码迷,mamicode.com
首页 > 其他好文 > 详细

CSE4705 - Introduction to Artitcial Intelligence

时间:2019-11-27 19:22:04      阅读:446      评论:0      收藏:0      [点我收藏+]

标签:rem   database   toc   therefore   ace   sch   学生   Dimension   sequence   


CSE4705 - Introduction to Arti?cial Intelligence November 3, 2019
Project - Konane ¯
1 Final Project Overview
?e ?nal project for CSE 4705 is to develop a player for the Hawaiian game of Konane which learns from its ¯
experience. At the end of the semester, you will join in a competition with other Konane players developed ¯
by your classmates.
While the overall goal of this project is to develop as e?ective a Konane player as possible (i.e., one that ¯
will perform well in the tournament), the speci?c goal is to develop a player that learns from its experience.
To this end, your player will choose its moves based solely on a search of the game tree and evaluation
of game states. Speci?cally, the encoding of openings, endings and other move sequences is expressly
prohibited. More information about what is allowed is provided below
?ese projects will be cone in teams of three to four students. Teams of other sizes must be approved
by the instructor. ?e expectations are the same for all teams, regardless of team size.
Konane Game Overview ¯
Konane is a native Hawaiian game played in preliterate Hawaii. K ¯ onane was popular among all classes ¯
and genders unlike some other games that were taboo to one class or gender. A game sometimes lasted an
entire day and o?en matches consisted of a large number of games.
?e Game Board
Konane is played on a rectangular game board divided into a grid of squares or indentations for game ¯
pieces. Boards of di?erent sizes and dimensions existed. ?e number of rows ranged from 8 to 13, and
the number of columns from 8 to 20. Each position on the game board had a hole or indentation for the
game pieces. ?e center of the board was called piko (navel). ?e row along the borders of the board
was termed kaka’i . Before starting play, the board positions were ?lled with alternating black and white
stones. Local beaches provided basalt and coral pebbles for game pieces, whose preferred size was under
an inch in diameter and slightly ?a?ened rather than spherical.
Your version of Konane will be played on an ¯ 18 × 18 board. ?e board rows are numbered from 1 to
18 and columns denoted by the le?ers a through r.
Game Play
Initial Player Moves
Each game begins with a full game board as shown in Figure 1. One of the two players is selected uniformly
at random to move ?rst. ?e selected player removes one game piece from the board. ?e player may select
either one of the four corner pieces or one of the four center pieces on the board as shown in Figure 1.
a b c d e f g h i j k l m n o p q r
Figure 1: ?e starting Konane board. ¯
With this selection the ?rst player to move selects the color of the pieces they can move for the remainder
of the game. ?e other player’s ?rst move is to select one of the pieces horizontally or vertically adjacent to
代写CSE4705留学生作业、代做Intelligence作业、Python,c/c++,Java编程作业代写
the piece removed by player 1, to remove from the board. ?e piece selected must be of the alternate color
from player 1 and will be the color of the pieces moved by player 2 for the remainder of the game.
Subsequent Player Moves
A?er the initial moves, Player 1 and Player 2 will alternate moves for the remainder of the game. Each
player, in turn, will select one of their pieces to move. A move consists of jumping, and removing from
play, a piece of the opponent player. Moves may only be made in the horizontal or vertical direction. ?at
is, no diagonal moves (jumps) are allowed in Konane. A player’s piece which jumps an opponent’s piece ¯
must land in the space immediately beyond the opponent piece. ?is space must, of course, be vacant in
order to be able to execute the jump. If a player’s piece is able to jump another piece of their opponent
in the same direction, then the player may jump the additional piece of their opponent at their discretion.
?ere is no penalty for not executing multiple jumps even if the opportunity exists. Any and all opponent
pieces which are “jumped” are captured and removed from the game board.
Endgame and Winning
?e ?rst player who is not able to make a move during their appointed turn loses the game. ?e number
of captured pieces is irrelevant. Players may be unable to move due to either all of their pieces being
a b c d e f g h i j k l m n o p q r
Figure 2: Selecting a piece for removal from the Konane board. ¯
captured or having a board con?guration which does not o?er them any moves from which to choose. ?e
last player to move is deemed the winner.
Client Players and Competition
Language and Environment
You are free to use any programming language (or languages) you choose to develop your player. However,
it must be able to compete as a client using the Artemis Konane server. See the document “?e Artemis ¯
Server Protocol” for more information on the server. A player who a?empts to make an illegal move loses
immediately and without warning. We will play using a game clock. If your player exhausts its allo?ed
time, your player loses the game.
You may choose to represent the game state however you like. But, you must communicate with the
server using its representation.
Wherever you care running your client, it will need to be available at the ?nal competition, which will
be in the location selected by the University for your ?nal examination. Make sure it is feasible to run your
client player with the server well before the competition.
Project Debrief
Each team will meet with the instructor (scheduled during class time as much as possible) during the ?nal
week of the semester. Each debrief will last approximately ten minutes and will involve discussion of your
3
program design, how your program updates its evaluation function, and how you trained it.
1.1 ?e Tournament
?e tournament will be held during ?nal exams. We will have a round-robin tournament (where each
client plays each other client), followed by a two-game ?nal round for the two top programs (best of three
wins including the earlier match). For this to be feasible, we will allow three minutes per player in each
match.
Documentation and Deliverables
Each team will submit a report that includes appropriate design documentation for your client, a description
of the factors used by your evaluation function and a description of how your program was trained
and evaluated. It should also include documented code for all of your programs. ?is report should document
how your program progressed over time. It should show how the evaluation function changed as a
function of the games it played.
What is Allowed?
Your program will choose moves based on limited lookahead, using a board evaluation function at the
leaves. Your board evaluations should be a function of observable board features that estimates the Minimax
value of the state. Samuel used a set of 25 board features for his Checkers player plus 12 “binary
connective” terms, as his parameters. ?ese may serve as a guide for parameters in your evaluation function.
Starting from an evaluation function with arbitrary coe?cients, your client should learn a be?er evaluation
function by playing games. We will look at a number of possible techniques for doing so (Chapter
21 in particular; Samuel also provides information on what he did for Checkers). However, feel free to
apply other techniques.
Bo?om line: Your program can learn to evaluate boards by playing with any other players, but it must
always make its own moves. So, Samuels method of playing book games is not allowed and the use of
opening or endgame databases is prohibited.
1.1.1 Endgame Variations
Experience shows that endgames are di?cult to do with the same evaluation function that was used in
earlier game play. ?ere are fewer pieces, hence fewer features. Also, di?erent features may have greater
value when trying to end the game, etc. We will, therefore, allow programs to change behavior once during
the game. From then on, it can use a di?erent board evaluation function and search depth. As with any
board evaluation, it must be based on learned feature weights, not encoded endgame knowledge. If you
choose to use this feature, your client can only change to endgame mode once in a game and must remain
there until the end of the game.
Grading
Grades will be assigned based on a linear function of the quality of your report, the debrief, and how well
your client does in the tournament with weights approximately 0.4, 0.3, and 0.3 respectively. However,
4
winning the tournament will be given extra weight, so the winner is virtually guaranteed an A unless the
team neglects to produce a report or participate in a debrief.

因为专业,所以值得信赖。如有需要,请加QQ99515681  微信:codehelp

CSE4705 - Introduction to Artitcial Intelligence

标签:rem   database   toc   therefore   ace   sch   学生   Dimension   sequence   

原文地址:https://www.cnblogs.com/codingjava/p/11944271.html

(0)
(0)
   
举报
评论 一句话评论(0
登录后才能评论!
© 2014 mamicode.com 版权所有  联系我们:gaon5@hotmail.com
迷上了代码!