対戦ゲームに学ぶ、フレームワークの設計技法とAIのアルゴリズム入門 / game model

by qsona

Slide 1

Slide 1 text

ରઓήʔϜʹֶͿ  ϑϨʔϜϫʔΫͷઃܭٕ๏ ͱAIͷΞϧΰϦζϜೖ໳ builderscon tokyo 2019 (2019-08-30) @qsona

Slide 2

Slide 2 text

whoami • @qsona • ιϑτ΢ΣΞΤϯδχΞ • ରઓήʔϜ޷͖ • কعΞϚνϡΞ࢛ʙޒஈ͘Β͍ • ήʔϜAIΛ͔ͬͨ͜͡ͱ͋Γ

Slide 3

Slide 3 text

ࠓ೔ͷ࿩ • ରઓήʔϜ͸ָ͍ͧ͠ • ରઓήʔϜʹؔ࿈͢Δٕज़ͷ࿩Λ͢Δ • ॳڃऀ޲͚ • ٕज़ͱ࣮ੈքͷήʔϜͱͷܨ͕ΓΛײͯ͡΋ Β͑ͨΒخ͍͠

Slide 4

Slide 4 text

ࠓ೔ͷ୊ࡐ͸ओʹ͜ΕΒͰ͢ • ࢦͷήʔϜ / কع / ຑ੃ / ΀Α΀Α / ਓ࿛  ͦΕͧΕҧ͏ੑ࣭Λ࣋ͭήʔϜ • ରઓήʔϜΛϞσϦϯά͢Δ͜ͱͰɺ  ήʔϜͷੑ࣭Λଊ͑ͯΈΔ • ϞσϦϯάΛੜ͔͠ɺήʔϜͷϓϩάϥϛϯά΍  AIΞϧΰϦζϜΛߟ͑Δ

Slide 5

Slide 5 text

Agenda • 1. ରઓήʔϜͷϞσϦϯά • 2. ରઓήʔϜAIͷΞϧΰϦζϜೖ໳ • 3. boardgame.io ʹֶͿ  ϑϨʔϜϫʔΫઃܭͷצॴ

Slide 6

Slide 6 text

1. ରઓήʔϜͷϞσϦϯά • 3ͭͷ؍఺ͰϞσϦϯά͢Δ • ήʔϜͷϧʔϧ • ήʔϜͷछྨͷ෼ྨ • ήʔϜͷঢ়ଶͷਐߦ (ήʔϜ໦) • ͦΕͧΕͷ؍఺ͰͷϞσϧͱɺ  ࣮ੈքʹ͓͚Δਓؒͷଊ͑ํͱͷؔ࿈Λߟ͑Δ

Slide 7

Slide 7 text

ରઓήʔϜͷϧʔϧͷఆࣜԽ • ήʔϜͷঢ়ଶͷू߹, ॳظঢ়ଶ • Move • ϓϨΠϠʔ͕ήʔϜΛਐΊΔखஈ • ঢ়ଶ͔Βঢ়ଶ΁ͷભҠ • ऴྃ৚݅ͱ݁Ռ

Slide 8

Slide 8 text

࣮ੈքͰͷఆࣜԽ ྫ: কعͷϧʔϧ • ࣮ࡍͷήʔϜͷϧʔϧ΋ɺ  ͦͷ3ͭͰߏ੒͞Ε͍ͯΔͷ? • কعΨΠυϒοΫ  (೔ຊকع࿈ໍ, 2003)  ΛಡΈղ͍ͯݕূ͢Δ • ϧʔϧ͸17߲໨ʹ෼͔Ε͍ͯΔ

Slide 9

Slide 9 text

• 1. ରہ • 2. ରہऀ • 3. ۨΛॳܕʹ഑ஔ • 4. ۨͷଘࡏͱͦͷۭؒ • 5. ֤ۭؒͷۨͷଘࡏ਺ • 6. ۨͷಈ͖ͱੑ࣭ • 7. ࢦ͠ख • 8. ࢦ͠खͷछྨ • 9. औΔ • 10. ੒Δ • 11. ۨͷؼଐ • 12. ࢦ͠खͷ׬ྃ • 13. ہ໘ͷఆٛ • 14. ൫্ͷۨͱ࣋ͪۨͷؔ܎ • 15. Ԧखͱ٧Έ • 16. ಛʹܾΊΒΕͨېࢭࣄ߲ • 17. ऴہ

Slide 10

Slide 10 text

Slide 11

Slide 11 text

ہ໘ (≒ήʔϜͷঢ়ଶͷू߹) • কع൫ = 9x9 • ֤Ϛεʹ͸ɺઌख·ͨ͸ޙखͷ͕ۨ  ࠷େ1ͭೖΔ • ۨ୆ɾ࣋ۨ • ઌखɾޙख1ͭͣͭ

Slide 12

Slide 12 text

ہ໘ (≒ήʔϜͷঢ়ଶͷू߹) • ൫໘ (൫্+ۨ), ࣋ۨ (ۨ୆+ۨ), ख൪ Λ߹Θͤ ͯʮہ໘ʯͱఆٛ͞Ε͍ͯΔ • ہ໘͸ɺকعʹ͓͚ΔήʔϜͷঢ়ଶ • ਖ਼֬ʹ͸ɺہ໘୯ମ͚ͩͰ͸ήʔϜͷঢ়ଶΛ දͤͳ͍ (ઍ೔ख: ϧʔϓΛ൑ఆͰ͖ͳ͍)

Slide 13

Slide 13 text

ॳܗ (ॳظঢ়ଶ)

Slide 14

Slide 14 text

Slide 15

Slide 15 text

ࢦ͠ख (Move) • ൫্ͷۨΛಈ͔͢ or ۨ୆ͷۨΛ൫্ʹଧͭ • ൫্ͷۨΛಈ͔͢৔߹ɺ͞Βʹ࣍ͷಈ࡞͕Մ ೳͳ৔߹͕͋Δ • (૬खͷۨΛ) औΔ • ੒Δ (ۨΛཪฦ͢)

Slide 16

Slide 16 text

߹๏खͱې͡ख • কعͷࢦ͠खʹ͸ɺ߹๏ख (Valid Move)ͱ  ې͡ख (Invalid Move) ͕͋Δ • ې͡ख͸ࢦ͠खͷू߹ʹؚ·ΕΔ͕ɺ  ͦΕΛࢦ͢ͱʮ൓ଇʯͷऴྃ৚݅ʹΑΓ  ήʔϜ͕ऴྃ͢Δ

Slide 17

Slide 17 text

ې͡खͷྫ • ਖ਼͘͠ͳ͍ۨͷಈ͔͠ํ • ೋา, ଧͪา٧Ί, etc... • ԦΛ૬खʹऔΒΕͯ͠·͏ঢ়ଶʹ͢Δ • ※ԦΛ࣮ࡍʹऔΔ͜ͱ͸ى͜Βͳ͍ɻ  औΕΔ࣌఺Ͱ૬ख͕൓ଇෛ͚

Slide 18

Slide 18 text

Slide 19

Slide 19 text

ऴہ (ऴྃ৚݅ͱ݁Ռ) • ԦΛ٧·͢ (٧·ͤͨํ͕উͪ) • ౤ྃ (౤ྃΛએݴͨ͠ํ͕ෛ͚) • ૬ೖۄ (఺਺Ͱউഊ͕ܾ·Δ) • ઍ೔ख (Ҿ͖෼͚) • ൓ଇ (ېࢭࣄ߲ɻ൓ଇΛͨ͠΄͏͕ෛ͚)

Slide 20

Slide 20 text

Ԧखͱ٧Έ • ࣍ʹ૬खͷԦΛऔΕΔঢ় ଶ͕ʮԦखʯ • ԦखͰ͋Γɺ͞Βʹ૬ख ͕Ͳ͏΍ͬͯ΋࣍ʹԦΛ औΕΔঢ়ଶ͕ʮ٧Έʯ • ٧Έͷ৔߹ɺԦखΛ͔͚ ͍ͯΔଆ͕উͪ

Slide 21

Slide 21 text

ଧͪา٧Ί • ൓ଇͷҰछɻ  าΛଧͬͯʮ٧Έʯʹͯ͠͸͍͚ͳ͍

Slide 22

Slide 22 text

"ଧͪา٧Ί" ͸ϝλతϧʔϧ • "٧Έ" = Ԧख͕͔͔͓ͬͯΓɺ͔ͭ૬खʹ  Valid Move ͕ଘࡏ͠ͳ͍ঢ়ଶ • าΛଧͬͯԦख͢Δ ͱ͍͏ Move ͸ɺ  ͦͷޙ૬खʹ Valid Move ͕ଘࡏ͠ͳ͍ͱ͖  Invalid Move Ͱ͋Δ

Slide 23

Slide 23 text

ઍ೔ख • 4ճಉ͡ہ໘͕ൃੜ͢Δ͜ͱ • Ҿ͖෼͚ • Ұൠతʹ͸ઌखͱޙखΛೖΕସ͑ͯ΍Γ௚͢ • ͨͩ͠ɺ͜ͷؒยํͷԦख͕ଓ͍͍ͯΔ  ʮ࿈ଓԦखͷઍ೔खʯ͸ɺ  ԦखΛ͔͚͍ͯΔํ͕൓ଇෛ͚

Slide 24

Slide 24 text

ʮ࠷ޙͷ৹൑ʯ(1997) • ๓ాޫ࢘ ࢯʹΑΔ  ٧কع • ٧কع ... Ԧखͷ࿈ଓ Ͱ૬खͷԦ༷Λ٧Ί Δύζϧ • কعͷϧʔϧͷໃ६Λ ಥ͍ͨ࡞

Slide 25

Slide 25 text

No content

Slide 26

Slide 26 text

No content

Slide 27

Slide 27 text

No content

Slide 28

Slide 28 text

தུ

Slide 29

Slide 29 text

No content

Slide 30

Slide 30 text

No content

Slide 31

Slide 31 text

No content

Slide 32

Slide 32 text

No content

Slide 33

Slide 33 text

3ճ܁Γฦ͢

Slide 34

Slide 34 text

No content

Slide 35

Slide 35 text

No content

Slide 36

Slide 36 text

No content

Slide 37

Slide 37 text

࿈ଓԦखͷઍ೔ख • ઍ೔ख ... 4ճಉ͡ہ໘͕ൃੜ͢Δ͜ͱ • ͜ͷہ໘͸ʮ࿈ଓԦखͷઍ೔खʯʹΑΓ  ઌख(ԦखΛ͔͚͍ͯΔํ)͕൓ଇෛ͚ • ͱ͍͏͜ͱ͸ɺઌख͕ҧ͏खΛࢦ͞ͳ͚Ε͹ ͳΒͳ͍?

Slide 38

Slide 38 text

No content

Slide 39

Slide 39 text

ଞʹख͕ͳ͍ • าͰԦख͕͔͔͍ͬͯΔ • ֯ͰऔΔҎ֎ͷબ୒ࢶ͕ͳ͍ • ͔͠͠ɺ֯Ͱऔͬͯ΋൓ଇෛ͚ • ͭ·Γ͜ͷہ໘͸ "٧Έ" Ͱ͸......?

Slide 40

Slide 40 text

าͷԦखͰ٧Μͩ • ͱ͍͏͜ͱ͸͜ͷہ໘͸ʮଧͪา٧Ίʯ • (࠶ܝ) าΛଧͬͯԦख͢Δ ͱ͍͏ Move ͸ɺ  ͦͷޙ૬खʹ Valid Move ͕ଘࡏ͠ͳ͍ͱ͖  Invalid Move Ͱ͋Δ • ͭ·Γ͜ͷา͸ଧͯͳ͍ɺͱ͍͏ओு

Slide 41

Slide 41 text

No content

Slide 42

Slide 42 text

͜ͷ٧কع͸੒ཱ͢Δ͔? • ͜ͷาଧͪ͸ʮଧͪา٧Ίʯͳͷ͔Ͳ͏͔ • ௚ײʹ͸͔ͳΓ൓͢Δɻ • ٧Έͱ͍͑͹ී௨͸  ʮͲ͏ࢦͯ͠΋࣍ʹԦ͕औΒΕΔʯ͜ͱ • ͜Μͳ͜ͱ͕ى͖Δ͜ͱ͸୭΋ߟ͑ͳ͔ͬͨ

Slide 43

Slide 43 text

͜ͷ٧কع͸੒ཱ͢Δ͔? • ݱߦͷϧʔϧͰ͸ɺউͪෛ͚͕ʮෆఆʯ  ͱ͍͏ͷ͕ఆઆ • ϧʔϧ͕ᐆດɺਖ਼͘͠ఆࣜԽͰ͖͍ͯͳ͍ • ࢲݟ: কعΨΠυϒοΫ (2003) Λਫ਼ಡ͢Δͱ... • ଓ͖͸WebͰ

Slide 44

Slide 44 text

ྫ: ਓ࿛ • ਓ࿛ͷϧʔϧΛͦͷ··ద༻͢Δͱɺࣗવݴޠղੳ͕ඞཁ ʹͳΓɺղੳ͕೉͍͠ • ʮਓ࿛ϓϩτίϧʯ... ֤ϓϨΠϠʔͷൃݴΛఆࣜԽ (ࣄ࣮΍ ҙਤͷද໌, ཁ੥ͳͲ) • ྫ: REQUEST [○͞Μ] (DIVINATION [×͞Μ])  →ʮ○͞Μɺ×͞ΜΛ઎͍·͠ΐ͏ʯ • ͜ΕʹΑΓݚڀର৅ʹ͠΍͘͢ͳΔ

Slide 45

Slide 45 text

ϧʔϧΛఆࣜԽ͢Δͱ • ʮ࠷ޙͷ৹൑ʯͷΑ͏ͳέʔεΛਖ਼͘͠൑ఆͨ͠Γɺ  ໃ६Λൃݟͨ͠ΓͰ͖Δ (͸ͣ) • ήʔϜ AI ͳͲͷݚڀର৅ʹͰ͖Δ • ಛʹ General Game Playing ... ະ஌ͷήʔϜΛ͏· ͘ϓϨΠͰ͖Δ൚༻ AI ͷର৅ʹ • ήʔϜࣗମͷੑ࣭ͷཧղʹܨ͕Δ

Slide 46

Slide 46 text

2. ରઓήʔϜͷ෼ྨ

Slide 47

Slide 47 text

جຊͷ෼ྨ߲໨ • ϓϨΠਓ਺ (2ਓʙ) • རಘͷ߹ܭ͸Ұఆ(θϩ)͔ • ঢ়ଶ͕ແݶʹଘࡏ͢Δ͔ • ϥϯμϜωε(ཚ਺)͸ଘࡏ͢Δ͔ • ͢΂ͯͷ৘ใ͕શһʹ։ࣔ͞Ε͍ͯΔ͔

Slide 48

Slide 48 text

ೋਓྵ࿨༗ݶ֬ఆ׬શ৘ใήʔϜ • কع, ғޟ, Φηϩ, νΣε, ... ͳͲ • ήʔϜAIݚڀͰ͸جຊͱͳΔΫϥελ ਓ਺ ਓ ঢ়ଶ ༗ݶݸ ϥϯμϜωε ͳ͠ ৘ใ શͯ։ࣔ

Slide 49

Slide 49 text

ྫ: ຑ੃ • ଟਓ਺, ϥϯμϜωε͋Γ, ෆ׬શ৘ใ ਓ਺ ਓ ঢ়ଶ ༗ݶݸ ϥϯμϜωε ͋Γ ہ͝ͱͷॳظঢ়ଶ͸ϥϯμϜ ৘ใ ࣗ෼ͷखṛ )BOE ͸ࣗ෼͚ͩʹ։ࣔ  ࢁṛ %FDL ͸શһʹඇ։ࣔ

Slide 50

Slide 50 text

ྫ: ΀Α΀Α • ϓϨΠϠʔ͸೚ҙͷλΠϛϯάͰૢ࡞Մೳͳ ͷͰɺ࿈ଓແݶͷہ໘͕͋Δ͕ɺ࣮࣭తʹ͸ ཭ࢄεςοϓͱଊ͑Δ͜ͱ͕Մೳ ਓ਺ ਓ ঢ়ଶ ࿈ଓແݶ࣮࣭తʹ͸༗ݶ ϥϯμϜωε ͋Γ ৘ใ ΄΅׬શ৘ใ खઌҎ্ͷ΀Α͸දࣔ͞Εͳ͍

Slide 51

Slide 51 text

ྫ: ਓ࿛ • ෼ྨ্͸͔ͳΓಛघͳήʔϜͱݴ͑Δ ਓ਺ ଟਓ਺ɾνʔϜ ঢ়ଶ ࿈ଓແݶϓϩτίϧͷ੍ݶʹΑΓ༗ݶ ϥϯμϜωε ॳظঢ়ଶҎ֎͸΄΅ͳ͠ ৘ใ ෆ׬શ৘ใ

Slide 52

Slide 52 text

ଟਓ਺ͷ೉͠͞ • ࣗ෼ͷߦಈ͕ଞਓͷউഊʹӨڹͯ͠͠·͏͜ ͱ͕͋Δ • શһ͕े෼ʹϨϕϧ͕ߴ͘ɺͦΕͧΕ͕ࣗ෼ ͷউͪͷͨΊʹߦಈ͢ΔͱԾఆͯ͠Α͍ͷͩ Ζ͏͔?

Slide 53

Slide 53 text

ϥϯμϜωεͷ೉͠͞ • ࠷ऴ݁Ռ͚͔ͩΒֶश͠Α͏ͱ͢Δͱ  ऩଋʹ͔͔࣌ؒΓ͗͢Δ • ≒ ݁ՌʹҰتҰ༕͢Δਓ͸ڧ͘ͳΒͳ͍ • ֬཰Ͱߟ͑Δඞཁ͕͋Δ • ྫ: "ṛޮ཰" ... ΞΨΕΔ֬཰͕ߴ͍ํ๏Λ୳͢

Slide 54

Slide 54 text

ෆ׬શ৘ใͷ೉͠͞ • ৚݅෇͖֬཰Λߟ͑Δඞཁ͕͋Δ • ྫ: ࣺͯṛ͔Β૬खͷ଴ͪΛਪଌ͢Δ • ૬खʹखͷ಺Λಡ·Εͳ͍Α͏ɺ  ࠞ߹ઓུ͕ඞཁʹͳΔ৔߹͕͋Δ

Slide 55

Slide 55 text

θϩ࿨ରઓήʔϜͷ෼ྨ·ͱΊ • ೋਓɾ༗ݶɾ֬ఆɾ׬શ৘ใ͕  ΋ͬͱ΋جຊͷΫϥελ • ͦΕͧΕҳ୤͢Δͱ৽ͨͳෳࡶ͕͞ੜ·ΕΔ

Slide 56

Slide 56 text

3. ήʔϜ໦  (ήʔϜͷਐߦͷϞσϦϯά)

Slide 57

Slide 57 text

ήʔϜͷਐߦͱ໦ߏ଄ • ॳظঢ়ଶ͔ΒɺMoveʹΑͬͯήʔϜͷঢ়ଶ͕ ॱʑʹҠΓมΘ͍ͬͯ͘ • ঢ়ଶ = ϊʔυ  Move = ࢬ  ͱΈͳͨ͠໦ߏ଄ͰදͤΔ

Slide 58

Slide 58 text

ઌख ޙख ઌख ޙख

Slide 59

Slide 59 text

ήʔϜ໦ͷϊʔυͷ஋ • ೋਓήʔϜ໦ͷ৔߹ • ࣗ෼ͷख൪ => max(ࢠϊʔυͷ஋) • ૬खͷख൪ => min(ࢠϊʔυͷ஋)

Slide 60

Slide 60 text

ઌख ޙख ઌख ޙख উͪ ෛ͚ Ҿ෼ ෛ͚ ઌखࢹ఺ ͷউഊ

Slide 61

Slide 61 text

ઌख ޙख ઌख ޙख উͪ ෛ͚ Ҿ෼ ෛ͚ ઌखࢹ఺ ͷউഊ Ҿ෼ উͪ

Slide 62

Slide 62 text

ઌख ޙख ઌख ޙख উͪ ෛ͚ Ҿ෼ ෛ͚ ઌखࢹ఺ ͷউഊ Ҿ෼ উͪ Ҿ෼ Ҿ෼

Slide 63

Slide 63 text

ήʔϜ໦ͱήʔϜͷੑ࣭ • ήʔϜ໦͕Ͳ͏ͳ͍ͬͯΔ͔͸ɺਓؒͷײ͡ ํͱϦϯΫ͍ͯ͠Δ͸ͣ • ྫ͑͹  ਓ͕ؒʮ໘ന͍ʯͱײ͡ΔήʔϜΛɺ  ήʔϜ໦ͷঢ়ଶͰදͤͳ͍ͩΖ͏͔?

Slide 64

Slide 64 text

ࢲݟ: ໘ന͍ήʔϜ • ϧʔϧ͕γϯϓϧ • ਂΈ͕͋Δ • ۓுײ͕࣋ଓ͢Δ  (ద౓ʹٯస͕ى͜Γ΍͍͢)

Slide 65

Slide 65 text

ήʔϜ໦ ઌख

Slide 66

Slide 66 text

ফԽࢼ߹ͷήʔϜ໦ ෛ͚ ෛ͚ ෛ͚ উͪ উͪ উͪ উͪ উͪ ઌख

Slide 67

Slide 67 text

ফԽࢼ߹ͷήʔϜ໦ • ࣗ෼͕ͲΕΛબΜͰ΋উͭ • ฆΕ͕গͳ͍

Slide 68

Slide 68 text

࠷ޙ·Ͱ೤͍ࢼ߹ͷήʔϜ໦ ෛ͚ ෛ͚ ෛ͚উͪ ෛ͚ উͪ ෛ͚ উͪ উͪ ઌख

Slide 69

Slide 69 text

࠷ޙ·Ͱ೤͍ࢼ߹ͷήʔϜ໦ • ࣗ෼͕બΜͰউͯΔख͸Ұ͚ͭͩ • উͪͷہ໘͕ଟ͍͕ɺ  ૬ख͕ਖ਼֬ʹ΍Δͱෛ͚Δख͕͋Δ • ٯసͷՄೳੑ͕͋Δ

Slide 70

Slide 70 text

ήʔϜͷ໘ന͞ͷఆࣜԽ • ଞʹ΋໘ന͞ͷई౓͸͋Δ͕ɺଟ͘͸  ήʔϜ໦ͷঢ়ଶͰදݱͰ͖ͦ͏ • ໘ന͞ΛఆࣜԽͰ͖Ε͹  ʮࣗಈήʔϜ࡞੒AIʯΛ࡞ΕΔ͔΋͠Εͳ͍

Slide 71

Slide 71 text

1ষ·ͱΊ • ରઓήʔϜΛ3ͭͷ؍఺ͰϞσϦϯάͨ͠ • ͦΕͧΕɺ࣮ੈքʹ͓͚Δ֓೦ͱͷؔ܎Λࣔ ͨ͠ • ࣍ͷষ͔Β͜ͷϞσϧΛ  AI΍ήʔϜϑϨʔϜϫʔΫʹద༻͍ͯ͘͠

Slide 72

Slide 72 text

2. ରઓήʔϜͷղੳͱ  AIͷΞϧΰϦζϜೖ໳

Slide 73

Slide 73 text

ରઓήʔϜAIͷྺ࢙ • ݚڀ͸1950೥͝Ζ͔Βଘࡏ • von Neumann (ϑΥϯɾϊΠϚϯ) • C. E. Shannon (γϟϊϯ) • Alan Turing (νϡʔϦϯά) • ௒߽՚

Slide 74

Slide 74 text

ରઓήʔϜAIͷྺ࢙ • 1993೥ 6x6Φηϩͷ׬શղੳ(ޙखউͪɺ16-20) • 1997೥ νΣεͰੈքνϟϯϐΦϯʹউར • 2007೥ νΣοΧʔͷ׬શղੳ(Ҿ͖෼͚) • 2013೥ কعͰτοϓϨϕϧͷϓϩع࢜ʹউར • 2016೥ ғޟͰτοϓϨϕϧͷϓϩع࢜ʹউར

Slide 75

Slide 75 text

ରઓήʔϜAIͷجຊ • ήʔϜ໦Λϕʔεʹ୳ࡧ͢Δ • ϛχϚοΫε๏, ࢬמΓ (α-β๏) • ϞϯςΧϧϩ໦୳ࡧ • ධՁؔ਺Λ࡞Δ • ήʔϜͷঢ়ଶΛείΞͰදؔ͢਺

Slide 76

Slide 76 text

ࢀߟ: কعʹ͓͚Δਓؒͷઓུ ͱͷൺֱ • ήʔϜ໦Λϕʔεʹ୳ࡧ͢Δ (ಡΈ) • ͋Γ͑ͦ͏ͳࢦ͠खΛ͍͔ͭ͘બΜͰಡΉ (બ୒త୳ࡧ) • AI Ͱ͸શ෯୳ࡧ͕ओྲྀ

Slide 77

Slide 77 text

ࢀߟ: কعʹ͓͚Δਓؒͷઓུ ͱͷൺֱ • ධՁؔ਺ (ܗ੎൑அ, େہ؍) • ඇৗʹ೉͘͠ɺτοϓϓϩಉ࢜Ͱ΋ҙݟ͕ ৯͍ҧ͏ • ७ਮͳධՁؔ਺(੩తͳہ໘ͷධՁ)Ͱ͸  ਓؒ > AI

Slide 78

Slide 78 text

୳ࡧ(1)  ϛχϚοΫε๏ / ࢬמΓ(α-β๏) • ήʔϜ໦ͷͱ͜ΖͰઆ໌ͨ͠ͷ͕  ϛχϚοΫε๏ • ࢬמΓ ... ಡ·ͳͯ͘ྑ͍ࢬΛΧοτ͢Δ • ྫ: ͋ΔMoveͰ10఺औΕΔͷ͕֬ఆ͍ͯ͠Δɻ => 10఺ҎԼ͕֬ఆͨ͠ࢬ͸ಡ·ͳͯ͘Α͍

Slide 79

Slide 79 text

ࢬמΓͷྫ ઌख

Slide 80

Slide 80 text

ࢬמΓͷྫ ઌख ? ͸ಡ·ͳͯ͘ྑ͍

Slide 81

Slide 81 text

୳ࡧ (2) ϞϯςΧϧϩ໦୳ࡧ • ͱΓ͋͑ͣɺήʔϜऴྃ·Ͱ΍ͬͯΈΔͷΛ  Կճ΋܁Γฦ͢ • উ཰͕ߴ͍खΛબͿ • ͨͩ͠ɺ୯ʹϥϯμϜʹબͿ͚ͩͩͱऑ͍ͷ Ͱɺ޻෉Λ͢Δ

Slide 82

Slide 82 text

ϞϯςΧϧϩ໦୳ࡧͷվྑ • Multi-Armed Bandit Problem ʹର͢ΔΞϧΰϦ ζϜΛԠ༻͢Δ (UCB1 ΞϧΰϦζϜͳͲ) • ͲΕ͘Β͍౰ͨΔ͔ෆ໌ͳεϩοτϚγϯ͕ͨ ͘͞Μ͋Δͱ͖ɺͲ͏͍͏ઓུͰ΍Δͱྑ͍͔

Slide 83

Slide 83 text

ධՁؔ਺ • ྫ: ΦηϩͰ۱ΛऔΔͱධՁ͕ߴ͍ • ػցֶशΛ࢖͏ͷ͕ओྲྀ • ಛ௃ྔͷநग़͸ਓ͕ؒߦ͏ • General Game Playing ʹ͓͍ͯ͸ɺಛ௃ྔ ͢Β΋ AI ͕ࣗ਎Ͱநग़͢Δ

Slide 84

Slide 84 text

͔͜͜Β࣮ࡍͷήʔϜͰͷྫ

Slide 85

Slide 85 text

ྫ: ࢦͷήʔϜ • ہ໘਺͕গͳ͍ • ͔͔ͨͩ 5^4 * 2 = 1250 ௨Γ • ରশੑ౳Λߟ͑ͯ (5+4+3+2+1) * 2 - 1 = 224௨Γ • ͦͷ͏ͪ14௨Γ͸ऴྃہ໘ • ׬શղੳ͕Մೳ

Slide 86

Slide 86 text

ૉ௚ͳղੳํ๏ • ॳظঢ়ଶ͔ΒήʔϜ໦Λల։͠ͳ͕Βܭࢉ • ϛχϚοΫε๏ • ಈతܭը๏ ... Ұ౓ܭࢉͨ͠ہ໘ͷ݁ՌΛอ ଘ͓͖ͯ͠ɺ࠶ར༻͢Δ • ໰୊: ήʔϜ໦ʹϧʔϓ͕ଘࡏ͢Δ

Slide 87

Slide 87 text

ޙୀղੳ • ऴྃͷہ໘͔Βٯࢉ͠ɺউͪ/ෛ͚ͷہ໘Λ֬ఆ͍ͤͯ͘͞ • ෛ͚ہ໘͔Β1ख໭ͬͨہ໘͸উͪہ໘ • উͪہ໘͔Β1ख໭ͬͨہ໘͔ΒɺࢬΛফ͢ɻࢬ͕0ຊʹ ͳ͍ͬͯͨΒͦͷہ໘͸ෛ͚ہ໘ • ֬ఆ͍ͯ͠Δہ໘͕૿͑ͳ͘ͳͬͨΒɺ  ࢒Γͷہ໘͸Ҿ͖෼͚

Slide 88

Slide 88 text

ྫ: কع • ׬શղੳ͸ෆՄೳ • ہ໘਺͕ଟ͗͢Δ (10^68 ఔ౓ͱਪଌ͞ΕΔ) • ήʔϜ໦ͰҰఆͷਂ͞·Ͱ୳ࡧ͠ɺ  ہ໘ΛධՁؔ਺ͰධՁ͢Δ • ධՁؔ਺͕ॏཁ

Slide 89

Slide 89 text

কعͷධՁؔ਺ • ੲ͸ϧʔϧϕʔεͩͬͨ • ։ൃऀͷ౒ྗͷ݁থ... • عྗ͸ΞϚνϡΞ༗ஈϨϕϧ  (12ࡀͷqsonaͱಉఔ౓. ϓϩ͸·ͣෛ͚ͳ͍) • ͜ͷධՁؔ਺ʹϒϨΠΫεϧʔ͕ى͖Δ

Slide 90

Slide 90 text

Ϙφϯβϝιου • Bonanza (2005, อ໦๜ਔ ࢯ) • ϓϩع࢜ͷعේΛڭࢣσʔλͱͯ͠ɺ  ػցֶशʹΑΓධՁؔ਺Λ࡞੒ • ύϥϝʔλ͸ɺۨͷՁ஋΍ɺ  3ۨͷҐஔؔ܎ͳͲ

Slide 91

Slide 91 text

কعͷධՁؔ਺ͦͷޙ • ࣗ෼ಉ࢜Λରઓֶͤͯ͞श͢Δ (ڧԽֶश) • ϓϩͷعේ౳Λશ͘࢖Θͣɺθϩ͔ΒࣗݾରઓͰ ڧ͘͢ΔࢼΈ΋͋Δ • AlphaZero ͕༗໊ • ͜ͷํ๏Ͱ΋ਓؒΛ௒͑ΔϨϕϧʹͳ͍ͬͯΔ

Slide 92

Slide 92 text

ͪͳΈʹ: ׬શղੳͱධՁؔ਺ • ׬શղੳʹ͓͍ͯ΋ධՁؔ਺͸ॏཁ • (݁ہશ෦ಡΉͳΒɺঢ়ଶΛධՁ͢Δඞཁ͕ͳͦ͞͏ʹࢥ͑Δ͕...) • ྑͦ͞͏ͳMove͔ΒಡΉ͜ͱͰ  ୳ࡧ͢Δہ໘ΛݮΒͤΔ • શہ໘਺ n => ୳ࡧہ໘਺ ໿ √n

Slide 93

Slide 93 text

ྫ: ғޟ • ғޟͷAI͸͠͹Βͣͬ͘ͱऑ͔ͬͨ • 2006೥, ϞϯςΧϧϩ໦୳ࡧ + UCB1 ͷొ৔ʹ ΑΓɺಥવΞϚνϡΞߴஈҐϨϕϧʹ • ΊͪΌͪ͘Ό෼͔Γ΍͍͢ࢀߟࢿྉ͕͋ΔͷͰ  ஔ͍͓͖ͯ·͢

Slide 94

Slide 94 text

ྫ: ΀Α΀Α • ࣮ػͰਓؒʹউͬͨྫ͋Γ • (࣮ը໘Λը૾ೝࣝ) • AI ಉ࢜ͷରઓେձͷه࿥΋͋Δ • ϨΪϡϨʔγϣϯ͕೉͍͠ • ൛ݖͷ໰୊, ϓϥοτϑΥʔϜ͕ͳ͍ • ਓؒͷೝࣝೳྗͷݶք (কع౳ʹൺ΂ͯAIѹ౗త༗ར)

Slide 95

Slide 95 text

ྫ: ਓ࿛ • ʮਓ࿛஌ೳϓϩδΣΫτʯ • ਓ࿛ϓϩτίϧͷ΋ͱͰAIಉ࢜ରઓ͢Δ  ϓϥοτϑΥʔϜ͕ଘࡏ͢Δ • ৽͍͠λΠϓͷήʔϜɻ  ͜Ε͔Βݚڀ͕ൃల͍ͯ͘͠෼໺

Slide 96

Slide 96 text

ରઓήʔϜAIͷ͜Ε͔Β • ݚڀʹ͓͍ͯ͸ɺ͋ΔಛఆͷήʔϜAIΛݚڀ͢Δ͜ ͱͰଞͷ෼໺ʹߩݙ͢Δ͜ͱ͕͋Δఔ౓ٻΊΒΕΔ • ೋਓɾ׬શ৘ใήʔϜ͸͔ͳΓݚڀ͠ਚ͘͞Ε͖ͯ ͍ͯΔ • ଟਓ਺΍ɺෆ׬શ৘ใήʔϜɺGeneral Game Playingʹؔ৺͕ҠΓͭͭ͋Δ

Slide 97

Slide 97 text

ରઓήʔϜAIͷ͜Ε͔Β (ࢲݟ) • झຯͱͯ͠ͷήʔϜAIϓϩάϥϛϯάΛ  ΋ͬͱؾܰʹָ͠ΊΔੈքʹͳΔͱ͍͍ • ద੾ͳ੍໿Λ΋ͬͨϓϥοτϑΥʔϜ͕͋Δͱ ྑ͍? (ܭࢉͷεςοϓ਺Λ੍ݶ͢Δͱ͔) • ࠃ಺Ͱ΋ e-sports ΍ AtCoder ͳͲؔ࿈෼໺͕ ੝Γ্͕͍ͬͯΔͷͰɺػӡ͸͋Δ

Slide 98

Slide 98 text

3. boardgame.io ʹֶͿɺ  ϑϨʔϜϫʔΫઃܭͷצॴ

Slide 99

Slide 99 text

boardgame.io ͱ͸ • λʔϯ(ख൪)ϕʔεͷήʔϜΛ࡞ΔͨΊͷ  ϑϨʔϜϫʔΫ • JavaScript, Node.js, (React, React Native) • TypeScript ରԠ͸ਐߦத

Slide 100

Slide 100 text

boardgame.io ͷػೳ (ެࣜΑΓ) • State Management • Multiplayer • AI • Game Phases • Prototyping • Logs • View-Layer Agnostic • Component Toolkit • Extendable

Slide 101

Slide 101 text

ϑϨʔϜϫʔΫͱ͸Կ͔ • A structure that forms a support or frame  for something • (Collins English Dictionary ΑΓ) • Կ͔ʹର͢Δࢧ࣋෺΍࿮૊ΈΛܗ੒͢Δߏ଄

Slide 102

Slide 102 text

ϑϨʔϜϫʔΫͷྫ • Ruby on Rails • Angular

Slide 103

Slide 103 text

ࢲݟ: ྑ͍ϑϨʔϜϫʔΫ • ର৅ͱ͢Δൣғ͕޿͍͕ɺ޿͗͢ͳ͍ • ର৅ͷߏ଄Λద੾ʹଊ͍͑ͯͯɺ  ؤ݈ੑ͕͋Δ • ଞͷٕज़ͱྑ͍ܨ͕Γ͕͋Δ  (ΤίγεςϜ)

Slide 104

Slide 104 text

boardgame.io ͷର৅ • λʔϯ੍ (Turn-Based) ͷήʔϜ • ޿͍͕ɺ޿͗͢ͳ͍ద੾ͳந৅Խ • ϦΞϧλΠϜͰಈ࡞͢ΔήʔϜΛؚΊΔͱ༨ ܭͳػೳ͕૿͑ɺҰํͰ݁ہ࢖ΘΕͳͦ͏

Slide 105

Slide 105 text

(࠶ܝ) ରઓήʔϜͷϧʔϧ • ήʔϜͷঢ়ଶͷू߹, ॳظঢ়ଶ • Move • ϓϨΠϠʔ͕ήʔϜΛਐΊΔखஈ • ঢ়ଶ͔Βঢ়ଶ΁ͷભҠ • ऴྃ৚݅ͱ݁Ռ

Slide 106

Slide 106 text

boardgame.io ʹΑΔଊ͑ํ (1) • G ... ։ൃऀ͕ ؅ཧ͢ΔήʔϜ ͷঢ়ଶ • ctx ... ϑϨʔϜ ϫʔΫଆͰ؅ ཧ͢Δঢ়ଶ

Slide 107

Slide 107 text

ଊ͑ํ͕ద੾ • ྫ: কع • ʮہ໘ʯ(൫໘ɾۨͷঢ়ଶ) Λঢ়ଶ(G)ͱͯ͠؅ཧ͍ͨ͠ • ࣮ࡍʹ͸ࢦ͠खͷཤྺ΋ήʔϜͷঢ়ଶʹؚ·ΕΔ • ཤྺ͕ͳ͍ͱઍ೔खͷ൑ఆ͕Ͱ͖ͳ͍... • ͦΕΒ͸ϑϨʔϜϫʔΫ͕؅ཧͯ͘͠ΕΔͷͰɺ։ൃऀ͕ҙ ࣝ͠ͳ͍Ͱ΋࢖͑Δ (͍͍ͪͪࣗ෼Ͱཤྺ؅ཧ͠ͳ͍͍ͯ͘)

Slide 108

Slide 108 text

boardgame.io ʹΑΔଊ͑ํ (2)

Slide 109

Slide 109 text

ଊ͑ํ͕ద੾ • Move Λ G => G ͷ७ਮؔ਺ͱͯ͠ଊ͍͑ͯΔ • ෭࡞༻ͳ͠ • ϝϦοτ͕ଟ͍ • ཤྺͱͯ͠ѻ͍΍͍͢ • React ͳͲ View ଆͷϑϨʔϜϫʔΫͱ૬ੑ͕͍͍

Slide 110

Slide 110 text

boardgame.io ʹΑΔଊ͑ํ (3) • Phase ͱ͍͏֓೦Λಋೖ • Move ͨ͠ޙ Phase ΛભҠͰ͖Δ • Phase ʹΑͬͯɺऔΓ͏Δ Move ͷछྨ͕  มΘΔ

Slide 111

Slide 111 text

ଊ͑ํ͕ద੾ • Phase ͸ཧ࿦తʹ͸ήʔϜͷঢ়ଶʹؚ·ΕΔ • ͔͠͠ɺଟ͘ͷήʔϜʹϚον͢Δ֓೦ • ਓ࿛ ... னͱ໷ͷ2ϑΣʔζͷ܁Γฦ͠ • ன͸શһ͕ձ࿩͠ɺ໷͸ਓ࿛͚͕ͩձ࿩͢ Δ

Slide 112

Slide 112 text

ϞσϦϯάͷؤ݈ੑ • ྑ͍ϞσϦϯά͕ग़དྷΔͱɺ෇ਵͯ͠  ͞·͟·ͳศརػೳΛ͚ͭΒΕΔ • "1ཻͰN౓͓͍͍͠"

Slide 113

Slide 113 text

࠶ܝ: boardgame.io ͷػೳ • State Management • Multiplayer • AI • Game Phases • Prototyping • Logs • View-Layer Agnostic • Component Toolkit • Extendable

Slide 114

Slide 114 text

Multiplayer • ෳ਺ਓͰͷಉ࣌ରઓ͕Մೳ • multiplayer: { server: 'localhost:8000' }  ͷΑ͏ʹઃఆ͢Δ͚ͩ • transport ૚͕ந৅Խ͞Ε͍ͯΔઃܭ • socket.io, ﬁrebase ͕࢖͑ΔͬΆ͍

Slide 115

Slide 115 text

Multiplayer - ෆ׬શ৘ใ • ෆ׬શ৘ใήʔϜͷ৔߹ɺϓϨΠϠʔʹ͸ݟ ͑ͳ͍ήʔϜͷ৘ใ͕͋Δ • ௨৴೷͍ͨΒݟΕΔͱ͔͸μϝ • ήʔϜͷঢ়ଶ => PlayerView ͷ  ७ਮؔ਺ͱͯ͠දݱ͢Δ͚ͩͰྑ͍

Slide 116

Slide 116 text

Logs • Move ͷݺͼग़͠Λཤྺͱͯ͠ѻ͑Δ • Undo, Redo ͕Ͱ͖Δ • عේͷอଘͱ͔͕؆୯ʹͰ͖Δ

Slide 117

Slide 117 text

Prototyping • ૢ࡞ը໘Λ࡞Βͳͯ͘΋ɺdebugը໘͕ࣗಈ Ͱ࡞ΒΕΔ • ήʔϜͷਐߦΛ؆୯ʹςετͰ͖Δ  (Move ΛςΩετͰॻ͍ͯήʔϜΛਐΊΔ)

Slide 118

Slide 118 text

View-Layer Agnostic • PDS(ϓϨθϯςʔγϣϯͱυϝΠϯͷ෼཭)͕ ׬શʹ͞Ε͍ͯΔ • React Ͱը໘ͷ࣮૷͕͙͢ʹ࢝ΊΒΕΔ • ΤίγεςϜ (Component Toolkit ͳͲ) Λ  ࣗવͱ࡞Δ͜ͱ͕Ͱ͖Δ

Slide 119

Slide 119 text

AI • ϞϯςΧϧϩ໦୳ࡧϕʔεͷAIΛࣗಈͰੜ੒Ͱ ͖Δ • ϞϯςΧϧϩ໦୳ࡧ͸ήʔϜࣗମͷ஌ࣝΛඞ ཁͱ͠ͳ͍ɻ • ֤ঢ়ଶʹ͓͍ͯ Valid Move ͷू߹Λฦؔ͢ ਺Λ࡞Δ͚ͩͰɺ؆қతͳAI͕࡞ΕΔ...!

Slide 120

Slide 120 text

AI (ิ଍) • ϞϯςΧϧϩ໦୳ࡧ͕୯७ʹ࢖͑Δͷ͸ɺϥ ϯμϜʹਐΊΕ͹ඞͣऴہ͢ΔήʔϜʹݶΔ • TicTacToe(○×), ғޟ, Φηϩ, etc... • ෆ׬શ৘ใήʔϜ͸೉͍͠

Slide 121

Slide 121 text

ίʔυͷDEMO ͠Α͏ͱࢥ͕ͬͨ......

Slide 122

Slide 122 text

໰୊ • ࣗ෼ͷίʔυ͕Ԛ͗ͯ͢ݟͤΒΕͳ͍... • PDS (ݟͨ໨ͱυϝΠϯͷ෼཭)͸׬શʹ͞Εͯ ͍Δɻͦͷ఺Ͱ࠷ѱͰ͸ͳ͍ • ͔͠͠ɺͦΕͧΕͷPͱDͷίʔυ͸ΘΓͱ  ͻͲ͍ঢ়ଶʹͳ͍ͬͯΔ

Slide 123

Slide 123 text

ϑϨʔϜϫʔΫͷର৅ͱݶք • ϑϨʔϜϫʔΫ͸ۜͷ஄ؙͰ͸ͳ͍ • ։ൃऀ͸ɺϑϨʔϜϫʔΫ͕ର৅ͱ͍ͯ͠Δ ྖҬΛݟۃΊΔඞཁ͕͋Δ • ର৅ͱ͍ͯ͠ͳ͍ྖҬʹ͍ͭͯɺࣗ෼ͷ಄Ͱ ߟ͑ͳ͚Ε͹ͳΒͳ͍

Slide 124

Slide 124 text

ଟ෼͜ͷ࿩ʹܨ͕Δ

Slide 125

Slide 125 text

·ͱΊ • ྑ͍ϑϨʔϜϫʔΫ͸ɺղܾ͢Δର৅ͷ޿͞ Λద੾ʹͱΓɺͦΕΛਖ਼͘͠ଊ͍͑ͯΔɻ • ։ൃऀ΋ͦΕΛཧղ͠ɺϑϨʔϜϫʔΫ͕α ϙʔτ͠ͳ͍෦෼ʹ͍ͭͯࣗ෼Ͱߟ͑Δඞཁ ͕͋Δɻ

Slide 126

Slide 126 text

·ͱΊͱࠓޙ

Slide 127

Slide 127 text

·ͱΊ • ϧʔϧɾήʔϜ໦ɾ෼ྨͷ3ͭͷ؍఺Ͱରઓ ήʔϜΛϞσϦϯά͠ɺͦΕͧΕ࣮ࡍͷରઓ ήʔϜͷੈքͱͷܨ͕ΓΛΈͨ • ϞσϦϯά͸ɺ AI ͷΞϧΰϦζϜ΍ϑϨʔϜ ϫʔΫΛ࡞Δͱ͖ʹ׆͔͞ΕΔ • ରઓήʔϜ໘ന͍ΑͶ!!

Slide 128

Slide 128 text

ࠓޙ • ෆ׬શ৘ใήʔϜ΍൚༻AIͳͲͷݚڀ͕ࠓޙ͞Βʹ ੝Γ্͕͍ͬͯ͘ͷͰɺ௥͍͖͍ͬͯͨ • ࣗಈήʔϜੜ੒AI͸͋Γ͏Δͷ͔...? • ΈΜͳ͕޷͖ʹήʔϜΛ࡞ͬͨΓ༡ΜͩΓAIΛ࡞ͬ ͨΓͯ͠ൃදͰ͖ͯɺָ͍͠͠ݚڀʹ΋ͭͳ͕ΔΑ ͏ͳϓϥοτϑΥʔϜ͕Ͱ͖ͯ޿·Δͱ͍͍ͳʙ

Slide 129

Slide 129 text

ิ଍, ࢀߟࢿྉͳͲ • https://note.mu/qsona/n/n137733801fd0    ʹ͢΂ͯ·ͱΊͯ͋Γ·͢