Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
An Investigative Study about Application of Supervised Learning for Making Predictions in Chess
KTH, School of Computer Science and Communication (CSC).
KTH, School of Computer Science and Communication (CSC).
2017 (English)Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesisAlternative title
En utredande studie kring tillämpning av supervised learning för att göra förutsägelser i schack (Swedish)
Abstract [en]

Supervised learning is not as popular as reinforcement learning in chess programming due to its inability to achieve as high prediction accuracies as reinforcement learning. However, through extensive search by the authors, there seems to be a few numbers of research conducted that focus on applying supervised learning into chess. Therefore, this study investigates how supervised learning could be used to make predictions in chess so that an empirical understanding of supervised learning using both logistic regression and convolutional neural networks is provided. Both the machine learning algorithms will be tested and compared to the prediction accuracies acquired by reinforcement learning through other studies (it will not be implemented in this study). The prediction task was to predict the position from which the next chess piece moves in a chess game.

It has been concluded from this study that convolutional neural networks are better at predicting than logistic regression, but had higher tendencies to suffer from overfitting compared to logistic regression. When comparing these two supervised learning algorithms to reinforcement learning, supervised learning algorithms do not achieve as high prediction accuracies as reinforcement learning in general, but could be used as heuristics in various programming contexts in the future.

Future research should investigate regularization techniques to overcome with overfitting tendencies in both machine learning algorithms and investigate how data representations may affect the prediction accuracy of respective machine learning algorithm.

Abstract [sv]

Supervised learning är inte lika populär som reinforcement learning i schackprogrammering på grund av dess oförmåga att uppnå samma höga förutsägelsenoggrannhet som reinforcement learning. Genom omfattande sökning av författarna verkar det dock finnas några få undersökningar som fokuserar på att tillämpa supervised learning i schack. Därför undersöker den här studien hur supervised learning kan användas för att göra förutsägelser i schack för att erhålla vidare kunskap om supervised learning med hjälp av både logistic regression och convolutional neural networks inom schackprogrammering. Båda maskininlärningsalgoritmerna kommer att testas och jämföras med de förutsägelsenoggrannheter som erhållits av reinforcement learning genom andra studier (det kommer inte att implementeras i denna studie). Förutsägelserna, som görs i denna studie, var att förutsäga den position, från vilken nästa schackpjäs kommer att röra sig. Slutsatserna från denna studie är att convolutional neural networks är bättre på att förutsäga än logistic regression, men hade högre tendenser att drabbas av overfitting jämfört med logistic regression. När man jämför dessa två ”supervised learning”-algoritmer med reinforcement learning visar resultaten att ”supervised learning”-algoritmerna inte uppnår lika höga förutsägelsenoggrannheter som reinforcement learning i allmänhet, men kan användas som heuristiker i diverse programmeringssammanhang i framtiden. Vidare forskning bör vara att undersöka regularization-tekniker för att överbrygga problemen med overfitting-tendenser i båda maskininlärningsalgoritmerna och undersöka hur datarepresentationer kan påverka förutsägelsenoggrannheten hos respektive maskininlärningsalgoritm.

Place, publisher, year, edition, pages
2017.
National Category
Computer Science
Identifiers
URN: urn:nbn:se:kth:diva-208909OAI: oai:DiVA.org:kth-208909DiVA: diva2:1108729
Supervisors
Examiners
Available from: 2017-06-17 Created: 2017-06-12 Last updated: 2017-06-17Bibliographically approved

Open Access in DiVA

fulltext(1919 kB)11 downloads
File information
File name FULLTEXT01.pdfFile size 1919 kBChecksum SHA-512
3c86b62d6dcfae7d3fbcb6431a772579182e0be08513045cbcb01c68524043e210f94865e71b52d126c67e2d9900a1281b61d87a1cbb923675af241284942d34
Type fulltextMimetype application/pdf

By organisation
School of Computer Science and Communication (CSC)
Computer Science

Search outside of DiVA

GoogleGoogle Scholar
Total: 11 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

Total: 297 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • harvard1
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf