Robert M. Hyatt

Using time wisely, revisited (extended abstract)

An earlier paper [l] described an algorithm for managing the CPU time used by a chess program. Simple linear allocations of time are unsatisfactory, since no heed is taken of the complexity of the task at hand. Also the linear model is not like the approach taken by humans, who devote much time in the early phases to possible themes and long range plans that may enable a simple and clear cut goal for the later stages when time may be short. The previous paper described the effect of using Factor*Target-Time as the time per move where Target-Time is the average time per move, and Factor = 2 Nmoves/lO. Here Nmoves is how many moves have passed which involved significant CPU usage (limited to a maximum value of 10). This produces a time allocation scheme along the lines of that shown in Figure 1. In this presentation we discuss the time affect of using different search partitions, referred to as narrow window aspiration search. For optimum performance, a chess program must search within the area of the tree where the true value of the position will be found, while avoiding the remaining areas of the tree whenever possible. Problems arise when the aspiration search “fails low”, indicating the loss of at least a pawn. In these positions major losses may also be possible, but the search time can be reduced significantly if the search window can be correctly positioned around the minimal loss. The new allocation scheme is based on the argument that the amount of additional time spent trying to avoid the loss should be proportional to the value of the material being lost (V). When using a window of (-m, -V) there may not be enough time to traverse the whole t.ree, and so the correct lo33 cannot be determined. To protect against this, one of five aspiration windows is used for the researches. If an initial search with (-Pawn, Pawn) produces a estimated value of V, the set of available windows becomes (-m, V-Queen), (V-Queen, V-Pawn) (V-Pawn, V+Pawn), (V+Pawn,V+Queen) and (V+Queen, co). The primary advantage of this scheme is that one always seeks an answer within the narrowest plausible window. In many cases it is not necessary to use (-a, V-Queen), since losing a queen is tantamount to losing the game. It is most important to spend what time is available on portions of the game tree that are relevant. Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given that copying is by permission of the Association for Computing Machinery. To copy otherwise, or to republish, requires a fee and/or specific permission. 1985 ACM 0-89791-170-g/85/1000-0271

A Solution to Short PVS Caused by Exact Hash Matches

00.75 Harry L. Nelson Lawrence Livermore National Laboratory Livermore, 94550 The five level search partitions described above eliminated the search time difficulties we encountered, because the lower search bound was not reduced to the point where the mate scores failed to cause alpha cutoffs. The program did not have t.o search the portion of the tree where scores were indicating checkmate, and so could spent more time analyzing move3 that were in the proper ‘band’ of the search tree. This algorithm also assisted the program in allocating time for the search, since it rather quickly determined that a pawn was lost, that more than a pawn was being lost, or that the program was getting mated. Having a quick estimate of the expected material 1035 was critical to the proper performance of the timing control algorithm originally proposed, since it may not be possible to obtain the actual value of the material being lost within the timing constraint present. Reference [l] Hyatt, R.M., “Using Time Wisely”, ICCA Journal, Vol. 7, No. 1, March, 1984.

The design and implementation of a digitally based remote cardiac monitoring system

The problem of short principal variations (PVs) due to exact transposition / refutation (ttable) matches has been discussed frequently over the past 50 plus years. Richard Greenblatt introduced the idea of a transposition/refutation table in his 1967 article on Mac Hack, and the topic has since been discussed many times (CCC11, Greenblatt et al., 1967, Slate, 1977). The issue of exact score hash matches creates several problems that have led to draconian solutions that are generally worse than the original problem. For example, the most popular solution is not allowing exact hash matches on PV nodes, which does avoid the short PV problem, but at a computationally expensive cost. We have already reached this position once, and completed a full search. We now have reached it again, but we choose to repeat the search to extract the remainder of the PV, rather than using the ttable information to avoid this search completely. The author describes a simple and effective solution that all but eliminates this problem with no measurable computational cost or loss of efficiency.

Book Learning - a Methodology to Tune an Opening Book Automatically

This paper gives an overview of the remote cardiac monitoring system developed at the University of Southern Mississippi in conjunction with NASA through the National Space Technology Laboratory. An overview of the proposed system along with details of the completed parts of the project will be discussed. The project encompasses such areas as satellite telemetry, real-time data collection and transmission, monitoring from a moving vehicle and the human interface with the system. Future goals including a medical data network are proposed.