Difference between revisions of "Supervised Learning"

From Chessprogramming wiki
Jump to: navigation, search
Line 6: Line 6:
  
 
=Move Adaption=
 
=Move Adaption=
[[Automated Tuning#MoveAdaption|Move adaption]] applies [[Automated Tuning#LinearRegression|linear regression]] using a [https://en.wikipedia.org/wiki/Loss_function cost function]
+
[[Automated Tuning#MoveAdaption|Move adaption]] can be applied by [[Automated Tuning#LinearRegression|linear regression]] to minimize a [https://en.wikipedia.org/wiki/Loss_function cost function] considering the rank-number of the desired move in a [[Move List|move list]] ordered by score <ref>[[Tony Marsland]] ('''1985'''). ''Evaluation-Function Factors''. [[ICGA Journal#8_2|ICCA Journal, Vol. 8, No. 2]], [http://webdocs.cs.ualberta.ca/~tony/OldPapers/evaluation.pdf pdf]</ref>.
to minimize the rank-number of the desired move in a [[Move List|move list]] ordered by score <ref>[[Tony Marsland]] ('''1985'''). ''Evaluation-Function Factors''. [[ICGA Journal#8_2|ICCA Journal, Vol. 8, No. 2]], [http://webdocs.cs.ualberta.ca/~tony/OldPapers/evaluation.pdf pdf]</ref>.
 
  
 
=Value Adaption=
 
=Value Adaption=

Revision as of 22:13, 15 November 2019

Home * Learning * Supervised Learning

Supervised Learning,
is learning from examples provided by a knowledgable external supervisor. In machine learning, supervised learning is a technique for deducing a function from training data. The training data consist of pairs of input objects and desired outputs [1]. In computer games and chess, supervised learning techniques were used in automated tuning or to train neural network game and chess programs. Input objects are chess positions. The desired output is either the supervisor's move choice in that position (move adaption), or a score provided by an oracle (value adaption).

Move Adaption

Move adaption can be applied by linear regression to minimize a cost function considering the rank-number of the desired move in a move list ordered by score [2].

Value Adaption

One common idea to provide an oracle for supervised value adaption is to use the win/draw/loss outcome from finished games for all training positions selected from that game. Discrete {-1, 0, +1} or {0, ½, 1} desired values are the domain of logistic regression and require the evaluation scores mapped from pawn advantage to appropriate winning probabilities using the sigmoid function to calculate a mean squared error of the cost function to minimize, as demonstrated by Texel's Tuning Method.

See also

Selected Publications

1960 ....

  • Arthur Samuel (1967). Some Studies in Machine Learning. Using the Game of Checkers. II-Recent Progress. pdf

1980 ...

1990 ...

2000 ...

2010 ...

Forum Posts

External Links

References

Up one Level