![]() Worked on a flask socket.io and eventlet server. In general these images look pretty muddled, but there's a painterly quality to it that I really like. (weekend, last week) Finished Tapestry #2, a 50x51 grid of ImageNet averages sorted by semantic similarity. I will have to do a lot more work to improve the code. The backend code at this point looks terrifying, and there is a display bug where the first word isn't recorded, but it is a significant improvement for the UX :) I also shut down the MySQL database and any queries to it, in preparation for moving to another platform. (today, yesterday) updated word.golf to use XML http requests between prompts, creating a seamless no-reload experience. ![]() () Learned some APL, worked on socketio & eventlet, modernized JS on golf. NORC is moving me from Amerispeak to GSS, which is awesome. Made this kind of scaffold (click to advance), experimenting with d3 as a tool for making slideshows. () Went to some chats with new Recursers, and readying a fun presentation for tomorrow on APL. Yesterday went to Powell's and read a bunch of picture books, and attended long NORC call from pioneer square. I wanted to cross the border but it is not accessible to pedestrians. () Spent the day walking around portland. Reading a book on the topography of seattle, a collection of japanese short stories, man is the measure, the Dawn of Everything, red mars, and a little bit of graph theory. () At home, working intermittently on various things. Today reentering the pairing pool, otherwise revising website. () Finished Red Mars and short stories, made some progress on the rest. Here's a bigger version that probably won't load The way DBSCAN works is that it looks for cluster cores, which at the current settings are two points (words) that are with a distance of 0.3 and with at least 1 neighbor (this is good for isolating extremely close vector pairs, like "map" and "maps"). got some clusters with k-means, then changed to DBSCAN which works surprisingly well, and the sklearn implementation makes it easy to assign words to clusters. () Paired on clusters for tSNE maps of word2vec. Both make sense taken together, it is actually a○n expression of the content/form distinction. The first is "spoon theory", or only having a certain number of spoons in a drawer (representing mental energy), and the second is The Matrix's "there is no spoon", which is a post-scarcity declaration of anything being possible. The second is that there are two major philosophies based on spoons that nicely complement one another. Maybe "sign down (here, below)", but why not just "sign (here, below)"? There are so many associated meanings (sign up, write down, sign in, write up) that you'd expect this phrase to suggest something, and yet it is a barren wasteland. Two delicious coincidences I've noticed recently. () Presented on tSNE language maps and then got some interviews done. ![]() Running this overnight and checking results then. , where there are four meaning-planes, formed by taking three of the four words. This also generalizes to higher dimensions, but with my current script producing four point/word/vectors I haveīelated position circumvented existential Given four equidistant points from a center point, they form a tetrahedron in 3D wordspace where each of the points is at the intersection of three planes of meaning. () I have this idea that solves the multiple meanings problem with language. html files are older works and that I actually enjoyed the previous format a good amount, but I still want to have some visual non-gridified aesthetic. I want to rewrite this website again because I'm not happy with a lot of the content.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |