home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!elroy.jpl.nasa.gov!ames!agate!triangle.Berkeley.EDU!grove
- From: grove@triangle.Berkeley.EDU (Eddie Grove)
- Newsgroups: rec.games.bridge
- Subject: Re: Which inference is better, WAS - "finesse or play for the drop"
- Date: 20 Nov 1992 21:01:08 GMT
- Organization: University of California, Berkeley
- Lines: 36
- Message-ID: <1ejjmkINN66i@agate.berkeley.edu>
- References: <BxwHCq.55K@irvine.com> <1ee8ukINN8ij@agate.berkeley.edu> <By11xy.GE6@irvine.com>
- NNTP-Posting-Host: triangle.berkeley.edu
-
- In article <By11xy.GE6@irvine.com> adam@irvine.com (Adam Beneschan) writes:
- >Didn't some researcher just determine recently that if you shuffle the
- >cards 7 times (I think), that the result will be close to a perfect
- >random shuffle?
-
- This is several years old. The theorem I remember is that 7 riffle
- shuffles produce a distribution whose Total Variation Distance from
- random is at most 1/2. The T.V.D. between two distributions is the sum
- of the absolute values of the differences in the probabilities of the
- points of the space (each possible permutation, in our case).
-
- Now, a single shuffle is not as good as a riffle. I, for example,
- tend to alternate cards from between the halves more than
- a true riffle would do. It is hard to say (at least for me) how
- much of a difference this makes.
-
- Secondly, the statistics on hand-dealt shuffles that motivated this
- work were consistent with a T.V.D. of less than .5. What I mean by
- this is that if you just look at the distributions of suits, and
- assume that for a given distribution the cards are random, then
- hand-shuffled cards would have been observed to have a T.V.D. small
- enough to compare to the result of the theorem.
-
- The result is that 7 shuffles suffice, but maybe they don't.
- However, I believe it is true that 6 truly random riffle shuffles
- DO NOT suffice. Of course in practice, for the most part, the
- starting configuration is sufficiently unknown to make up for
- a few shuffles. But if you start with a new deck, maybe
- you should shuffle 8 or more times.
-
- The actual shuffle in the paper is to take the deck, and for each
- card move it to the left or right half with prob 1/2, and then put
- one half on top of the other. It is actually the inverse of the
- shuffle we all think about.
-
- Eddie Grove
-