I guess this is up to v5 now. The only difference of any substance is the new prediction feature. Basically, this uses the OPRs computed so far to predict the remaining matches at a regional, and prints out the simulated rankings afterwards. If all the matches have been played, it does a self-analysis to see how accurate it would have been. Based on this, I can say that it is about 65-70% accurate when predicting after 40-50 matches, though this seems to depend heavily on the regional. It is much better at predicting GTR 2008 (80%) than it is predicting DC 2009 (55-60). Still, it is generally a good bit better than a coinflip.
Command-line is:
Code:
oprnet <regional> <year> predict <sort style> <quiet flag>
So...
oprnet il 2009 predict t q
or
oprnet dc 2009 predict t
or
oprnet oh 2009 predict
or
oprnet on 2008 predict (this has pretty cool results)
The main problem with predicting match results is that robots evolve quite a bit over the course of a regional. Some break, some get fixed, some get improved tactics or mechanical parts. Without actual 'on-the-ground' knowledge of what is going on with individual robots, any purely mathematical model is going to fail. Getting 65-70% of guesses right isn't bad. But still, the 'predict' feature is an entertainment-only kind of feature.