First off, I started by digging into some old papers and stuff I could find online. I stumbled upon a few things about first-order and second-order methods. Now, I'm no math whiz, but it seemed like these methods are all about how quickly you can get to an answer, not necessarily how accurate that answer is. That was a bit of a letdown, to be honest.
Then, I found this one paper from way back in '92. It talked about using these optimization methods for neural networks. It sounded fancy, but it was way over my head. I tried to make sense of it, but I ended up more confused than when I started.
After that, I switched gears and looked into some basic probability stuff. Figured, maybe there's a simple way to do this. I read something about there being 120 ways to rank things, and how to calculate the odds of specific outcomes. Like, if you have 'A' in first place and 'B' in second. It felt like I was getting somewhere, but it wasn't quite what I was looking for.
So, I decided to try a different approach. I found this thing called the "probabilistic method." It's this idea that you can prove something exists without actually finding it. Sounds weird, I know, but it seemed worth a shot. I delved into it, but it got really abstract really fast. Lots of talk about first-order and second-moment methods. My brain started to hurt.
But I didn't give up! I kept poking around and found this French article. Yeah, I don't speak French, but I used a translator. It was talking about improving a method by using a parabola instead of a straight line. I think it was related to forecasting, which is kind of what I'm trying to do. It seemed promising, but I couldn't really figure out how to apply it to my problem.
Finally, after all that, I decided to just build something myself. I used a simple machine learning model and fed it a bunch of data from past competitions. It wasn't perfect, but it actually did a pretty good job of predicting the first and second place winners. It's not foolproof, but it's way better than just guessing randomly.
Here's what I learned
- Sometimes, the fanciest methods aren't the best.
- Simple probability can be surprisingly powerful.
- Don't be afraid to just try building something, even if you're not sure it will work.
So, yeah, that's my journey trying to predict first and second place. It's been a rollercoaster, but I've definitely learned a lot along the way. And who knows, maybe my little model will actually win me some money someday!
