For fun, I decided Wolfrat Bot needed a bit more, err, personality. So, I set off to make him one. After training an RNN for a few hours on my desktop’s GPU, I had interesting results but ultimately failed at achieving quite what I wanted. Anyway, I encountered some huge hurdles I found out AFTER training for ~8 hours that the way I was encoding the characters for the network input (and thus, the way I was decoding the network output) behaved non-deterministically between Python 3.5 runtimes!
The content of this website is on GitHub here. A nifty feature about GitHub are its Webhooks. I wanted to be able to add content here without having to SSH into my server, cd-ing into the www directory, git pulling, and makeing, as those steps are tedious. So, I set up a GitHub webhook to POST to a PHP page on the server. The basic form of this PHP page is just: <?php passthru("MYSCRIPT.sh"); ?> Where MYSCRIPT.sh is just a script that pulls and makes the site.
For our final project for our Data Structures course, my group decided to make a GPS app that finds the path with the minimum time spent outdoors. We wrote a python program to allow us to map our campus as an undirected graph, with vertices at the intersection of every outdoor path, and at every door of every building. To decide paths, it uses Dijkstra’s Algorithm. To achieve our goal of deciding paths with the least time outdoors, we assign weights of 0 to all edges that connect nodes within buildings, and for each other edge the weight is the linear distance between the nodes.