Nolan,
Any idea when the next rating batch will be posted to MSA? I’ve got an event long overdue - partly my fault!
Thanks
We’re hoping to do the first rate under the new programming on Tuesday or Wednesday of this week (ie, Feb 1 or 2.)
The ratings staff is still learning how to enter events using the new programming, so I don’t know how many events that were mailed to the USCF office will be ready to rate by the time we’re ready to do the first run. The last time I ran a count it looked like there were about 90 events submitted online that are ready to be rated.
Events that had been entered but were still pending in the old ratings system after the last rate run using the old software have been transferred over to the new system so the ratings department can continue to work on getting them to pass validation.
While we’re at it, I ran a tournament in six sections, but only four of them were rated in the last run. Should I expect the other two to make the next run?
Alex Relyea
Presumably the other two sections didn’t get rated because of some validation problem, usually missing IDs or non-current memberships.
I think all of the unrated sections have been converted over to the new database so that the ratings department can continue working on getting them to pass validation.
E-mail me the details of your event (mnolan@uschess.org) and I can check on it. That’ll help me check that we’ve got them all, too.
Whether or not they make the first run on the new system is hard to say, without checking them to see what the problems are. Probably not, since we’re hoping to do that in the next day or two. But once we get a few batches under our belts, we’ll switch over to running batches several times an hour, probably with a weekly re-rate to get events into chronological order.
In preparation for rating events, I will be shutting down the tournament entry/edit module in TD/A this evening at 9PM (CST).
It should be back online Tuesday morning.
And now for some good news. As of earlier this morning the new ratings module is computing ratings that match the old ratings system, so that computational obstacle has been overcome.
There are a few more administrative tasks to complete, but it once those are done we’re ready to rate events again.
WOW !!
As a fellow programmer, I understand the huge HIGH you must be experiencing right now – especially considering how complicated the rating system is.
Congratulations, Mike, and GREAT WORK !
Bill Smythe
Actually, by the time I finally found the last of the anomalies, I was so tired of looking at the formula and so sleep-deprived that to celebrate I took a nap!
That’s an even better story. The career marathon runner, after finishing in first place in the best race of his life, is so drained that he collapses into the arms of his supporters at the finish line, unable to celebrate until he regains his strength a few hours later. Congrats again, Mike!
Don’t tell me, let me guess. The last hurdle was that “special formula”, especially with provisionals with prior rated games, and all those nodes and broken-line approximations, right?
Bill Smythe
While the linear programming portion of the special formula was a bit challenging (and IMHO unnecessary as there are several other
ways to converge a monotonic function that are easier to program but not quite as efficient), that wasn’t the real stumbling block.
Iit was figuring out all the places that the previous implementation deviated from the formula or at least my interpretation of the formula.
Another stumbling block was where the different ways the two programs determine the pre-event rating made a section a bad test case. This has to do with some extra work that the new programming goes through in order to be rerunnable and for events to be reratable, plus the USCF’s tendency over the years to make changes to people’s rating without keeping a log of what was changed, when and why.
It is amusing that when I finally figured out the last piece in the puzzle, the members of the ratings committee were not in unanimous agreement on whether the old implementation was correct or the new one is.
As a result, a discussion is taking place within the RC regarding whether the special formula should be invoked in some additional cases in step 3. I’m hoping the RC won’t dawdle with their decision on that.
Yeah, that’s sort of what I thought too. If you just start with a too-low estimate of 100 and a too-high estimate of 3000, and keeping doing binary splits until you find a stable rating, the result should be the same, and the extra CPU time is trivial nowadays.
Well, as soon as a bunch of ratings come out differently than what the RC thinks they should be, their tendency to dawdle will diminish considerably! 
Bill Smythe
If the MSA is being held back because of the rating committee. Could the MSA be held back after the scheduled April 2005 Supplement?
There are still a number of administrative steps that have to be working in order to rate, and one additional data issue that arose during testing of the formula having to do with discontinuities in the crosstable history. This is important because in order for events to be reratable there has to be a way to determine what someone’s pre-event rating was on a consistent basis.
If someone’s previous event has a post-event rating of 1150 and the pre-event rating for their next event is something other than 1150, then that means that something happened in between those two events. This could be any of several things including a deleted event, an event with a changed ID, merging of multiple (duplicate)IDs into a single rating, or an additional event played in between those two dates.
The RC discussions are proceeding well, I suspect I’ll have an answer from them before I’m actually ready to rate any events anyway.
We know you’re working hard on the new program. Its’ just the official statement back in December, that the programing would be up and going the first of the year. With the ‘individual ratings’ no longer updated since December 18, 2004. With the statement the next release of the MSA is close to being ready. With the final run of the old rating program January 12, 2005.
We know any ratings on the MSA from January 12 till now would not be official till April 1, 2005. With the old system, it was updated on average once a week, or 52 times a year. Making each updating being 1/52 of the total tournaments in the given year. With that, the data did take a long time to upload on the MSA, even making the server crash. Since its going to be over a month before the next update, that would be 1/12 of all the tournaments in the given year. If the server was having problems with 1/52 of all the tournaments in a given year, would it have a harder time with the date of 1/12 of all the tournaments?
Under the old ratings system, when we updated the crosstables on MSA we had to delete and reload an entire calendar year’s worth of data at a time because there was no way on the old system to track changed or deleted events, all we knew was that events in 2004 needed to be updated, not which events were affected. This was a flaw in the original MSA design and in the old ratings system, but one that wasn’t a problem until we had 12 years worth of history in place.
Since most of the time a ratings batch had events from at least two calendar years we generally had to process both 2004 and 2003 most of the time. Thus we were updating more like 1/6 of the ENTIRE crosstable history each time.
On the new system we should only have to upload those events that are new plus update those that have been changed or deleted, so in fact the bandwidth to update MSA should go down significantly once we have done an initial reload of all crosstable data under the revised data structure.
It has taken far longer to get the new ratings system up than I had originally estimated back in June and again in October. I think the benefits will be worth the effort and will reward patience on the part of USCF members.