JAIST Repository: 機械学習による囲碁の着手の日本語表現

全文

(1)JAIST Repository https://dspace.jaist.ac.jp/. Title. 機械学習による囲碁の着手の日本語表現. Author(s). 宍戸, 崇音; 池田, 心; ビエノ, シモン. Citation. 研究報告ゲーム情報学（GI）, 2015-GI-33(4): 1-7. Issue Date. 2015-02-26. Type. Journal Article. Text version. publisher. URL. http://hdl.handle.net/10119/13463. Rights. 社団法人情報処理学会, 宍戸崇音, 池田心, ビエノシモン, 研究報告ゲーム情報学（GI）, 2015-GI33(4), 2015, 1-7. ここに掲載した著作物の利用に関する注意: 本著作物の著作権は（社）情報処理学会に帰属します。本著作物は著作権者である情報処理学会の許可のもとに掲載するものです。ご利用に当たっては「著作権法」ならびに「情報処理学会倫理綱領」に従うことをお願いいたします。 Notice for the use of this material: The copyright of this material is retained by the Information Processing Society of Japan (IPSJ). This material is published on this web site with the agreement of the author (s) and the IPSJ. Please be complied with Copyright Law of Japan and the Code of Ethics of the IPSJ if any users wish to reproduce, make derivative work, distribute or make available to the public any part or whole thereof. All Rights Reserved, Copyright (C) Information Processing Society of Japan.. Description. Japan Advanced Institute of Science and Technology.

(2) Vol.2015-GI-33 No.4 2015/3/5. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. ‫ػ‬ցֶशʹΑΔғ‫ޟ‬ͷணखͷ೔ຊ‫ޠ‬ද‫ݱ‬ ࣡‫ ށ‬ਸԻ1,a). ஑ా ৺1,b). ϏΤϊ γϞϯ1,c). ֓ཁɿۙ೥ͷғ‫ޟ‬ϓϩάϥϜͷ‫͞ڧ‬͸ɼϓϩ‫ ʹ࢜ع‬4 ࢠͷϋϯσͰউͭͳͲɼ΄ͱΜͲͷΞϚνϡΞʹͱͬ ͯॆ෼ͳҬʹୡͭͭ͋͠ΔɽͦͷͨΊɼ࣍ͷஈ֊ͱͯ͠ਓؒΛ‫͑ڭ‬Δɾָ͠·ͤΔͱ͍ͬͨ໨తͰͷ‫ڀݚ‬ ΋੝Μʹͳ͖͍ͬͯͯΔɽࢦಋ‫ޟ‬΍઀଴‫Ͱޟ‬ਓؒΛָ͠·ͤΔཁૉͷ 1 ͭʹʮ‫ײ‬૝ઓɼ‫ݕ‬౼ɼର‫ہ‬தͷ͓ ஻Γʯ͕͋Δ͕ɼ͜ͷͨΊʹ͸ʠ‫ܗ‬ʡΛද‫͢ݱ‬Δ୯‫ޠ‬ʢπέɼϋωͳͲʣΛίϯϐϡʔλʹද‫ͤ͞ݱ‬Δ͜ͱ ͕๬·͍͠ɽͦ͜Ͱຊ࿦จͰ͸ɼ‫ػ‬ցֶशΛ༻͍ͯ൫໘ͱணख͔Β୯‫ޠ‬Λಋ͘͜ͱΛ໨ࢦͨ͠ɽ·ͣɼ‫ܗ‬ ͷ୯‫ޠ‬Λ໿ 70 छྨʹߜͬͨ͏͑ͰɼΞϚνϡΞߴஈऀ 6 ਓʹ‫ع‬ේΛ౉֤ͯ͠ணखʹϥϕϧ෇͚Λͯ͠΋ Βͬͨɽ͜ͷࡍɼʮϋωͱ΋‫͑ݴ‬Δ͠ɼΦαΤͱ΋‫͑ݴ‬ΔʯΑ͏ͳख͕සൟʹ͋Δͱ͍͏ࠔ೉͞Λߟྀ͠ɼ ෳ਺ͷϥϕϧΛ෇͚Δ͜ͱ͕Ͱ͖ΔΑ͏ͳϑΥʔϚοτͱ͠ධՁͷࢀߟͱͨ͠ɽֶशʹ͸ɼணखͷपғͷ ഑ੴύλʔϯҎ֎ʹɼ‫఺ٵݺ‬ͷมԽ΍ੴ͕Կઢʹ͋Δ͔ͳͲғ‫ޟ‬ಛ༗ͷಛ௃ྔΛ༻͍Δ͜ͱͰੑೳ޲্Λ ਤͬͨɽਓؒಉ࢜Ͱ͋ͬͯ΋୯‫ޠ‬ͷҰக཰͸໿ 82 ˋʹ͗͢ͳ͍͕ɼൺֱత୯७ͳ‫ػ‬ցֶशͰ΋͜Εʹ͍ۙ ஋Λग़͢͜ͱʹ੒ޭͨ͠ɽணखͷ೔ຊ‫ޠ‬ද‫ʹݱ‬Αͬͯɼίϯϐϡʔλͱͷ‫ײ‬૝ઓɼ‫ݕ‬౼ɼ͓஻Γͷ࣮‫ʹݱ‬ ۙͮ͘ͱͱ΋ʹɼॳ‫ऀڃ‬ͷ஌ࣝఆண΋ਤΔ͜ͱ͕Ͱ͖Δɽ Ωʔϫʔυɿָ͠·ͤΔ AIɼғ‫ޟ‬ɼ‫ػ‬ցֶशɼ೔ຊ‫ޠ‬ද‫ݱ‬ɼ‫ܗ‬ͷ໊લ. Japanese expression of the move of Go by machine learning SHISHIDO Takanari1,a). IKEDA Kokolo1,b). VIENNOT Simon1,c). Abstract: Computer Go programs have recently won against professional players with a 4-stone handicap, which is a level of strength sufficient for most amateur players. A new target for research is then to create programs able to entertain or teach Go to human players, but communication is a major obstacle, especially because moves in the game of Go are described by many specific terms such as Tsuke or Hane. In this research, our goal is to make the program able to label the moves with their associated specific term. We used machine learning to deduce the term for a move from the local patterns of stones. First, 6 strong amateur Go players recorded for each move of some game records the corresponding specific term, or possibly multiple terms, from a pre-selected list of 71 terms. Secondly, a machine learning algorithm was executed and the performance was improved by using not only the local patterns of stones but also features specific to the game of Go, such as changes of liberties or distances to the edge of the board. The human players associated the same specific term to a move at a rate of 82% and our progam succeeded to achieve a similar rate although the machine learning method was rather simple. Such derivation of the terms for moves is a first step towards Go programs able to chat with human players during game reviews or matches. Keywords: Entertainment Computing, Go, Machine Learning, Japanese Expression, Specific Term. 1. ͸͡Ίʹ 1 a) b) c). ๺཮ઌ୺Պֶٕज़େֶӃେֶ JAIST, Asahidai 1-1, Nomi, Ishikawa, 923–1211, Japan [email protected] [email protected] [email protected]. ⓒ 2015 Information Processing Society of Japan. ௕͍ؒɼίϯϐϡʔλғ‫ͯͬͱʹޟ‬ͷத৺త՝୊͸ʮ‫ڧ‬ ͘͢Δ͜ͱʯ͕ͩͬͨɼZen ͕෢‫ٶ‬ਖ਼थ‫۝‬ஈʹ 4 ࢠஔ͍ͯ উͭͳͲɼ΄ͱΜͲͷϓϨΠϠʹͱͬͯίϯϐϡʔλғ. 1.

(3) Vol.2015-GI-33 No.4 2015/3/5. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. ‫ޟ‬ͷ‫͞ڧ‬͸ॆ෼ͳҬʹୡͭͭ͋͠ΔɽͦͷͨΊɼ࣍ͷஈ֊. ͢Δ޻෉΋͞Ε͍ͯΔɽ࠷ۙͷࢢൢιϑτʮఱ௖ͷғ‫ ޟ‬5ʯ. ͱͯ͠ਓؒΛ‫͑ڭ‬Δɾָ͠·ͤΔͱ͍ͬͨ໨తͰͷ‫ڀݚ‬΋. ʹ΋ঁྲྀ‫࢜ع‬ͷ੠Ͱணखͷ‫ܗ‬ΛಡΈ্͛ͯ͘ΕΔ‫ػ‬ೳ͕͋. ੝Μʹͳ͖͍ͬͯͯΔɽ‫ࡏݱ‬ɼॳ‫ऀڃ‬ͷࢦಋ͸ଟ͘ͷ৔߹. ΓɼϓϨΠϠͷຬ଍౓ΛߴΊΔ͜ͱʹߩ‫͍ͯ͠ݙ‬Δ [3]ɽ͜. ਓؒͷ্‫͍ͯͬ୲͕ऀڃ‬Δɽ͔͠͠ɼ্‫ऀڃ‬͸‫ॆ͕͞ڧ‬෼. ΕΒͷιϑτͷ಺෦ͰͲͷΑ͏ʹணखΛ‫ܗ‬ͷ໊લʹม‫͠׵‬. Ͱ͋ͬͯ΋‫͑ڭ‬Δɾָ͠·ͤΔٕज़͸ॆ෼Ͱͳ͍৔߹͕͋. ͍ͯΔ͔͸ෆ໌Ͱ͋Δ͕ɼ͓ͦΒ͍͔ͭ͘͘ͷ৚݅จΛ༻. Δɽ·ͨɼͦΕΒͷٕज़Λ࣋ͭࢦಋऀ͸গ਺ͰɼࢦಋΛड. ͍ͨϧʔϧϕʔεͷ൑ఆͰ͋Ζ͏ͱ૝૾͞ΕΔɽຊ࿦จͷ. ͚Δͷ͸ߴίετʹͳΔͨΊɼࢦಋ‫ޟ‬΍઀଴‫Ͱޟ‬ਓؒΛָ. ໨తͷҰͭ͸ɼ͜ͷΑ͏ͳٕज़Λ࠶‫ݱ‬Մೳͳ‫هͰܗ‬ड़͢Δ. ͠·ͤΔ͜ͱ͕Ͱ͖Δίϯϐϡʔλғ‫͕ޟ‬๬·ΕΔɽ஑ా. ͜ͱʹ͋Δɽ. Β͸ɼίϯϐϡʔλ͕઀଴‫͍͓ͯʹޟ‬ਓؒΛָ͠·ͤΔͨ Ίʹඞཁͳཁૉͱͯ͠ɼ1) ૬खϞσϧͷ֫ಘɼ2) ‫ܗ‬੎ͷ. ʠ൫໘ͷঢ়ଶͱணख͔Βɼͦͷணखͷ‫ܗ‬ͷ໊લΛਖ਼͘͠. ༠ಋɼ3) ෆࣗવͳணखͷഉআɼ 4) ଟ༷ͳઓུɼ5) ணख΍. ؔ࿈෇͚Δʡ͜ͱ͸ɼ΋໊֤͠લ͕໌͔֬ͭ؆ܿʹఆٛͰ. ౤ྃͷద੾ͳλΠϛϯάɼ6) ‫ײ‬૝ઓɼ‫ݕ‬౼ɼ͓͠Ό΂Γɼ. ͖ΔͳΒ͹ɼख࡞‫Ͱۀ‬ϧʔϧΛ࡞੒͢Ε͹Α͍ɽ͔࣮͠͠. ͷ 6 ͭΛ‫͍ͯ͛ڍ‬Δ [1]ɽຊ‫ڀݚ‬ͷ໨త͸ɼ‫ײ‬૝ઓɾ‫ݕ‬౼ɾ. ࡍʹ͸ɼ ʮϚΨϦͱΦαΤʯ ʮϊϏͱώΩʯ ʮπϝͱώϥΩʯ. ͓஻Γ͕Ͱ͖Δίϯϐϡʔλͷ࣮‫ͨͮۙ͘ʹݱ‬Ίʹɼ ʠ‫ܗ‬ʡ. ͷΑ͏ʹҧ͍͕ඍົͰ໌จԽ͠ʹ͍͘΋ͷ΋ଟ͍ɽ͜ͷΑ. Λද‫͢ݱ‬Δ୯‫ޠ‬ʢπέɼϋωͳͲʣΛίϯϐϡʔλʹද‫ݱ‬. ͏ͳ৔߹ʹ͠͹͠͹༻͍ΒΕΔͷ͕ɼ‫ػ‬ցֶशͷҰͭɼ‫ڭ‬. ͤ͞Δ͜ͱͰ͋ΔɽͳͥͳΒɼਓؒಉ͕࢜ғ‫ޟ‬ͷ‫ײ‬૝ઓ΍. ࢣ͋ΓֶशͰ͋Δ [4]ɽ‫͋ࢣڭ‬ΓֶशͰ͸ɼೖྗͱग़ྗͷਖ਼. ‫ݕ‬౼Λߦ͏ࡍɼʮ͜͜͸े࿡ͷ 12 ͡Όͳͯ͘े࿡ͷ 13 ͔. ղྫΛଟ͘༩͑ͨ͏͑Ͱɼؔ਺ϞσϧΛબ୒ͯͦ͠ͷύϥ. ेࣣͷ 11 ͩͱࢥͬͨʯͱ͍͏Α͏ͳணखͷҐஔΛ࠲ඪͰ. ϝʔλΛࣗಈͰ࠷దԽ͢Δɽྫ͑͹ɼ͜͏͍͏ঢ়‫Ͱگ‬͸ϊ. ड़΂Δ͜ͱ͸‫Ͱك‬ɼ΄ͱΜͲͷ৔߹͸ʮ͜͜͸έΠϚ͡Ό. Ϗͩɼ͜͏͍͏ঢ়‫Ͱگ‬͸ώΩͩɼͱ͍͏ਖ਼ղྫΛ 100 ྫͣ. ͳͯ͘πέ͕ίεϛͩͱࢥͬͨʯͳͲ‫ܗ‬Λද‫͢ݱ‬Δ୯‫ޠ‬Λ. ͭ༩͑ΒΕΕ͹ɼϊϏ΍ώΩ͕ຊདྷͲ͏͍͏ҙຯͰ͋Δ͔. ༻͍Δ͔ΒͰ͋Δɽίϯϐϡʔλʹ‫ܗ‬Λ೔ຊ‫ޠ‬ද‫ͤ͞ݱ‬Β. ͷఆٛΛ஌Βͣͱ΋ɼͲͪΒͳͷ͔ະ஌ͷྫʹରͯ͋͠Δ. ΕΕ͹ɼॳ‫ऀڃ‬ͷ‫ؔ͢ʹܗ‬Δ஌ࣝఆணʹ໾ཱͭ͜ͱ΋Ͱ͖. ఔ౓ਖ਼͘͠౴͑ΒΕΔΑ͏ʹͳΔͩΖ͏ɽ. Δɽຊ࿦จͰ͸ɼ ʠ‫ܗ‬ʡΛද‫͢ݱ‬Δ୯‫ޠ‬Λίϯϐϡʔλʹද. ‫͋ࢣڭ‬Γֶशʹ͸ඇৗʹଟ͘ͷλΠϓ͕͋Γɼ·ͨͦΕ. ‫ͤ͞ݱ‬ΔͨΊʹɼ‫ػ‬ցֶशΛ༻͍ͯ൫໘ͱணख͔Β୯‫ޠ‬Λ. Ώ͑ʹඇৗʹଟ͘ͷؔ਺Ϟσϧ΍ֶश๏͕༻͍ΒΕΔɽೖ. ಋ͘͜ͱΛ໨ࢦͨ͠ɽ. ྗ͕཭ࢄ஋ͳͷ͔࿈ଓ஋ͳͷ͔ɼೖྗཁૉ਺͕ଟ͍͔গͳ. 2. ؔ࿈‫ڀݚ‬. ͍͔ɼग़ྗ͕ yes/no ͷ 2 ஋ͳͷ͔ɼෳ਺‫ݸ‬ͷϥϕϧͳͷ ͔ɼ࿈ଓ஋ͳͷ͔ɼͦΕͧΕʹ߹Θͤͯؔ਺Ϟσϧ΍ֶश. ਓ޻஌ೳٕज़ͷൃలͱ‫ػࢉܭ‬ੑೳͷ޲্ʹΑΓɼଟ͘ͷ. ๏ΛબͿඞཁ͕͋Δɽ୅දతͳ΋ͷͱͯ͠͸ɼຊ࿦จͰ༻. ήʔϜͰίϯϐϡʔλϓϨΠϠͷ‫͞ڧ‬͸े෼ͳ΋ͷʹͳΓ. ͍Δܾఆ໦ͷଞʹɼχϡʔϥϧωοτϫʔΫ΍ɼαϙʔτ. ͭͭ͋Γɼࣗવ͞΍ָ͠͞Λ໨ࢦͨ͠‫͕ڀݚ‬஫໨͞ΕΔΑ. ϕΫλʔϚγϯͳͲ͕‫͛ڍ‬ΒΕΔɽܾఆ໦͸ଞͷೋͭʹൺ. ͏ʹͳ͍ͬͯΔɽྫ͑͹εʔύʔϚϦΦϒϥβʔζʹ୅ද. ΂ͯม਺ؒґଘੑͷߴ͍࿈ଓ஋໰୊Ͱ͸෼ྨੑೳ͕ྼΔ. ͞ΕΔԣεΫϩʔϧΞΫγϣϯήʔϜʹؔ͢Δίϯϐϡʔ. ͕ɼֶश͕ߴ଎Ͱɼ·ͨಘΒΕͨ݁Ռ͔ΒʮͲͷΑ͏ͳ৚. λ‫ٕڝ‬ձͰ͸ɼ୯ʹʮ͏·͘ϚϦΦΛૢ࡞ͯ͠ૣ͘ΫϦΞ. ݅Ͱ൑ఆ͍ͯ͠Δ͔ʯΛཧղ͠΍͍͢ͱ͍͏ར఺͕͋Δɽ. ͢Δʯͱ͍͏໨తͷ‫ٕڝ‬ͷଞʹɼ ʮਓ͕ؒϓϨΠ͍ͯ͠ΔΑ. 3. ఏҊख๏. ͏ʹϚϦΦΛૢ࡞͢Δʯ͋Δ͍͸ʮਓ͕ؒϓϨΠָͯ͠͠ ͍ͱࢥ͑ΔεςʔδΛੜ੒͢Δʯͱ͍͏‫ߦ͕ٕڝ‬ΘΕɼ஫ ໨ΛूΊͨ [2]ɽ ஑ాΒ͸ɼίϯϐϡʔλ͕઀଴‫͍͓ͯʹޟ‬ਓؒΛָ͠·. ຊ‫ڀݚ‬͸‫ػ‬ցֶशΛ༻͍ͯ൫໘ͱணख͔Β୯‫ޠ‬Λಋ͘͜ ͱΛ໨ࢦ͢ɽ‫ڀݚ‬खॱΛҎԼʹࣔ͢ɽ. ( 1 ) ‫ط‬ଘιϑτͷ೔ຊ‫ޠ‬ද‫ػݱ‬ೳͷௐࠪΛߦ͏ɽ. ͤΔͨΊʹඞཁͳཁૉͱͯ͠ 6 ͭͷཁૉΛ‫͛ڍ‬ɼͦͷ͏ͪ. ਓ‫ؾ‬ιϑτʮఱ௖ͷғ‫ ޟ‬5ʯ͸ணखͷ‫ܗ‬ΛಡΈ্͛Δ. ଟ༷ͳઓུͷԋग़΍ɼෆࣗવͳணखΛ཈੍͠ͳ͕Β‫ܗ‬੎Λ. ‫ػ‬ೳΛ͍࣋ͬͯΔɽͦΕ͕೔ຊ‫ޠ‬ද‫ػݱ‬ೳͱͯ͢͠Ͱ. ༠ಋ͢Δํ๏ʹ͍ͭͯ͸۩ମతͳΞϓϩʔνΛఏҊ͍ͯ͠. ʹॆ෼Ͱ͋Δ͔Ͳ͏͔ɼ্‫ʹऀڃ‬ΑΔධՁΛߦ͏͜ͱ. Δ [1]ɽ͔͠͠‫ײ‬૝ઓɾ‫ݕ‬౼ɾ͓͠Ό΂Γ౳ʹ͍ͭͯ͸ඞ. Ͱௐࠪ͢Δɽ. ཁੑ͕ड़΂ΒΕ͍ͯΔ͚ͩͰ۩ମతͳख๏͸ఏҊ͞Ε͍ͯ ͳ͍ɽ. ( 2 ) ҰखͷΈʢख‫ے‬Ҏ֎ʣʹ͍ͭͯ‫͋ࢣڭ‬ΓֶशΛߦ͏ɽ ғ‫ޟ‬ͷखͷ‫ʹܗ‬͸ɼख‫ݺͱے‬͹ΕΔෳ਺ख͔ΒͳΔ΋. ࢢൢιϑτ΢ΣΞͰ͸͜ͷΑ͏ͳ‫ײ‬૝ઓɾ‫ݕ‬౼͸͍ͭ͘. ͷ΋͋Δ͕ɼ·ͣ͸ҰखͷΈͷ‫ܗ‬ɼதͰ΋‫ج‬ຊతͳ‫ܗ‬. ͔ࢼΈΒΕ͍ͯΔɽͨͱ͑͹ʮ΍͍͞͠ғ‫ޟ‬ʯͰ͸Ωϟϥ. ʹߜͬͯ‫͋ࢣڭ‬ΓֶशΛߦ͍ɼ‫ܗ‬ͷ෼ྨͱ೔ຊ‫ޠ‬ද‫ݱ‬. Ϋλʹ‫Ͱޠޱ‬஻ΒͤΔ͜ͱͰٖਓԽΛਤΓɼ·ͨ࠲ඪͰ͸. Λ໨ࢦ͢ɽ‫͋ࢣڭ‬Γֶशʹඞཁͳֶशσʔλ͸ɼ্‫ڃ‬. ͳ͘Ұ෦ʠ੾ΓͷखͰྑ͍खͰ͢ͶʡͳͲ‫ܗ‬ͷ໊લͰද‫ݱ‬. ऀͷ‫ྗڠ‬Λಘͯ‫ہ‬໘ɾखͱରԠ͢Δ‫ܗ‬ͷηοτΛूΊ. ⓒ 2015 Information Processing Society of Japan. 2.

(4) Vol.2015-GI-33 No.4 2015/3/5. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report ද 1. Δɽ࣍ʹɼੴͷઈରҐஔ΍ύλʔϯͳͲ‫ܗ‬ͷ෼ྨʹӨ ‫ͳ͏ͦ͠ڹ‬ಛ௃ྔΛઃ‫͠ܭ‬ɼ‫ہ‬໘ͱख͔Βಛ௃ྔΛந ग़ͨ͠ͷͪɼೖྗʢ‫ہ‬໘ɾखʣͱग़ྗʢ‫ܗ‬ͷ໊લʣͷ ηοτΛ࡞੒͢Δɽͦͯͦ͠ΕΛܾఆ໦ֶश๏ͷҰͭɼ. J4.8 Ͱֶश͢Δɽ ( 3 ) ϓϩ‫ʹ࢜ع‬ධՁͯ͠΋Β͏ɽ. πΪ (1404). બఆͨ͠‫ͱܗ‬ग़‫ݱ‬ճ਺ ΧΧΤ (135). Χυ (44). ΦαΤ (1062). ίϞΫ (133). τϏαΨϦ (40). ϋω (940). ϑΫϥϛ (123). ϋαϛπέ (39). ΞλϦ (827). ੕ (105). େήΠϚ (37). ϊϏ (639). ίεϛπέ (103). ώϥΩρϝ (37). σ (612). Ξςίϛ (101). ϗ΢Ϧίϛ (36). ಘֶͨश݁ՌʹΑΔ‫ܗ‬ͷ෼ྨ͕ɼਓؒʹͱͬͯద੾ͳ. τϏ (575). άζϛ (88). ΦΩ (35). ͷ͔ධՁͯ͠΋Β͏ɽ͜ͷධՁʹ͸ਖ਼ղ͔ෆਖ਼ղ͔ɼ. ΩϦ (531). τϏπέ (87). ϋβϚ (26). Ͱ͸ͳ͘ຬ଍౓Λ༻͍Δɽຬ଍౓Λ༻͍Δͷ͸ɼ‫ܗ‬ͷ ൑அ͸ਖ਼ղෆਖ਼ղ͕໌֬ʹఆΊΒΕΔ΋ͷͰ͸ͳ͍͔ ΒͰ͋Δɽ. 4. ֶशσʔλͷ࠾औͱ‫ط‬ଘख๏ͷੑೳ 4.1 ‫ܗ‬ͷ‫ݶ‬ఆ ຊ‫Ͱڀݚ‬͸ɼ·ͣ֓Ͷ‫ج‬ຊతͳ‫ܗ‬ͷ೔ຊ‫ޠ‬ද‫ݱ‬ΛࢼΈ ΔɽͦͷͨΊɼ‫ܗ‬Λද͋͢ΒΏΔғ‫ޠ༻ޟ‬Λ࠷ॳ͔Βίϯ ϐϡʔλʹ‫ݴ‬ΘͤΑ͏ͱ͢ΔͷͰ͸ͳ͘ɼද 1 ʹࣔͨ͠ զʑ͕બఆͨ͠‫ج‬ຊతͳ‫ ܗ‬71 ‫ݶʹݸ‬ఆͯ͋͠Δɽ ྫ͑͹ɼ্πέ΍Լπέɼ֎πέ΍಺πέ͸શͯπέɼ ҰؒߴΨΧϦ΍ೋؒΨΧϦ΋શͯΧΧϦͱ͍ͯ͠Δɽ͞Β ʹɼ ʮ߈Ίʯ ʮकΓʯ ʮγνϣ΢ΞλϦʯ ʮ͙͘͢Γʯ ʮί΢μ ςʯ ʮ༷ࢠ‫ݟ‬ʯ ʮ͖͔͠ʯͳͲɼ‫͏͍ͱܗ‬ΑΓ͸ʮखͷҙຯʯ. πέ (441). ϋαϛ (84). λν (26). έΠϚ (386). ϫλϦ (80). πέίγ (20). ίεϛ (352). ϋωμγ (67). ϋωίϛ (18). ψΩ (351). τϦ (67). αγίϛ (18). Φγ (302). γϚϦ (66). πΩμγ (18). ϊκΩ (295). Χέ (66). ϫϦ΢ν (16). ϚΨϦ (251). ΢νίϛ (65). τϏίϛ (10). αΨϦ (223). εϕϦ (64). έΠϚπΪ (9). ώϥΩ (209). Ϙ΢γ (62). ϋαϛฦ͠ (7). ϒπΧϦ (203). ϫϦίϛ (62). ੕Լ (7). ϋΠ (193). πϝ (60). ྆ΨΧϦ (7) ϔίϛ (6). ώΩ (192). λέϑ (54). ί΢τϦ (176). ࡾʑ (54). ήλ (4). ΧΧϦ (170). Χλ (50). ໨ϋζγ (5). ΧέπΪ (151). ιΠ (46). ߴ໨ (4). χή (139). φϥϏ (46). ʹ͋ͨΔΑ͏ͳ༻‫ޠ‬΋আ͍ͯ͋Δɽ. 4.2 ֶशσʔλͷ࠾औ ֶशσʔλͱ͢Δ‫ہ‬໘ɾखͱ‫ܗ‬ͷ໊લͷηοτͷ࠾औ͸ɼ ਓؒͷ্‫ऀڃ‬ͷ‫ྗڠ‬ΛಘͯߦͬͨɽϑϦʔͷғ‫عޟ‬ේ࠶ ੜɾฤूιϑτʮMultiGoʯ[5] Λ༻͍ͯɼਤ 1 ͷΑ͏ʹ‫ع‬ ේதʹදΕͨ‫ܗ‬Λೖྗͯ͠΋Βͬͨɽ ೖྗͷࡍɼೖྗϑΥʔϚοτ͸ʮτϏɼώϥΩʢ90ʣ ʯͷ Α͏ʹୈҰީิ͚ͩͰͳ͘ୈೋީิͷ‫ܗ‬Λ఺਺Λ‫ʹڞ‬ೖྗ Ͱ͖ΔΑ͏ʹͨ͠ʢτϏʹ఺਺͕ॻ͔Ε͍ͯͳ͍ͷ͸ 100 ఺Λҙຯ͍ͯ͠Δɽೖྗͷ؆୯ͷͨΊʣɽ͜Ε͸ɼ‫ܗ‬͸། Ұͷਖ਼ղʹఆ·Βͣ४ਖ਼ղͷΑ͏ͳ΋ͷ͕͋Δ͕࣌͋Δͨ ΊͰ͋Δɽ఺਺͸ 70ʙ100 ఺ͷൣғͰೖྗͯ͠΋Β͍ɼ࠷ ΋ద੾ͱࢥ͏‫ܗ‬Λ 100 ఺ɼʮࢲͳΒ A ͩͱࢥ͏͚ͲɼB Ͱ ΋͞΄Ͳҧ࿨‫ײ‬͸‫͍ͳ͡ײ‬ʯఔ౓ͳΒɼB90 ఺ɼ ʮB ͱ‫ݴ‬Θ. ਤ 1 ‫ܗ‬ͷೖྗͷ༷ࢠ. ΕΔͱগ͠ҧ࿨‫͕͋ײ‬ΔʯͳΒ 80 ఺ɼ ʮB ΋͋Γ͔΋͠Ε. ৔ճ਺ 10 ճҎԼͷ΋ͷ΋͍͔ͭ͘ଘࡏ͠ɼ͜ΕΒ͸ࣗಈత. ͳ͍͕ɾɾɾʯఔ౓ͳΒ 70 ఺ɼͱྫࣔͨ͠ɽ. ͳ‫ػ‬ցֶशͰ͸ਫ਼౓ͷߴ͍෼ྨ͸ࠔ೉Ͱ͋Δͱ༧૝͞ΕΔɽ. 4.3 ֶशσʔλͷಛ௃. ‫ع‬ේʢ૯ख਺ 117 खʣ͚ͩ͸‫ڞ‬௨ͯ͠ೖྗͯ͠΋Βͬͨɽ. 6 ໊͸‫ج‬ຊతʹ͸ҟͳΔ‫ع‬ේʹ‫ܗ‬Λೖྗ͕ͨ͠ɼ̍ຕͷ ຊઅͰ͸࣮ࡍʹֶशʹར༻ͨ͠σʔλͷಛ௃ʹ͍ͭͯड़. 2 ਓͷೖྗऀ͕ಉ͡खʹରͯ͠Ͳͷఔ౓ಉ͡‫ܗ‬ΛୈҰީิ. ΂Δɽೖྗͨ͠ͷ͸ 6 ໊ͷΞϚߴஈऀͰɼ๺཮ΞϚ໊ਓΛ. ͱ͔ͨ͠ௐ΂ͨͱ͜Ζɼͦͷׂ߹͸ฏ‫ Ͱۉ‬82.2%ɼୈೋީ. ‫ؚ‬Ίɼ֓Ͷ kgs4d Ҏ্Ͱ͋Δɽ‫ع‬ේʹ͸ϓϩ‫͋࢜ع‬Δ͍͸. ิͱͷҰகΛ‫ؚ‬Ίͨ৔߹Ͱ΋ 87.0% ʹա͗ͳ͔ͬͨɽೖྗ. τοϓΞϚͷ 60 ‫ہ‬Λ༻͍ɼ૯࠾औख਺͸ 11,526 खͱͳͬ. ऀ͸‫ڞ‬ஶऀʢ஑ాʣΛআ͍ͯۚ୔େֶғ‫ޟ‬෦һͰ͋Γɼಉ. ͨɽ71 छྨͷ‫ܗ‬ͷ͏ͪ࠷΋සൟʹొ৔ͨ͠ͷ͸πΪͰ 1404. ͡ίϛϡχςΟʹଐ͍ͯ͠ͳ͕Β 2 ׂۙ͘΋ҧ͏ҙ‫ݟ‬Λ࣋. ճɼଓ͍ͯΦαΤͷ 1062 ճͳͲͱͳ͍ͬͯΔʢද 1ʣɽ. ͭͱ͍͏ࣄ࣮͸ɼ‫ܗ‬ͷ໊લΛ‫͕ۀ࡞͏͍ͱ͏ݴ‬ᐆດ͔ͭࠔ. ొ৔ճ਺ʹ͸େ͖ͳ։͖͕͋Γɼ༗໊ͳ‫ͯͬ͋Ͱܗ‬΋ొ. ⓒ 2015 Information Processing Society of Japan. ೉ͳ΋ͷͰ͋Δ͜ͱΛ͍ࣔͯ͠Δɽ. 3.

(5) Vol.2015-GI-33 No.4 2015/3/5. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. 4.4 ‫ط‬ଘख๏ͷੑೳ ຊઅͰ͸ɼຊ‫ڀݚ‬ҎલʹߦΘΕͨணखͷ೔ຊ‫ޠ‬ද‫ݱ‬ͷࢼ Έʹ͍ͭͯɼࢢൢιϑτʮఱ௖ͷғ‫ ޟ‬5ʯͱғ‫ޟ‬ϓϩάϥ ϜʮNomitanʯͷੑೳΛड़΂Δɽ ఱ௖ͷғ‫ ޟ‬5ʢҎԼʮఱ௖ʯ ʣ͸ਓ‫ؾ‬ͷࢢൢιϑτͰ͋Γɼ ணखͷࡍʹͦͷணख͕ͲΜͳ‫͍ͯͬͳʹܗ‬Δͷ͔ಡΈ্͛ Δ‫ػ‬ೳΛ࣋ͭɽ ఱ௖ʹ‫ع‬ේ 4 ຕΛ༩͑ɼ૯ख਺ 262 ख෼ͷಡΈ্͛݁Ռ. ͳ͘ɼਤ 2 ͷΑ͏ʹண໨఺͔Β (δx, δy) ͚ͩ཭Εͨ఺ͷ ‫཭ڑ‬Λ d(δx, δy) = δx + δy + max(δx, δy) ͱͨ͠΋ͷͰ͋ Δ [7]ɽ. • PosXɼPosY ɿ (x,y) ࠲ඪΛ y ≤ x ≤ 10 ͱͳΔΑ͏ ʹճసɾ൓సͤͨ͞΋ͷɽ੕΍ίϞΫͳͲΛ෼ྨ͢Δ ͷʹඞཁɽ. • HeightɿԿઢ͔ɽҰ൪͍ۙ൫୺·Ͱͷ‫཭ڑ‬ɽ • DistToMyNearestɿ࠷‫د‬Γͷຯํͷੴ·Ͱͷ R ‫཭ڑ‬ɽ. Λ‫ه‬࿥ͨ͠ɽ࣍ʹɼຬ଍Ͱ͖Δ͔Ͳ͏͔Λɼ1 ໊ͷΞϚߴ. पғʹଞͷੴ͕ͳ͚Ε͹ɼ͜Ε͕ 2 ͳΒφϥϏɼ3 ͳ. ஈऀʹධՁͯ͠΋Βͬͨɽͦͯ͠ɼ262 खΛ࣍ͷ 4 ͭʹ෼. Βίεϛɼ4 ͳΒτϏɼ5 ͳΒέΠϚͳͲͱͳΔɽ. ྨͨ͠ɽ. • DistToOpNearestɿ࠷‫د‬Γͷఢͷੴ·Ͱͷ R ‫཭ڑ‬ɽप. 1). ఱ௖ͷಡΈ্͛Ͱਖ਼ղ. ғʹଞͷੴ͕ͳ͚Ε͹ɼ͜Ε͕ 2 ͳΒπέɼ3 ͳΒΧ. 2). ఱ௖ͷಡΈ্͛Ͱ͸ෆࣗવ. υ΍ΧλͳͲͱͳΔɽ. 3). ಡΈ্͛ແ͠ 4). ఱ௖ͷಡΈ্͛Ͱ͸ؒҧ͍ 4 ͕ͭͦΕͧΕ 262 खதʹ઎ΊΔׂ߹͸ɼ1) 65.6 ˋɼ2)2.3. • HeightOfMyNearestɿ࠷‫د‬Γͷຯํͷੴ͕Կઢʹ͋ Δ͔ɽ. • HeightOfOpNearestɿ࠷‫د‬Γͷఢͷੴ͕Կઢʹ͋Δ͔ɽ. ˋɼ3)30.2 ˋɼ4)1.9 ˋͰ͋ͬͨɽ͢ͳΘͪɼಡΈ্͛ͨ৔. ྫ͑͹͜Εͱ Height Λൺ΂Ε͹ɼΧυͱΧλɼΦγ. ߹ʹ͸ؒҧ͍͕গͳ͍͕ɼಡΈ্͛ͯ͘Εͳ͍͜ͱ͕͔ͳ. ͱϋΠͳͲ͕۠ผͰ͖Δ৔߹͕ଟ͍ɽ. Γͷׂ߹Ͱ͋ͬͨɽಡΈ্͛ͳ͍‫ͯ͠ͱܗ‬͸ɼΞςίϛɼ. • Lib1OpɿଧͨΕͨՕॴͷ্Լࠨӈʹɼఢͷ‫ ఺ٵݺ‬1 ͷ. άζϛɼϋαϛπέͳͲগ͠ߴ౓ͳ‫·ؚ͕ܗ‬Εͨɽࢢൢι. ੴͷूஂ͕͍ͭ͋͘Δ͔ɽ͜Ε͕͋Ε͹ψΩʹͳΔ͜. ϑτͰ͋ΔͷͰɼಡΈ্͛ͯؒҧ͏ΑΓ͸ɼಡΈ্͛ͳ͍. ͱ͕ଟ͍ɽ. ΄͏͕ྑ͍ͱͷ൑அͩͬͨͱਪଌ͢Δɽ. • Lib2OpɿଧͨΕͨՕॴͷ্Լࠨӈʹɼఢͷ‫ ఺ٵݺ‬2 ͷ ੴͷूஂ͕͍ͭ͋͘Δ͔ɽ͜Ε͕͋Ε͹ΞλϦʹͳΔ. Nomitan ͸๺཮ઌ୺Պֶٕज़େֶӃେֶͷ൧ా‫ݚ‬ɾ஑ా ‫Ͱݚ‬։ൃ͞Εͨғ‫ޟ‬ϓϩάϥϜͰ͋Γɼ‫ػ‬ցֶशͰ͸ͳ͘ ਓ͕ؒߟҊͨ͠ 554 ͷൺֱจʹΑΔϧʔϧʹΑͬͯ‫ܗ‬ͷ෼ ྨΛߦ͍೔ຊ‫ޠ‬ද‫͢ݱ‬Δ‫ػ‬ೳ͕͋Δɽ4.2 અͰಘͨ 11,526 खʹ͍ͭͯɼ֤‫ہ‬໘ɾखͰͷ Nomitan ͷग़ྗΛಘͨͱ͜ ΖɼਓؒͷୈҰީิͱ Nomitan ͷग़ྗͷҰக཰͸ 73.7 ˋɼ. ͜ͱ͕ଟ͍ɽ. • Lib1MyɿଧͨΕͨՕॴͷ্Լࠨӈʹɼຯํͷ‫ ఺ٵݺ‬1 ͷੴͷूஂ͕͍ͭ͋͘Δ͔ɽ. • Lib2MyɿଧͨΕͨՕॴͷ্Լࠨӈʹɼຯํͷ‫ ఺ٵݺ‬2 ͷੴͷूஂ͕͍ͭ͋͘Δ͔ɽ. • NewLibɿͦͷखΛଧͭ͜ͱʹΑͬͯɼͦͷஔ͔Εͨੴ. ୈೋީิ·Ͱ‫ؚ‬Ίͨ৔߹͸ 76.6 ˋͰ͋ͬͨɽ͜Ε͸ 4.3 અ. ͱɼͦͷੴͱ࿈͍݁ͯ͠Δੴͷूஂͷ‫͍͕ͭ͘఺ٵݺ‬. Ͱड़΂ͨਓؒಉ࢜ͷҰக཰ 82.2 ˋɼ87.0 ˋʹൺ΂Δͱ 10. ʹͳΔ͔ɽ. ˋҎ্ྼ͍ͬͯΔɽ΋ͱ΋ͱ͜ͷ‫ػ‬ೳ͸ 9 ࿏൫༻ [6] ʹ࡞ ΒΕͨ΋ͷͰ͋ΓɼώϥΩɾϘ΢γͳͲ޿͍൫Ͱొ৔͢Δ ‫ొ͕ܗ‬࿥͞Ε͍ͯͳ͔ͬͨ͜ͱ΋‫ݪ‬ҼͰ͋Δɽ ͜ΕΒͷ͜ͱ͔Βɼզʑ͸·ͩணखͷ೔ຊ‫ޠ‬ද‫ݱ‬ͷ‫ڀݚ‬. • CutNumɿࠨͱԼʹఢੴɼࠨԼʹຯํͷੴ͕͋ΔΑ͏ ͳ֨޷͔Ͳ͏͔ɽ௚઀తʹΩϦͱؔ܎͢Δɽ. • R ‫ ͕཭ڑ‬2ʙ4 ͷपғ 12 Ϛεͷঢ়ଶʢ0ɿۭɹ 1ɿຯํ ͷੴ͕͋Δɹ 2ɿఢͷੴ͕͋Δɹ 3ɿ൫֎ʣ. Ձ஋͕͋Δͱߟ͑ͨɽ. 5. ‫ػ‬ցֶशͱ༧උ࣮‫ݧ‬ 5.1 ಛ௃ྔͷઃ‫ܭ‬ ‫͋ࢣڭ‬ΓֶशͰ͸ɼೖྗΛ൫໘ͦͷ΋ͷͰ͸ͳ͘ɼͦ͜ ͔Β͍͔ͭ͘ͷಛ௃ྔΛநग़ͯ͠ΞϧΰϦζϜʹ౉͢͜ͱ ͕๬·͍͠ɽͲͷΑ͏ͳಛ௃ྔΛநग़͢Δ͔͸ੑೳʹ௚݁ ͠ɼૈ͗͢Δಛ௃ྔͰ͸ߴ͍ਫ਼౓͸๬Ίͣɼ‫͔͗͢ࡉʹٯ‬ Δಛ௃ྔͰ͸աֶशΛҾ͖‫͍ߴ͜͠ى‬൚Խੑೳ͕๬Ίͳ͍ɽ ·ͣզʑ͸ɼNomitan ͷϧʔϧϕʔεख๏ͷதͰ࢖ΘΕ ͍ͯΔม਺Λಛ௃ྔͷީิͱ͠ɼ͔ͦ͜Β໌Β͔ʹෆཁͳ ΋ͷΛআ͍ͨҎԼͷ 25 ‫ݸ‬ͷಛ௃ྔΛ༻͍Δ͜ͱʹͨ͠ɽ ͜ͷதͰొ৔͢ΔʮR ‫཭ڑ‬ʯ͸ɼϢʔΫϦου‫Ͱ཭ڑ‬͸. ⓒ 2015 Information Processing Society of Japan. ਤ 2. R ‫཭ڑ‬ͷྫɽ਺ࣈ͸֤఺ͱ˛ͱͷ‫཭ڑ‬. 4.

(6) Vol.2015-GI-33 No.4 2015/3/5. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. 5.2 ‫ػ‬ցֶशͷํ๏. ͷϧʔϧϕʔεͰߴ͍ਖ਼౴཰ͱͳ͍ͬͯΔͱ͍͏͜ͱ͸ɼ. ࠾औֶͨ͠शσʔλΛ‫ͯ͠ͱࢣڭ‬σʔλϚΠχϯάιϑ. ಛ௃ྔΛ͏·͘ઃ‫͢ܭ‬Ε͹ۤख෦෼͕ղফ͞ΕͯશମͷҰ. τ Weka ͷ J4.8ʢC4.5[8] Λ java Ͱ࣮૷ͨ͠΋ͷʣΛ༻͍. க཰΋޲্͢Δ͜ͱ͕‫ظ‬଴Ͱ͖Δɽ͜ΕΒͷ‫ܗ‬͸ग़‫ݱ‬ճ਺. ͯ‫ػ‬ցֶशΛߦ͏ɽͦͷखॱΛҎԼʹࣔ͢ɽ. ͕ଟ͍ͱ͍͏఺Ͱ΋Ұக཰޲্͕‫ظ‬଴Ͱ͖Δɽ. ( 1 ) ࠾औֶͨ͠शσʔλ͸‫ہ‬໘ɾखͱ‫ܗ‬ͷ໊લͷηοτͷ sgf ϑΝΠϧͰ͋ΓɼWeka Ͱ͸ѻ͏͜ͱ͕Ͱ͖ͳ͍ɽ. ද 2. पғͷύλʔϯͷҰ෦ʹؔ͢Δ‫ܗ‬ͷਖ਼౴཰ ਖ਼౴཰. Nomitan. ‫ػ‬ցֶश. ͦͷ sgf ϑΝΠϧ͔Β Nomitan ͱεΫϦϓτʹΑͬͯ. ϚΨϦʢ251ʣ. 76.6. 41.6. 5.1 અͰड़΂ͨಛ௃ྔΛநग़͠ɼ‫ہ‬໘ɾखɼ‫ܗ‬ͷ໊લɼ. σʢ612ʣ. 83.2. 59.3. ಛ௃ྔΛηοτʹͨ͠ csv ϑΝΠϧΛ࡞੒͠ɼWeka. Φγʢ302ʣ. 85.4. 65.2. Ͱѻ͑ΔΑ͏ʹͨ͠ɽ. ( 2 ) ࡞੒ͨ͠ csv ϑΝΠϧʢֶशσʔλʣΛ Weka ʹಡΈ ࠐΈɼֶशʹෆཁͳଐੑʢ‫ع‬ේ൪߸΍ख਺ʣΛ࡟আ͢. 6. ಛ௃ྔͷվળͱධՁ࣮‫ݧ‬. ΔલॲཧΛߦ͏ɽͦͯ͠ɼ෼ྨ‫ ʹث‬J4.8 Λબͼܾఆ໦. ຊষͰ͸ɼલઅͰ‫ݟ‬ΒΕͨ‫ػ‬ցֶशͷۤख෦෼Λղফ͢. Λ࡞੒ͤ͞ɼҰக཰ΛಘΔɽͳ͓ɼܾఆ໦࡞੒ʹཁ͢. Δ΂͘޻෉ͨ͠ಛ௃ྔͱɼͦΕΛ༻͍࣮ͨ‫݁ݧ‬Ռʹ͍ͭͯ. Δ࣌ؒ͸Ұൠతͳ PC Ͱ໿ 1 ඵɼ10 folding ަࠩ‫ূݕ‬. ड़΂Δɽ. ʹΑΔҰக཰ධՁΛߦͬͯ΋ 10 ඵఔ౓ͰऴΘͬͨɽ. ( 3 ) ͞Βʹɼग़ྗ݁ՌΛೖྗͯ͠΋Βͬͨୈೋީิͱ΋ൺ ֱ͠ɼͦͷҰக཰ʢ४ਖ਼ղ཰ʣ΋ಘΔɽ Ұக཰ΛಘΔ·ͰͷσʔλͷॲཧͷྲྀΕͷ֓೦ਤΛਤ 3 ʹࣔ͢ɽ. 6.1 ಛ௃ྔͷվળͱ࣮‫݁ݧ‬Ռ લઅͰ༻͍ͨपғͷύλʔϯʹؔ͢Δಛ௃ྔ͸ɼपғ 12 Ϛεͷঢ়ଶͰ͋Δɽ͜ΕΒͷಛ௃ྔͰ͸ɼ ʮࣄ্࣮ಉ͡഑ஔ ͷੴͰ΋ɼҧ͏ಛ௃ྔͱͯ͠ѻΘΕΔʯͱ͍͏໰୊͕͋Δɽ ྫ͑͹पғ 3 ʷ 3 Ϛε͕ਤ 4 ͷΑ͏ͳঢ়‫گ‬Λߟ͑Δɽ͜Ε Β͸શͯɼճస͓Αͼ൓సʹΑͬͯॏͶ߹ΘͤΔ͜ͱ͕Ͱ ͖ΔύλʔϯͰ͋Γɼࠇ൪Ͱ͋Ε͹ʮσʯͱ‫ݺ‬͹ΕΔΑ͏ ͳύλʔϯͰ͋Δɽ͜ΕΒΛผͷύλʔϯͱͯ͠ѻ͏ͱͦ Ε͚ͩ৚݅෼‫ذ‬ͷ਺΋૿͑Δ͏͑ɼͳʹΑΓ֘౰͢Δֶश σʔλ͕গͳ͘ͳͬͯ͠·͏ɽͦ͜Ͱɼ͜ΕΒͷύλʔϯ ΛಉҰͷ΋ͷͱͯ͠ѻ͏Α͏ʹͨ͠ɽ۩ମతʹ͸ɼҎԼͷ ༏ઌॱͰύλʔϯʹճసͱ൓సΛՃ͑ɼ8 ௨Γͷύλʔϯ Λ།Ұͷ΋ͷʹஔ͖‫͑׵‬Δɽ. ( 1 ) Ͱ͖Δ͚ͩࣗ෼ͷੴ͕௚ԼʹདྷΔΑ͏ʹ͢Δɽແཧͳ Βఢͷੴ͕௚ԼʹདྷΔΑ͏ʹ͢Δɽ. ( 2 ) ্ͷ৚݅ͷ࣍ʹɼͰ͖Δ͚ͩࣗ෼ͷੴ͕௚ࠨʹདྷΔΑ ͏ʹ͠ɼແཧͳΒఢͷੴ͕௚ࠨʹདྷΔΑ͏ʹ͢Δɽ ਤ 3. σʔλͷॲཧͷྲྀΕͷ֓೦ਤ. ( 3 ) ಉ༷ʹɼࣗ෼ͷੴ͔ఢੴ͕௚ӈʹདྷΔΑ͏ʹ͢Δɽ ( 4 ) ಉ༷ʹɼࣗ෼ͷੴ͔ఢੴ͕ࠨԼʹདྷΔΑ͏ʹ͢Δɽ ( 5 ) ಉ༷ʹɼࣗ෼ͷੴ͔ఢੴ͕ӈԼʹདྷΔΑ͏ʹ͢Δɽ. 5.3 ༧උ࣮‫ݧ‬ͷ݁Ռͱߟ࡯ ‫ػ‬ցֶशͷۤख෼໺Λ‫ݟ‬ΔͨΊͷ༧උ࣮‫ ͯ͠ͱݧ‬5.1 અ Ͱड़΂ͨಛ௃ྔΛ༻͍ͯɼֶशΛߦ͍Ұக཰Λಘͨɽͦͷ ݁Ռ͸. ( 6 ) ಉ༷ʹɼࣗ෼ͷੴ͔ఢੴ͕ࠨ্ʹདྷΔΑ͏ʹ͢Δɽ ਤ 4 ͷྫͰ͋Ε͹ɼ·ͣࣗ෼ͷੴ͕௚ԼʹདྷΔ (b)(e) ͕༏ ઌ͞Εɼଓ͍ͯ௚ࠨʹఢੴ͕དྷΔ (e) ʹ౷Ұ͞ΕΔɽ ͜ͷޮՌ͸ܶతͰɼ5 ˋҎ্ͷҰக཰޲্Λ΋ͨΒͨ͠ɽ. • ୈҰީิ·ͰͷҰக཰ɿ75.3 ˋ. ·ͨɼपғͷύλʔϯͷҰ෦ʹؔ͢Δ‫ܗ‬ͷਖ਼౴཰΋ϚΨ. • ୈೋީิ·ͰͷҰக཰ɿ76.8 ˋ. Ϧɿ75.5 ˋɼσɿ83.6 ˋɼΦγɿ81.9 ˋͱେ෯ʹ޲্ͨ͠ɽ. ͱͳΓɼNomitan ͷग़ྗ݁Ռ 73.7 ˋɼ76.6 ˋΑΓগ͠ྑ. ͳ͓ɼ͜ͷ౷Ұํ๏ͷલʹʮఱ‫ݩ‬ʢ‫ޟ‬൫ͷத৺ʣʹ͍ۙํ. ͍͕ɼਓؒಉ࢜ͷҰக཰ 82.2 ˋɼ87.0 ˋʹ͸·ͩ·ͩ‫ٴ‬͹. ޲Λ্ɼଓ͍ͯӈʹདྷΔΑ͏ʹճసɾ൓స͢Δʯ͜ͱΛࢼ. ͳ͔ͬͨɽ. Έ͕ͨɼ͜Ε͸ 1 ˋఔ౓ͷ޲্ʹͱͲ·ͬͨɽྫ͑͹ʮΦ. Nomitan ͱ‫ػ‬ցֶशͰ‫ͱ͝ܗ‬ͷਖ਼౴཰Λ‫ͯݟ‬ΈΔͱɼද 2. γͱϋΠʯͷΑ͏ʹʢԿઢ͔ͷҙຯͰʣ্͔Լ͔͕ॏཁͳ. ͷΑ͏ʹपғͷύλʔϯʹؔ͢Δ‫ܗ‬͸ Nomitan ͷํ͕Α. ৔߹ʹ͸͜Ε͸༗ӹ͕ͩɼຆͲͷ‫Ͱܗ‬͸ͦΕΑΓ΋ಉҰࢹ. ͘ਖ਼ղ͍ͯ͠Δ͜ͱ͕෼͔ͬͨɽ͜ΕΒͷ‫ ͕ܗ‬Nomitan. ʹΑΔֶशσʔλ਺૿ՃͷԸ‫ܙ‬ͷ΄͏͕େ͖͔ͬͨΑ͏Ͱ. ⓒ 2015 Information Processing Society of Japan. 5.

(7) Vol.2015-GI-33 No.4 2015/3/5. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. ͋Δɽ ͜ΕʹՃ͑ͯɼҎԼͷࡉ͔͍ಛ௃ྔͷఴ࡟ͱύϥϝʔλ ௐ੔ΛߦͬͨɽҰͭҰͭͷߩ‫౓ݙ‬͸Ұக཰ʹͯ͠࠷େͰ. 0.3% ఔ౓Ͱ͋ΓɼֶशσʔλʹΑͬͯ͸ෆཁ·ͨ͸༗֐ ͳมߋ͔΋͠Εͳ͍ɽ. • HeightMy ͷಛ௃ྔΛ࡟আɽ • पғͷੴύλʔϯʹɼR ‫ ͕཭ڑ‬5 ͷ 8 ఺Λ௥Ճɽ • पғͷੴύλʔϯʹɼʢճస‫ޙ‬ͷʣ3 ϚεԼɼ3 Ϛε্ Λ௥Ճ. • ʢճస‫ޙ‬ͷʣࠨԼɼԼɼӈԼɼࠨԼͷԼɼԼͷԼɼӈ ԼͷԼʹ͋Δࣗ෼ͷੴͷ߹‫਺ܭ‬Λ௥Ճ. • J4.8 ͷύϥϝʔλͰ͋Δ Confidence ͷ஋Λ 0.25 ͔ Β 0.1 ʹมߋ. • J4.8 ͷύϥϝʔλͰ͋Δ Subtree Raising Λ False ʹ มߋ ͜ΕΒͷ޻෉ͷ݁Ռ࠷ऴతʹ͸ɼୈҰީิ·ͰͷҰக཰. ਤ 5. ‫ܗ‬ͷग़‫ݱ‬ճ਺ͱਖ਼౴཰. ͕ 82.0 ˋɼୈೋީิ·ͰͷҰக཰ɿ85.4 ˋͱͳͬͨɽ͜ Ε͸ Nomitan ੑೳΛ໌Β͔ʹ্ճΓɼਓؒಉ࢜ͷҰக཰. ͕ͨɼ͜Ε͸ώϥΩρϝͱ͸શ͘ҟͳΔ΋ͷͰɼ͔ͳΓҹ. 82.2 ˋɼ87.0 ˋʹ΋͔ͳΓഭΔੑೳͰ͋Δɽ. ৅͸ѱ͍ɽ͜ΕΛώϥΩρϝͱ൑ఆͯ͠͠·ͬͨཧ༝͸ɼ ʮ3 ઢ͔ 4 ઢʹ͋ΓɼҰ൪͍ۙࣗ෼ͷੴͱ 6 ͷ R ‫཭ڑ‬ɼҰ൪ ͍ۙ૬खͷੴͱ 4 Ҏ্ 6 ҎԼͷ R ‫཭ڑ‬ʯͱ͍͏Α͏ͳϧʔ ϧ͕ώϥΩρϝʹ͍ͭͯ࡞ΒΕͯ͠·͔ͬͨΒ͔΋͠Εͳ ͍ɽຊ౰ͷώϥΩρϝ΋͜ͷ৚݅Λຬͨ͢ɽ࣮ࡍʹ͸ɼҰ ൪͍ۙࣗ෼ͷੴ΍૬खͷੴ΋ 3 ઢ͔ 4 ઢʹͳ͚Ε͹ͳΒͳ ͍ͳͲ௥Ճͷ৚͕݅ඞཁͰ͋Δ͕ɼώϥΩρϝ͸શମͰ 35 ճʢ໿ 0.3 ˋʣ͔͠ग़‫͓ͯ͠ݱ‬Βͣɼͦ͜·Ͱ͸ֶशͰ͖ ͳ͔ͬͨͱ༧૝Ͱ͖Δɽ͜ͷΑ͏ʹɼಛʹग़‫ݱ‬ճ਺ͷগͳ ͍‫͍ͯͭʹܗ‬ɼ ʮ໌Β͔ʹؒҧ͍ͬͯΔʯ໊લΛ‫·ͯͬ͠ݴ‬ ͏ͷ͸‫ࡏݱ‬ͷγεςϜͷ໰୊఺Ͱ͋Δɽ. ਤ 4. ࣄ্࣮ಉ͡഑ஔͷੴɽଟ͘ͷ৔߹ʮσʯʹͳΔɽ. 6.2 ‫ͱ͝ܗ‬ͷਖ਼౴཰ͱɼ‫ࡏݱ‬ͷ໰୊఺ ૯߹తͳҰக཰͸ 82 ˋఔ౓Ͱ͋Δ͕ɼ‫ʹܗ‬Αͬͯಘखෆ ಘख͸͋Δɽਤ 5 ͸ɼԣ࣠Λ‫ع‬ේதͷग़‫ݱ‬ճ਺ʢlogscaleʣ ɼ ॎ࣠Λਖ਼౴཰ʢద߹཰ͱ࠶‫཰ݱ‬ͷฏ‫ۉ‬஋ʣʹͱͬͨ΋ͷͰɼ ਖ਼౴཰ 10 ˋҎԼͷ΋ͷ͔Β 100 ˋͷ΋ͷ·Ͱ෯޿͍ɽશ ମతͳ܏޲ͱͯ͠͸ग़‫ݱ‬ճ਺͕ଟ͍΄Ͳਖ਼౴཰΋ߴ͘ͳ Δ͕ɼಉ͘͡Β͍ͷग़‫ݱ‬ճ਺Ͱ΋্Լʹ͸෯͕͋Δɽྫ͑ ͹੕ɾখ໨ɾ໨ϋζγɾߴ໨ͳͲ͸ۭ͖۱ʹର͢ΔணखͰ ৚͕݅࡞Γ΍͘͢ɼొ৔ճ਺͸͞΄Ͳଟ͘ͳ͍͕΄΅ 100 ˋͷਖ਼౴཰ͱͳ͍ͬͯΔɽ. ਤ 6. ໌Β͔ͳؒҧ͍ͷྫ. ྫ͑͹ɼਤ 6 ͷനͷखʢϓϩ‫ʹ࢜ع‬ΑΕ͹΢νίϛͱ‫ݺ‬ Ϳ΂͖΋ͷʣΛ‫ػ‬ցֶशͰ͸ώϥΩρϝͱ൑ఆͯ͠͠·ͬ. ⓒ 2015 Information Processing Society of Japan. 6.

(8) Vol.2015-GI-33 No.4 2015/3/5. ৘ใॲཧֶձ‫ڀݚ‬ใࠂ IPSJ SIG Technical Report. 6.3 ϓϩ‫ʹ࢜ع‬ΑΔධՁ. ਓؒͷΞϚνϡΞߴஈऀʹ͔ͳΓ͍ۙੑೳΛಘΔ͜ͱ͕Ͱ. ‫ػ‬ցֶशʹΑΔ‫ܗ‬ͷ೔ຊ‫ޠ‬ද‫͕ݱ‬ϓϩ‫࢜ع‬ͷ໨͔Β‫ͯݟ‬. ͖ͨɽ·ͨɼ ʢ2ʣ ʢ3ʣ ʢ4) ͷ਺Λ‫ݟ‬ΔͱɼඍࠩͰ͸͋Δ͕ɼ. ͲΕ͚ͩຬ଍Ͱ͖Δ͔ɼ‫ػ‬ցֶशʹΑΔग़ྗ݁ՌΛࡌͤͨ. ‫ػ‬ցֶश͸ܰඍͳϛε͕গͳ͍ҰํͰલઅͷྫʹ΋͋ΔΑ. ‫ع‬ේΛϓϩ‫࢜ع‬ʢ೔ຊ‫ع‬Ӄ࿡ஈʣʹ‫ͤݟ‬ධՁͯ͠΋Βͬͨɽ. ͏ʹॏେͳϛε͕ଟ͍ͱ͍͏܏޲͕‫ݟ‬ΒΕͨɽ. 3 ষͰड़΂ͨͱ͓ΓɼධՁʹ͸ਖ਼ղ͔ෆਖ਼ղ͔Ͱ͸ͳ͘ຬ. ͳ͓ɼ ʮ͋ΔϨϕϧͷֶशσʔλΛ༻͍͍ͯͨΒɼͦΕҎ ্ͷ݁Ռ͸๬Ίͳ͍ͷͰ͸ͳ͍͔ʯͱ͍͏‫ݒ‬೦͸͜ͷ৔߹. ଍౓Λ༻͍ͨɽ ·ͣɼ4.2 અͰೖྗͯ͠΋Βͬͨ‫ع‬ේ͔Β‫ڞ‬ஶऀΛআ͘ 5. ඞͣ͠΋ਖ਼͘͠ͳ͍ɽΞϚνϡΞͷதʹ͸Ұ෦ͷ‫ܗ‬ͷΈਖ਼. ໊෼ 1 ຕͣͭΛແ࡞ҝʹબͼɼ͏ͪ 3 ຕ͸ॳख͔Β 100 ख. ͘͠‫͍ͳ͑ݴ‬ਓ΋ଟ͘ɼͦͷ‫͕ܗ‬ॏෳ͠ͳ͍‫ݶ‬Γ͸ɼଟ͘. ໨·Ͱɼ2 ຕ͸ 101 ख໨͔Β 200 ख໨·ͰΛ࢒ͯ͠ɼͦΕ. ͷೖྗऀͷσʔλֶ͕श͞ΕΔ͜ͱͰଟ਺ܾతʹਖ਼͍͠‫ܗ‬. Ҏ֎ͷೖྗ͞Εͨ‫ܗ‬͸࡟আͨ͠ɽೖྗ͞Εͨୈೋީิ΋࡟. ͕‫͑ݴ‬ΔΑ͏ʹͳΔ৔߹͕ଟ͍͔ΒͰ͋Δɽखಈͷϧʔϧ. আͨ͠ɽଓ͍ͯɼಉ͡‫ع‬ේ߹‫ ܭ‬500 ख෼ʹ͍ͭͯɼWeka. ௥ՃͳͲಛผͳௐ੔Λࢪ͞ͳ͘ͱ΋ɼֶशσʔλ΍ಛ௃ྔ. ͷ൑ఆ݁ՌΛಉ༷ͷ‫ͳʹࣜܗ‬ΔΑ͏ʹ sgf ϑΝΠϧʹ‫ه‬࿥. ͷ௥ՃʹΑͬͯΑΓߴ͍Ϩϕϧʹ౸ୡ͢ΔՄೳੑ͸͋Δͱ. ͨ͠ɽ. ߟ͑Δɽ. ͦͷ͏͑Ͱɼϓϩ‫͜ʹ࢜ع‬ΕΒͷೖྗऀΛ໌͔ͣ͞ʹධ ՁΛͯ͠΋ΒͬͨɽධՁ߲໨ͱͯ͠͸ɼ֤खΛ. 7. ·ͱΊ ຊߘͰ͸ɼίϯϐϡʔλʹ‫ج‬ຊతͳ‫ܗ‬ͷ೔ຊ‫ޠ‬ද‫ͤ͞ݱ‬. ( 1 ) ࣗ෼Ͱ΋͜͏‫Ϳݺ‬ ( 2 ) ࣗ෼ͳΒผͷ‫͕ͿݺͰܗ‬ɼ͜ΕͰ΋͞΄Ͳ͓͔͘͠͸. Δ͜ͱΛ໨ࢦ͠ɼͦͷͨΊͷख๏ͱͯ͠ɼਓؒͷߴஈऀʹ ೖྗͯ͠΋Βͬͨ‫ہ‬໘ɾखʹରԠ͢Δ‫ͱܗ‬ɼͦΕΒ͔Βந. ͳ͍ɽ. ( 3 ) ͜Ε͸ΘΓͱҧ࿨‫͕͋ײ‬Δɽ. ग़ͨ͠ಛ௃ྔΛ༻͍Δ‫͋ࢣڭ‬ΓֶशΛఏҊͨ͠ɽಛ௃ྔʹ. ( 4 ) ͜Ε͸໌Β͔ʹ͓͔͍͠ɽ. ޻෉ΛՃ͑Δ͜ͱͰɼ‫ܗ‬ͷҰக཰ͱɼϓϩ‫ʹ࢜ع‬ΑΔຬ଍. ͷ 4 ߲໨ʹ෼ྨͯ͠΋Βͬͨɽͦͷ্Ͱɼ֤‫ع‬ේʢ100 खʣ. ౓ධՁͷ૒ํͰɼਓؒͷΞϚνϡΞߴஈऀʹ͔ͳΓ͍ۙੑ. ͝ͱʹɼͦͷ૯߹఺Λग़ͯ͠΋Βͬͨɽ૯߹఺͸ʮ90 ఺ʹ. ೳΛಘΔ͜ͱ͕Ͱ͖ͨɽ. ̣̝̠ͰͷಡΈ্͛ʹ΋࢖͑ΔϨϕϧʯʮ80 ఺ʹΞϚࡾஈ. ܰඍͳϛε͕গͳ͍ҰํͰॏେͳϛε͕ଟ͍՝୊΋͋Δ. ͷձ࿩Ͱ௨༻͢ΔϨϕϧʯʮ70 ఺ʹΞϚ 6 ‫͘ڃ‬Β͍ͱ͍͍. ͕ɼֶशσʔλΛ௥Ճ͢Δ͜ͱͰग़‫ݱ‬ճ਺͕গͳ͔ͬͨ‫ܗ‬. উෛͷϨϕϧʯΛ໨҆ͱͯ͠΋Βͬͨɽ. ͷग़‫ݱ‬ճ਺Λ૿΍ͨ͠ΓɼΑΓྑ͍ಛ௃ྔͷઃ‫ʹͲͳܭ‬. ෼ྨ݁Ռͱ૯߹఺Λද 3 ͱද 4 ʹࣔ͢ɽ(2)(3)(4) ͷྻ. Αͬͯɼࠓ‫͞ޙ‬ΒͳΔੑೳͱຬ଍౓ͷ޲্͕‫ࠐݟ‬ΊΔɽͦ. ͸ऑ͍ҧ࿨‫( ײ‬2) ͔Β‫͍ڧ‬ҧ࿨‫( ײ‬4) ·Ͱ͕ 100 खதʹԿ. ΕʹΑΓίϯϐϡʔλͱͷ‫ײ‬૝ઓɼ‫ݕ‬౼ɼ͓஻Γͷ࣮‫ݱ‬ɼ. ख͔͋ͬͨͦͷճ਺Λࣔ͠ɼগͳ͍΄͏͕ྑ͍݁ՌͱͳΔɽ. ॳ‫ऀڃ‬ͷ஌ࣝఆண΁ͷߩ‫ظ͕ݙ‬଴Ͱ͖Δɽ. ද 3. ΞϚνϡΞͷ‫ܗ‬ೖྗʹର͢Δϓϩ‫࢜ع‬ͷධՁɽҧ࿨‫ײ‬ͷճ਺ ͱ૯߹఺ɽ ‫ع‬ේ. ද 4. ँࣙ. ຊ‫ڀݚ‬ͷҰ෦͸ɼՊֶ‫ڀݚ‬අิॿۚ ‫ج‬൫ C ‫ڀݚ‬. (2). (3). (4). ૯߹఺. ʮਓؒϓϨΠϠΛʠָ͠·ͤΔʡғ‫ޟ‬ϓϩάϥϜͷ‫ڀݚ‬ʯͷ. A. 5. 6. 4. 82. ॿ੒ΛಘͯߦΘΕͨɽ·ͨɼֶशσʔλͷ࠾औɾධՁʹ‫ڠ‬. B. 7. 3. 3. 84. ྗ͍͍ͨͩͨۚ୔େֶғ‫ޟ‬෦ɼ೔ຊ‫ع‬Ӄ‫͢ँਂʹ࢜ع‬Δɽ. C. 3. 0. 3. 91. D. 2. 2. 5. 86. E. 10. 8. 5. 80. ฏ‫ۉ‬. 5.4. 3.8. 4.0. 84.6. ‫ػ‬ցֶशͷ‫ܗ‬ͷ෼ྨʹର͢Δϓϩ‫࢜ع‬ͷධՁɽҧ࿨‫ײ‬ͷճ਺ ͱ૯߹఺ɽ ‫ع‬ේ. (2). (3). (4). ૯߹఺. A. 4. 4. 5. 83. B. 4. 5. 3. 85. C. 4. 1. 4. 88. D. 2. 4. 2. 90. E. 8. 4. 9. 73. ฏ‫ۉ‬. 4.4. 3.6. 4.6. 83.8. ࢀߟจ‫ݙ‬ [1]. [2]. [3] [4] [5] [6]. [7]. ‫ػ‬ցֶश͸ΞϚνϡΞߴஈऀͷฏ‫ۉ‬૯߹఺ʹ 0.8 ఺͚ͩ ྼ͍ͬͯΔɽલઅͷҰக཰ͷΈͳΒͣɼຬ଍౓ʹ͓͍ͯ΋. ⓒ 2015 Information Processing Society of Japan. [8]. ஑ాɹ৺ɼSimon ViennotɼϞϯςΧϧϩ‫͚͓ʹޟ‬Δଟ༷ ͳઓུͷԋग़ͱ‫ܗ‬੎ͷ੍‫ޚ‬ɹʙ઀଴‫ ޟ‬AI ʹ޲͚ͯɼGame Programming WorkShopɼ2013 IEEE-CIG (Computer Intelligence and Games) Competitions, http://geneura.ugr.es/cig2012/competitions.html http://batora1992.blog.fc2.com/blog-entry-17.html C.M. Ϗγϣοϓ, ύλʔϯೝࣝͱ‫ػ‬ցֶश, Springer, 2007. http://www.ruijiang.com/multigo/ JAIST CUP 2012 ήʔϜΞϧΰϦζϜେձ ғ‫ ޟ‬9 ࿏൫ʮ઀଴‫ޟ‬ʯίϯςε τ, http://www.jaist.ac.jp/jaistcup/2012/jc/9ro.html Remi Coulom, Computing Elo Ratings of Move Patterns in the Game of Go, ICGA Workshop, 2007 Quinlan, J. R. C4.5: Programs for Machine Learning. Morgan Kaufmann Publishers, 1993. 7.

(9)