chris@wiley.UUCP (Chris Stassen) (06/30/87)
(I) Overview
Bob Silverman is basically correct that there are no computer bridge programs
which can provide much challenge for an expert player. Unfortunately, bridge
is a "hidden information" game, which makes it very difficult to write good
bridge-playing programs. In addition, most PCs do not have much in the way
of CPU power, making the "brute force" decision methods impractical. The best
programs can give a novice practice or entertain an expert, but one cannot
expect to learn too much from them. (I have not seen a program that can
perform an endplay, let alone a squeeze; however, I learned to recognize and
set up endplays from reading books and practicing with my PC).
Also, there are some incurable card players (like myself) that have trouble
finding three other players who are in the mood for a few dozen hands at 2AM.
In cases like these, my computer is the best solution, though I readily
acknowledge that it isn't the perfect one.
Included in this review are all bridge-playing programs that I've used for 20
hours or more. Not included are those which only play pre-set deals (*) or
those which are not available for the IBM PC.
(*) For those experts who want a real challenge, but don't mind a limited
set of hands, I suggest "Tournament Bridge". I will kiss the feet of
any player who gets all of the hands right the first time through.
(II) Mini-review
This is a short table of the programs reviewed in this article. The
manufacturer, cost, and an overall rating are given. I used a scale
of 1-10 for ratings, 10 being "the best there is" (some would expect
10 to be reserved for "perfect," but then I couldn't use it).
+-------------------+------+------------------------+------+--------+
| Program | Abbr | Manufacturer | Cost | Rating |
+-------------------+------+------------------------+------+--------+
| Grandslam Bridge | GR | Electronic Arts | $60 | 10 |
| Bridge Parlor | PA | Parlor Software | $75 | 9 |
| Turbo Bridge | TU | Borland |* $35?| 6 |
| Bridge Baron | BB | Great Game Products | $50 | 5 |
| Bridge 4.0 | 40 | Artworx | $25 | 1 |
+-------------------+------+------------------------+------+--------+
* I don't know what Turbo Bridge costs by itself. Source included (see review)
Program = the name of the computer program.
Abbr = abbreviation that will be used to refer to it later.
Manufacturer = the name of the company that distributes the game.
Cost = purchase price retail, last I heard.
Rating = overall 1-10 rating (considering performance more than price).
I used ratings in a number of different categories to arrive at the final
ratings listed above. I did not give all categories equal weight.
Ratings in various categories (5 = acceptable, 10 = best of the five):
+----+-------+-------+----------+-------+------+-------+-------+-------+
|prog|playing|playing|defensive |bidding|bid |needed |extra |copy |
|abbr|speed |quality|signalling|system |robust|options|options|protect|
+----+-------+-------+----------+-------+------+-------+-------+-------+
| GR | 5 | 9 | 10 | 10 | 10 | 10 | 10 | 10 |
| PA | 6 | 10 | 9 | 7 | 10 | 9 | 9 | 5 |
| TU | 2 | 7 | 1 | 5 | 8 | 3 | 3 | 1 |
| BB | 10 | 3 | 3 | 5 | 5 | 2 | 3 | 10 |
| 40 | 7 | 1 | 1 | 2 | 1 | 1 | 1 | 2 |
+----+-------+-------+----------+-------+------+-------+-------+-------+
playing speed = how quickly the program plays.
playing quality = how well the program plays.
defensive signal = how well the program uses defensive signals.
bidding system = how well the bidding system is implemented.
bid robust = how well the bidding reacts to unexpected bids.
needed options = how many of the necessary options are present (see below).
extra options = how much I was impressed by extra frills.
copy protection = how well the program is copy protected (1 = no protection).
(III) Important feature comparison:
There are several features that I consider very important in a bridge playing
program. Quality of play is most important, but I don't want to sacrifice
speed (it must play at least as quickly as a human). Defensive signals are
also important, because it is hard to mount a competent defense without them
(and because I feel that defensive practice without them is not worthwhile).
Some program features, like the ability to "claim," are also important, because
they help to speed up the play a lot. I don't like programs that cheat.
+------+-------------+-----------------------+---------------------------------+
| Prog | Playing | Leading conventions | Important |
| abbr | Speed | Qual| and defensive signals | features |
+------+-------+-----+-----------------------+---------------------------------+
| GR | >= | Hi | 4B TON AS CSS SP | CLV SCR ENT RDS SAV NOC MEN PAR |
| PA | > | Hi | [IS CSA 4B TON] AS FC | CLV SCR ENT RDS NOC MEN PAR |
| TU | << | Med | none (see review) | SCR NOC MEN |
| BB | >> | Low | CSA | CHL MEN |
| 40 | > | 0 | none (see review) | CLU |
+------+-------+-----+-----------------------+---------------------------------+
Speed = Program's playing speed compared to human playing speed.
Quality = How infrequently the computer makes playing mistakes.
Leading = What sorts of leads and signals the computer makes, ones that may be
selected or turned off in brackets [].
4B = fourth best TON = "top of nothing" (3 cards)
CS = count signals in certain suits CSA = count signals always
AS = attitude signals when appropriate ASA = attitude signals always
SP = suit preference signals IS = inverted signals
FC = opponents occasionally falsecard
Feature = Things that the program lets you do.
CLV = human may claim, verified NOC = computer won't cheat (peek)
CLU = human may claim unverified CHL = cheats only on opening leads
SCR = computer keeps score ENT = user can enter deals
RDS = can request distributions SAV = can save hands for later
PAR = partial entry allowed (i.e. don't have to always enter rank & suit)
MEN = menu-driven (don't need to keep manual handy)
(IV) Bidding convention comparison
Almost all of the programs seemed to prefer strong 2-bids to weak, and all
implemented Blackwood, most implemented Stayman. If you want your own weird
and wonderful conventions, you'll have to buy Turbo Gameworks and Turbo Pascal,
and write several hundred lines of PASCAL (or more).
+------+-----------------+-----------------------------------------------------+
| Prog | Bidding system | Bidding conventions |
| abbr | basis (claimed) | Major | 2-bids | Other (see below) |
+------+-----------------+---------+--------+----------------------------------+
| GR | Sheinwold | 4 or 5 | either | BWD GBR STM UNT PRE BDB [GFC] |
| PA | Sheinwold | 4 or 5 | strong | BWD GBR STM PRE BDB [GFC] |
| TU | Goren | 4 only | strong | BWD GBR STM |
| BB | "Std American" | 5 only | strong | BWD GBR STM BDB |
| 40 | Goren | 4 only | strong | BWD-FUBAR (see review) |
+------+-----------------+---------+--------+----------------------------------+
Bid basis = The bidding system from which the computer's system is derived.
Major = Whether the computer opens 4 or 5 card majors, or it is selectable.
2-bids = Whether the computer uses strong or weak 2-bids.
Other = Other bidding conventions used by the computer players, ones that
may be selected or turned off are in brackets [].
UNT = unusual notrump PRE = 3-5 level pre-emptive openings
BWD = blackwood BDB = balancing double
GBR = gerber STM = stayman
GFC = game force cue bid of opponents' suit, implies 3-card support
(V) Where to get the programs:
If you are interested in acquiring one of the programs, I would suggest that
you first check your local computer software stores. Electronic Arts (Grand-
slam bridge) has displays in all computer stores I've seen, as well as most
game stores and even some record stores. If you visit a store, you can test
the program for a while before you buy it -- your preferences may not match
mine.
However, if you do wish to buy and cannot find the product in a local store,
you can try the following:
Grandslam Bridge: Available through Baron Bridge Supplies (502) 895-1354.
Bridge Parlor: Call (716) 865-8412 for free trial version of the program.
Turbo Bridge: Call (800) 556-2283 for nearest outlet of Borland software.
Bridge Baron: Available through Baron Bridge Supplies (502) 895-1354.
Bridge 4.0: You don't want to know (read the review).
(VI) Individual reviews:
(A) [Grandslam Bridge]
Grandslam Bridge, the most recent entry, has obviously been written with the
shortcomings of the other programs in mind. Almost everything may be
configured.
Hands can be entered by the user; hands can be saved to disk. Plays that
don't work may be backed up and re-tried; hands may be replayed. An option
may be set so that the <enter> key is not necessary, and so that not all of
the information need be given for every play (for example, typing '4' is
sufficient to play the only 4 in your hand, or the 4 you must play to follow
suit -- you don't have to enter the suit when you don't need to). [Believe
me, this becomes important after a few hundred hands!]
There are various practice options to test bidding only, test play only,
or to test both. Hands may be automatically replayed double dummy. There
are sound effects, a "boss key", configurable colors (I *like* bright green
spades! :-).
If it weren't for the quality of play, this is the program of my dreams. Until
the state of the art in programming this sort of problem gets better (or I get
a home computer with a lot more horsepower), it's about the best compromise
that I can see.
Its main disadvantage is its speed. This program is probably the slowest
other than Turbo. It still plays more quickly than humans. More bidding
conventions would be nice, but it is still the best that there is. A minor
annoyance is its copy protection. You can move Grandslam to your hard disk,
but must have its "key disk" present to start it up.
The deal configuration is simply awesome. Examples (of single hands, you can
configure all four):
S: ..... \
H: ..... | This configuration always deals hands that are exactly
D: + | 5-5 in the majors (any 5 cards in each major).
C: + /
S: + \ This configuration deals hands with at least 6 diamonds,
H: | a void in hearts, exactly two clubs, and the high cards
D: AKxxxx+ | listed (additional diamond or spade high cards possibly;
C: Ax / x is low card, but + is zero or more cards of any rank).
I've used this feature to test weak spots in my bidding system. (I suggest
experts read Goldman's "Aces Scientific" if they think this isn't of value).
Different numbers of humans may play (from 0 to 4), but it is a bit of a hassle
(all but the player whose turn it is must look away from the screen, or
printouts of the hands must be made for each human player). However, it is
still less of a hassle to use (and bids and plays better) than a "Bridge
Challenger".
(B) [Bridge Parlor]
Bridge Parlor plays about as well as Grandslam, plays a little more quickly,
and has just about as many features. I disliked the configuration process --
I found it difficult to do (the options are all two-character abbreviations,
which are a little hard to remember). I much prefer weak 2-bids to strong,
and Bridge Parlor does not permit them (though there are noises indicating
that the next version -- which owners will receive for free) will.
Bridge Parlor should be of particular interest to tournament bridge players.
When you buy the game, you receive a regular newsletter of Bridge Parlor
tournaments (you enter a code to the program to play a particular tournament).
Each tournament may only be played once, and the score printout may be sent
to the company and best scores are printed in the next newsletter. Also
of interest to the tournament crowd is that Bridge Parlor has several scoring
variations: Chicago, Rubber, IMPS, Duplicate.
The hand distributions may be set in a vague way. A number between 0 and 9
indicates various levels of distribution, from a bias towards 3-3-3-4 to
a bias towards monstrously long suits. There is a similar number to adjust
the high-card distribution. Particular hands may also be entered.
Like Grandslam Bridge, colors and other nice options are configurable. This
program has an amusing option (CC - carnival comments) which causes it to
make occasional rude comments on your playing ability (especially when you
fail to make contracts!).
The program is not copy protected, but tournaments may only be played on the
original disk. (i.e. all copies that are made are limited in their playing
ability -- just as the free trial copy is).
(C) [Turbo Bridge]
Turbo Bridge plays reasonably well, if slowly, and does not make the same
types of mistakes that plague Bridge Baron and Bridge 4.0. However, just
tonight, I made a grand slam against it by leading low from Kx towards my
singleton queen... if you've played against it enough to know how it plays,
you can goad it into making mistakes.
There aren't too many configurable options, and I was rather dismayed at
its lack of a claim. There is a bug in the scorekeeping algorithm, too,
which causes 1 to be added to the score of the vulnerable party in a rubber.
Color changes, etc., are easy to do, provided you have the compiler so that
you can modify the source. It is intentionally (it says so in the manual)
a no-frills program. If you want an option, you can &*$%@ implement it
yourself. If you're an incurable hacker and bridge player, this may be the
toy for you.
I don't know exactly what the cost of Turbo Gameworks is (I bought a
combination package with it and the editor toolkit for $70 about 18 months
ago). The package comes with source and executable for Bridge, Go-Moku, and
Chess. The algorithms used are explained in good detail in the supplied manual
(which runs about 150 pages). If you want to make modifications to the source,
you will also need the Turbo Pascal compiler, which costs about $70.
This is the only program that (to my knowledge) uses a brute-force method
to decide what to play. It randomly shuffles all unknown cards, and then
tests each possible play. The plays that produce the best results (on the
average, over several shuffles) are used. This makes it slooooooow. (You
can make it faster, by limiting the number of random tests, but there is a
pretty significant playing quality loss for the speed increase). This method
is only used (I think) when leading. Without modification, it takes about
35 seconds to make an opening lead, less time thereafter to lead to any trick,
getting faster and faster as the size of the hand decreases.
A second disadvantage is that the program moves faster when its choices are
limited. If you lead an ace, and count how long it takes RHO and LHO to make
their plays, you can just about figure out the distribution. If you lead
a lower honor, an opponent that cannot cover will decide what to play in less
than half of the time that it takes an opponent who has to consider covering
the honor (i.e. the program "hesitates").
Since source code is provided, the program is obviously not copy protected.
(D) [Bridge Baron]
When I'm in a mood to play a mindless game (like hearts or euchre), I often
whip in the Baron disk. I can play as quickly as I can type, and it always
makes its play immediately.
It is a little frustrating that you always have to type in the card rank
and suit (and <enter>), except when you have only one card to follow suit
with (in which case it is automatically played for you). This may sound
trivial, but imagine doing without it for several hundred hands...
There are no options to configure deals. Hands are always played in sequence
number (sequence number is the random number generator's seed, I bet), and
you can't enter hands that interest you. You also must write down the last
hand that you played (its seed number) if you don't want to play the same set
of hands every time you sit down to play.
There is an option to peek at the opponents' cards. This (and the escape
key to end playing a hand early) are the only options that I've used. I
don't like the fact that it doesn't keep score, and that there's no claim.
It appears to cheat on the opening lead, because it always leads the longest
and strongest suit of *both* defenders when making an opening lead. It does
not appear to cheat when playing.
The level of play is not bad (especially considering the speed), but it tends
to give away tricks when long suits are run. The fact that it doesn't signal
anything but count on defense makes it a little difficult to defend.
This program was my mainstay until I found Bridge Parlor. I've played the
first 1,800 hands. Playing against it, though, has become more a matter of
fooling the predictable program into making mistakes than playing bridge.
Bridge Baron is copy protected.
(E) [Bridge 4.0]
Nobody should ever buy this program, and I will cite and document several
reasons. The first is level of complexity. While most of the other programs
are tens of thousands of lines of C or assembly code, Bridge 4.0 is about
a thousand lines of BASIC -- and runs in interpreted mode. Therefore, it
is slow, and clearly there is not enough room for a lot of the features that
make a reasonable bridge program.
The bidding is horrendous. First of all, it doesn't take previous bidding
into account. It just adds up the partnership point total (i.e. cheats)
and bids until an appropriate level of the best fit (cheats!) is reached.
When it decides it has enough to overcall (11 points & 5-card suit), it will
always overcall. I once opened a powerhouse hand 6NT (what the hell, there's
no information from my partner's bidding, anyway, and it would bid 7NT if we
could make it), and LHO (holding the remaining high cards) overcalled 7 hearts!
(When I open the same hand 7NT, 4.0 crashes trying to bid 8 of a nonexistent
suit).
The bidding can get pretty confusing as, when the final level of the contract
is being approached, 4.0 will simply bid whatever suit the two hands have the
most of. I once kept a seven-card minor suit unbid, and watched my 4.0 partner
bid that suit twice, while holding only two cards in it.
The *one* bidding convention that is implemented for 4.0 is not implemented
correctly. Any bid of 4NT is *always* blackwood. With 4 aces, 4.0 responds
6NT! 5NT asks for kings, and with all four, 4.0 responds 7NT! *Get this*,
6NT asks for queens! (No, I'm not kidding!) With all four queens, 4.0 will
crash trying to bid 8NT!
Its play is also horrendous, in spite of absolute cheating. I've had (on the
second trick, mind you) 4.0 finesse the 6 (from AQ986) behind my KJT73!
Even this advantage does not help the program. When I'm declarer, I just
run a long suit (even trump) and it is guaranteed to unguard one of the
side suits (running the side suit causes it to drop another suit...).
I once made 3NT on 12 points between the two hands (and only a five-card
running suit, and no finesses). Just like Bridge Baron, Bridge 4.0 has
become for me a game of trying to get the predictable algorithm to make
mistakes, not a game of bridge.
The user interface is not too great, either. The color scheme you get is
blue symbols on a green background (eyestrain-O-vision), and you can't
configure it. (The BASIC program is written in protected mode so that you
can't read or modify it -- though we bypassed that just to see what the code
looked like).
You can't configure anything. The bidding is worthless in spite of its
cheating. Its level of play fluctuates between plays that it has no right
to make (based on knowing all of the cards), to stupid give-aways that the
rankest novice wouldn't commit.
Bridge 4.0 is not copy protected, and the "protected" BASIC is trivial
to bypass (hint: run BASICA from the debugger, load the file and exit).
(VII) About the author:
I'm not an expert on bridge; I've got only 25 points (7 red). But, when I
play in a field that's 75% life masters (like Bridge Week open games), I
usually place in the top 1/3. I just don't play duplicate very often.
I've owned all of the computer programs reviewed above for several months
(Grandslam Bridge was my most recent acquisition, which I received for
Christmas). I play computer bridge a few times a week, averaging about
two hours per session. I play "Chicago" bridge daily at lunch, plus for
several hours a couple of evenings a week.
I was going to write a bridge-playing program, so I did a lot of research
on AI papers on the subject and the current programs on the market. When
I bought Grandslam bridge, I decided that I didn't need to write my own
program, because it had all of the features that I wanted.
I'm considering seeing if the ACBL Bulletin would have any interest in a
greatly revamped version of this posting; please give me your comments if
you are familiar with any of the programs I've reviewed, or if your priorities
for features are different than mine.
-- Chris
--
"It is good to have an open mind, but | Chris Stassen
not so open that your brains fall out." | {cit-vax,trwrb}!wiley!chrisjking@jplpub1.jpl.nasa.gov (Jennifer King) (07/01/87)
Great review Chris! The only thing that you could add is if the games run using EGA (Enhanced Graphics Adapter). I played Grand Slam Bridge and had the same comments about it as you did. It won't run on my PC Limited AT compatible though, so you might mention that too. EA, though, is moving toward eliminating copy protection on all their IBM software, so perhaps a new version will be uncopy protected. Jennifer King (Lorini) Jet Propulsion Laboratory jking@jplpub1.jpl.nasa.gov elroy!jplpub1!jking@csvax.caltech.edu CIS: 72466,3413 VOICE: (818)354-8617 "Great leaders are rare, so I'm following myself."