|
|
|
Special Features Bridge Computers Play World Champion Northamptonshire company Thinking Games writes bridge software for computers. Their program Oxford Bridge recently played bridge champion Zia Mahmood, in an event held at the Robson Bridge Club in London. Out of seven programs participating, Oxford Bridge came joint top, sharing a prize of $2,000. While no program beat the world champion, Oxford Bridge lead the field jointly with Zia for several rounds. Andrew Bracher, author of the program, says: "Bridge computers are still a good few years behind their chess counterparts. Programming bridge is a much harder task than chess, stemming from the fact that you can only see about half the "pieces" (i.e. cards). This leads to very many more possibilities to consider, and thus to a more heuristic rather than pure computational approach." The Author Oxford Bridge is written by Andrew Bracher, and distributed through a partnership with his wife Rachel called Thinking Games. Andrew was a student at Worcester College, Oxford between 1977 and 1984, where he read Chemistry, and then continued with a doctorate in Chemical Pharmacology. He and his wife moved to the Northampton area in 1994, after coming back from expatriate assignment with Shell Research in Amsterdam. Andrew writes: "My involvement in computers grew out of my doctoral research. I was using computer graphics and computational chemistry techniques to study the way drug molecules interacted with proteins, and got involved with programming. Also while at college I took up bridge, and played in the University ‘C’ and occasionally ‘B’ teams. On leaving Oxford, I took my first job with a small company doing Artificial Intelligence research. In 1986, I joined Shell Research in Sittingbourne, Kent, supporting their molecular modelling and computational chemistry systems. In 1991, Shell moved me to the Netherlands on a three-year expatriate assignment, where I worked on a variety of projects involving image processing and laboratory automation. Shell gradually closed all its research facilities from around 1994 (they are now proud to do only "development"), and so I left to develop Oxford Bridge full time. I now do a variety of contract programming jobs while continuing to keep Oxford Bridge up to date. The Program Oxford Bridge grew out of programs I wrote at University. Using a VIC 20 computer with 20 Kb memory, I wrote programs to deal cards, calculate distribution probabilities and practice bidding using a completely natural bidding system. Also, I wrote a simple play program that used hard coded rules of the type ‘second player plays low’ to see how far such rules could mimic reasonable play. These early attempts were interesting in that they highlighted the things that programs do badly, yet humans intuitively do well. After leaving University, I moved the software to a CBM 64 computer. With more memory and speed, I was able to bring the program fragments together and make a complete system. ‘The Oxford Bridge Program’, was thus born. I was encouraged by family and friends to sell this into the home market, and so together with my wife, trading under the name Thinking Games, started selling Oxford Bridge in late 1986. The program was in CBM cassette tape format, and we advertised through computer magazines aimed at CBM users. A few dozen were sold, not enough to cover costs, but the experience and feedback concentrated our minds on the developments that would be necessary to make and sell a really useful bridge product. I worked on the program in my spare time through 1987. The approaches to the bidding and play were really the wrong way around. The bidding was algorithmic, yet a conventional bidding system needed rules to define the non-natural bids. The play was rule based, yet accurate play required calculation. I therefore rewrote the bidding to be based on rules - still hard coded at this stage - and implemented a simple Acol system. I added new play algorithms to calculate derived information from the hand, and used this to formulate better rules for the play. The result, Oxford Bridge 2, came out in late 1987, now available on floppy disk. With the rise and rise of the IBM PC, and the falling away of almost everything else, the target market now had to be DOS. The advent of Amstrad’s cheap PC clones in particular, made the IBM format available to the home market. Also, the program was now exceeding the 48 Kb of user memory available on the CBM 64. I transferred the program to a PC, and a DOS version of Oxford Bridge 2 came out late in 1988. We switched our advertising to the EBU magazine and other bridge press. This proved successful, and we started to turn a profit. During 1989, I rewrote the bidding program once again, so that the rules for the bidding system were separated from the program. This now meant that users could define their own rules, or modify the supplied rules, and different bidding systems could be developed. Play enhancements also continued throughout the year. In August 1989, we took part in the first Computer Olympiad in London, organised by David Levy. Four programs took part in the Bridge section, and we managed third place, just beating Bridge Baron. Much was learnt from the Olympiad, and a number of corrections were made to the program. The result, Oxford Bridge 3, was launched in October 1989. In August 1990 we took part in the second Computer Olympiad. Only two programs competed this time, and the Bridge Baron got his revenge by beating us into second place. However, development continued and by late 1991 Oxford Bridge 3.4 was looking good. This version was our first real commercial success, selling about three hundred copies over the next two years. Development continued through 1992-3, and Oxford Bridge 4 was launched - still running under DOS - in October 1993. A number of significant enhancements were added. The ability to turn conventions on and off enabled users to select from a list of options, such as weak / strong NT, weak / strong twos, and so on. Standard American and 5-card majors bidding systems were added. There were significant rewrites to much of the declarer play and defence. Increased computer speed meant that more computation was possible, and programming shortcuts previously added to make early machines play quickly enough could be removed. By this time, Windows 3.1 had been released, and was gaining in popularity. This gave the opportunity of developing a graphical user interface, and also developing the program beyond the 640 Kb limit of DOS. During 1994, the software was ported to Visual Basic, and the graphical user interface developed. The first release of the Windows version - Oxford Bridge 4.2 - came out later that year. Oxford Bridge for Windows was further developed during 1995, and a better user interface was included in version 4.4. Support for Windows 95 was added, and version 4.5 followed in April 1996. The increased popularity of home PCs, and Windows in particular, meant that sales had by now increased to several thousand units. During 1997-8, the underlying bidding and play code was translated into the ‘C’ language to increase speed, and so allow for further complexity in play computations. The internal structures used to describe bidding rules were made more elaborate, allowing an improved Acol system to be produced, and also to allow one bidding system to play another. Several other features were added, including teams play. Version 5.0 was released in October 1997, and version 5.1 - the version that played Zia - in 1998. The Zia Match Oxford Bridge performed well in the Zia match. Although rated as an outsider by some (dare I say US?) commentators, I always thought that we had a good chance. Our software is well developed and stable, and the play engine in particular is quite accurate. Oxford Bridge was joint second (with GIB) after the first session, and actually led with Zia at one stage. In the end we fell away slightly, and finished joint second with three of the other programs. There was never any real doubt from our side that Zia would win. There are still many subtleties in Bridge that software doesn’t even address yet, let alone do well. Our main interest was to see how well and how differently the favourite ‘GIB’ would do. GIB is supposed to be very different, and use efficient calculation to play much better than ‘rule based’ programs. However, our view was that the approach adopted by GIB - generation of many possible double dummy hands and their solution - is fatally flawed (sorry Matt!). It just doesn’t deal with the ‘discovery’ element, which is so significant in bridge. In the end GIB failed to show, so I guess the rule-based programs still have the edge, for now. Future Development I am currently working on the 32-bit version of Oxford Bridge - Oxford Bridge 6 - due for release in the year 2000. This will be faster - allowing for further improved play with a larger searching element - and it will have several new features and improved graphics. Improvements to the bidding system editor and ‘post mortem’ facility will make bidding system definition easier, the intention being to get the program playing your bidding system quicker, with no ‘misunderstandings’ ". Contact Information You can contact Thinking Games for sales or product information, by:
|
|
Designed by Web Design Factory E-mail: webmaster@northampton.org.uk Hosted by Internetters Internetters Limited is an ICANN Accredited Registrar |