Posted in char-rnn | Tagged , , , , | Leave a comment

Neural-style retro-post

Posted on August 15, 2016 by underwhelmingforce

Seeing as I’m starting to get more serious about machine learning, I’ve created this website to document my adventures. Most of my previous posts have been in rather impermanent media, so my first few posts will simply be playing catchup … Continue reading →

Posted in neural style | 2 Comments