Tuesday 11 September 2018

You're only as good as your last game

In a slow day at work today, I was thinking about how I calculate winning chances and whether I could come up with a better methodology. Right now, I'm using the simple method of seeing how often players win legs within four, five, six and seven or more visits and then working out the probability that they win a leg on their own throw from totals of those numbers. This seems to be working well enough, but it's running on a rather crude date slicer - I can pick any date range in my database, and then have it use only those legs won and lost during those dates.

Then again, do we want a leg someone has thrown in the UK Open qualifiers that we had at the start of the year to have the same significance as those that were thrown in Maastricht last weekend? I kind of think that we shouldn't - having a larger sample size to run projections is almost always beneficial, but players do get better and worse over the course of a year, and if someone's throwing like crap right now it's something that we probably ought to take into consideration.

What we can do is amend the way that I sum up the legs, and have it work on a sliding scale depending on recency - perhaps have any legs thrown within the last 30 days counting in full, those from 31-60 days earlier counting at 75%, those between 61-90 days earlier counting 50% and anything older counting just 25%. That way, you're still getting the benefit of a larger sample, but it weights recent games more heavily, which is kind of what we want.

The issue comes with something like we had in the Euro Tour that van Gerwen last won - prior to that, there was the Matchplay and nothing else in terms of competitive darts for a couple of months, so you're in danger at certain times of the year of reducing the section of your sample to more or less zero. That said, if you're just dividing all the figures for someone by four, then you're just getting the same methodology that you're using right now, so it really shouldn't make a difference, and if someone has had a good showing in the one recent tournament that we've had, then it ought to influence our decision making somewhat.

I may play with some coding at the weekend when I'm not making bets for the Riesa event. Is it really already a year ago since I was there?

No comments:

Post a Comment