[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-gnubg] Luck rates ?

From: Massimiliano Maini
Subject: Re: [Bug-gnubg] Luck rates ?
Date: Tue, 13 Nov 2007 09:38:37 +0100

> I agree that the luck is probably better represented this way for a
> single match length or a money game session. But the same argument
> could be made for the error rate, i.e. should your errors be measured
> in relation to the amount of match equity that you loose rather than
> the amount of EMG. The reason that it is done this way for error rates
> is obviously to be able to keep just on statistic independent of match
> length.

Not only: in fact you want to be able to compare mistakes independently
of the stake (cube value for money and match score for matches).

Hence the notion of normalization:
- unnormalized equities/MWC : depends on the cube value/match score
- normalized equities/EMG : do not depend on the cube value/match score
(well this last is not really realy true, but ...)

Unnormalized equity and MWC tends both to emphasize errors committed
when the stakes were high (typically, high cube money games and last
games in matches). Of course, this is not really interesting from a
learning point of view, since something that would cost you 0.1% in
the first game of a match (and which would be considered a no-error)
could cost you 5% in a last one (a giant blunder). In order to compare
errors, in money games equities are normalized (i.e. the cube value
is considered 1) and in matches we use EMG.

> To me the error and luck are very similar, i.e. it doesn't
> matter if you loose 0.2 point by bad luck or by making a major
> blunder.

> That is the explanation, I believe, for the layout being as it is. But
> I agree that the labels could be clearer, for example
> Total luck - ME (EMG) == Luck rate (total)
> Luck rate - ME (EMG) == Luck rate (per move)
> Total error - ME (EMG) == Error rate (total)
> Error rate - ME (EMG) == Error rate (per move)
> (where ME == ME or points. The term Luck rate(total) is certainly
> meaningless.)
> with these clearer labels I believe that we can interchange the the
> values as you and Ian wish.

This is much better. My only additional remarks would be:

- use the terms equity (E) and normalized equity (NE), both make sense
for both money and match and for both luck and error. I know EMG is more
or less standard, but using EMG in a money game is somewhat ... ugly.
For match, the equity is of course just the MWC%/100, while the normalized
equitity is the EMG.

- I would use E (NE) for luck, but NE (E) for error:
(1)        Total luck  - E (NE)
       Luck rate   - E (NE)

        Total error - NE (E)
       Error rate  - NE (E)

This because the thing the user would look at at the end of a match or
session would be: the normalized error (total or rate) and the unnormalized
luck (total most of the time). These values should appear in the "main"
column (i.e. not in brackets).

However, I do agree on your point: luck and error are both apples and can
be compared (normalized ones with normalized ones and unnormalized ones
with unnormalized ones). The above format (1) makes the comparison less
clear, hence the following one seems to me reasonable too:
(2)        Total luck  - NE (E)
       Luck rate   - NE (E)

        Total error - NE (E)
       Error rate  - NE (E)

I still have a small preference for my (1) one, but I won't complain
if you decide the (2) one is better.


Is "unnormalized" good english ?

reply via email to

[Prev in Thread] Current Thread [Next in Thread]