Previous: Chess
Up: Examples of Material Analysis
Previous Page: Chess
Next Page: Expected Outcome

Discussion

Despite its simplicity, the analysis produced useful piece values for a wide variety of games, which agree qualitatively with the assessment of experts on some of these games. This illustrates that the general knowledge encoded in METAGAMER's advisors and analysis methods is an appropriate generalisation of game-specific knowledge. This appears to be the first instance of a game-playing program automatically deriving material values based on active analysis when given only the rules of different games. It also appears to be the first instance of a program capable of deriving useful piece values for games unknown to the developer of the program [\protect\citenamePell, 1993b]. The following sections compare METAGAMER to previous work with respect to determination of feature values.

• Expected Outcome
• Automatic Feature Generation
• Evaluation Function Learning
• Hoyle