I did similar analysis here http://karpathy.ca/myblog/?p=559 http://karpathy.ca/myblog/?p=559 but it's a little less clean presentation. I'm quite excited abou

I did similar analysis here http://karpathy.ca/myblog/?p=559 http://karpathy.ca/myblog/?p=559 but it's a little less clean presentation. I'm quite excited about this data release- there are many interesting ML models that could be trained here. One I hacked on previously on my own data was a comment ranker which uses a ranking loss to rank comments consistent with their observed order in data (which roughly reflects their number of upvotes, I believe). In principle I think it could be converted to a browser extension that gives a score for how well received your comment will be conditioned on the parent comment, as you write it in the text box. One of the main issues I ran into when I hacked a bit on it was space complexity, since you need to keep all the word embeddings (usually on order of 50-200D / word) around in memory of the extension, and there are many words.