[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

Ken's proposal

Posted By: Henrik Bukkjaer
Date: Tuesday, 17 December 2013, at 10:41 p.m.

In Response To: Ken's proposal (Rod)

While PR is a very good measure, it's not at all the golden solution to all rankings in backgammon.

There are many factors and reasons apart from PR, as to why some players succeed more than others, and what should be considered.

But simply the fact that the currently available PR is not "correct" makes a difference.

Let me give you some examples or ask some questions for consideration:

Playing against a very very strong player. If you were able to play the best possible moves in all situations (the absolute best, not the best according to the current bot-flavour of the year). What PR would you imagine on the WorldClass setting that performance would get?

If you play against a strong player (a guy with a PR avg. of say 5), what would optimal play then be? Take it even further, what would the optimal play be rated as PR-wise against a "10" player?

Another thing to consider, and this is a key thing that people either forgets or simply is not aware of: An error is not just an error, and the normalized error-size is not at all corresponding to the impact on you win chances. And I'm not even thinking about cube-levels here, it's much more skewed than that, really affecting some player types more than others:

If you play at a reasonably strong level, say you're what we call a "5 player", and your opponent is a "stock player" worthy of an average of say 7.

Now, if you start to induce just a small amount of small errors, doubling a bit too early, and taking a bit too much. Just a tad bit. then you'll see that: 1) Your PR average will DROP as a result of those errors, and 2) Your win rate will also DROP!

On the other hand, if you systematically starts to induce just a few very small errors doubling a bit too late on occasions, and passing a bit too conservative on some borderline cubes. Then guess what: 1) Your PR will rise, but your win rate will also rise!

This is contradicting what you would expect. And the real problem here is, that it will clearly follow the patterns of certain player types, thus making them consistently look good/bad PR-wise in relation to their win rates. An "aggressive" player really gets rewarded by the bot, but often finds that he's not as successful in the "bread-and-butter" games as he should be. He often fares well against better players though.

So, in my opinion, you should always consider both PR input AND results when doing your thinking for the GIANTs ballot. There probably are no better way to "consider" results, than taking a look at a good working ELO ranking system, like the one DBgF produces or similar data sources that systematically covers a lot of matches.

Cheers,
Henrik

PS: In the last part above, I'm not at all thinking about "adjusting" your checkerplays or anything like that. Just that fact that evenly sized cube-errors from optimal play, will yield a positive effect on your PR in one direction, and a negative PR effect in the other direction! And it's not just an idea that I have, I actually simulated this to verify that my feelings were correct.

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.