Sorry things have gone silent on Spawning Tool recently. Various personal factors (some mentioned in my personal blog) took precedence over Spawning Tool, but I’m somewhat back now. Although most things tend to slow down around the holiday season, I’ll be trying to ramp up as I find more free time to be working on it. As such, I have a few different updates.
First, props first to TL’s Dakota_Fanning on releasing Scelight 1.0. I haven’t had a chance to try it myself, but it sounds like he basically accumulated too much cruft building sc2gears and wanted to start over. If you’re not using it, you’re seriously missing out.
And another shout-out to Aligulac for releasing the beta of their API. I’m not sure how people are going to use it, but they should be. I think it’s great that they developed this, since I would like to see the StarCraft analytics community grow.
Last shoutout to GraylinKim on GitHub for patching sc2reader so quickly. Maybe a week ago, I tried to upload the Red Bull Battle Grounds replays to Spawning Tool, but they weren’t parsing correctly. Graylin fixed it within a day, and those replays are now parsed and available at http://spawningtool.com/replays/?tag=612
So Spawning Tool continues to float along well, though I admit that it hasn’t gotten as much traction as I would have hoped. Most of that is my fault: I don’t know how to market these sorts of things, and I haven’t dedicated the time to really get it out there. I do, however, know how to code, and I want to continue to work on things for that over the next 2 months or so.
First, I want to be more visual with the data. It isn’t one of my strengths, but it’s a chance to learn. I think that the right graphics can be very compelling, so I’ll try to get some of those put together. As a preview, check out this rough graph of TLO’s wins (green) and losses (red). The Y axis is the length of the game in minutes, and the X axis is the date the game was played. Do you see any patterns?
Second, I’m going to try to push again for the machine learning (ML) side. There are lots of great questions that I think that ML can help with, but I have 2 in mind. First, given a replay, can we guess what build they used? Second, given a snapshot of a game, can we predict who won? The second question is just a matter of me building it and hoping that my approach works. The first question is one of having enough training examples to know what various builds look like, which flows well into my last part.
Third, I want to figure out better ways to engage users in a helpful way. The site has the research tool, but it’s something that you have to figure out: the actions on the site aren’t obvious. One thing that I’m considering is allowing anyone to tag any replay, registered or not. Currently, you must be the uploader of the replay or a superuser to tag a replay, but having done a lot of labeling myself, it’s a lot of work. My concern is anonymous users vandalizing the data, but at our current pace, the data isn’t useful in the first place.
Let me know what your thoughts on the above are, especially the part about opening up tagging. I would love to know more about what you see as the value of Spawning Tool and the best ways to make it better.