自分のための機械学習をしてみた話

ࣗ෼ͷͨΊͷػցֶशΛ ͯ͠Έͨ࿩ !UIJOL"NJ 3FDBQPG5FOTPS'MPX%FW4VNNJUJO৴भ -5

ࣗݾ঺հ • UIJOL"NJ ◦ ͸ͯͳ 5XJUUFS (JU)VC • झຯɿΓΜͨ͝΂͋Δ͖ ◦
໿छྨ͘Β͍ • ࢓ࣄɿ1ZUIPO ◦ %KBOHP$POHSFTT • (%(৴भɺ/4&( • ΪʔΫϥϘ௕໺ IUUQSJOHPUBCFUUFSIFSPLVBQQDPNIDUPUBM

ࣗ෼ͷͨΊͷػցֶशΛͯ͠Έͨ࿩ ◦ ΄͍݁͠ՌΛಘΔͨΊʹɺ͍Ζ͍ΖͱࢼߦࡨޡΛͯ͠Έͨ࿩ • ͋ͱͰεϥΠυ͸ެ։͠·͢

࣭໰ ೔ ͸Կͷ೔ʁ

౴͑ *1"ͷ৘ใॲཧࢼݧ

ߴ౓ࢼݧͷޕલ͸ ಉ೔ͷԠ༻৘ใٕज़ऀࢼݧ ͷޕલͱಉ͡໰୊͕ग़Δ Β͍͠

͔ͤͬ͘ͳͷͰ ػցֶशͰௐ΂ͯΈͨ ͱ͍͏-5

࣭໰ • ػցֶशɺಛʹࣗવݴޠॲཧΛ΍ͬͨ͜ͱ͕͋Δํ • 1ZUIPOΛॻ͍ͨ͜ͱ͕͋Δํ

ௐ΂ͯΈͨྲྀΕ աڈ໰Λೖख 0$3ॲཧ ղੳ݁ՌΛલॲཧͯ͠ɺσʔλΛΩϨΠʹ͢Δ ػցֶशͰࣗવݴޠॲཧ
ֶश݁ՌΛදࣔ ιʔείʔυ 1ZUIPO ͸ͪ͜Β IUUQTHJUIVCDPNUIJOL"NJJQB@JTTVFT@NPEFM@CZ@EPDWFD

աڈ໰Λೖख • ͪ͜Β͔Β ◦ *1"ಠཱߦ੓๏ਓ ৘ใॲཧਪਐػߏɿաڈ໰୊ ▪ IUUQTXXXKJUFDJQBHPKQ@IBOOJ@TVLJSV@JOEFY@NP OEBJIUNM •
0$3ॲཧ͞Ε͍ͯͳ͍ͬΆ͍QEG ◦ ͭΒ͍

0$3ॲཧ • 1ZUIPOͷ৔߹ɺQZPDS 5FTTFSBDUͰ͍͚ͦ͏ • ͱ͸͍͑ɺ͜͜͸(%(৴भ ◦ (%((PPHMF%FWFMPQFST(SPVQT • (PPHMFΛ࢖Θͳ͍ͱʂ

(PPHMFͰ0$3ॲཧͰ͖ͦ͏ͳ΋ͷ • (PPHMF$MPVE7JTJPO"1* • (PPHMF%SJWFͷ0$3ॲཧ ◦ ࠓճ͸ͪ͜ΒΛ࢖͏

(PPHMF%SJWF"1*Ͱ0$3ॲཧͰ͖Δ͔ • ύϥϝʔλ͕͋ͬͨ ◦ 1045IUUQTXXXHPPHMFBQJTDPNVQMPBEESJWFWGJMFTͷ PDS-BOHVBHF ▪ IUUQTEFWFMPQFSTHPPHMFDPNESJWFWSFGFSFODFGJMFTDSFBUF • 1ZUIPOͰॻ͍ͯΈͨ
◦ 0$3ॲཧ͕Ͱ͖ͨ ◦ HPPHMF@ESJWF@VQMPBEFSQZ • ͜Μͳײ͡ NFEJB@CPEZ.FEJB'JMF6QMPBE G NJNFUZQF.*.&@5:1& SFTVNBCMF5SVF CPEZ\OBNFGOBNF NJNF5ZQF.*.&@5:1& QBSFOUT<EJSFDUPSZ@JE> ^ STFSWJDFGJMFT DSFBUF CPEZCPEZ NFEJB@CPEZNFEJB@CPEZ PDS-BOHVBHFKB FYFDVUF

0$3ྫ ࠨɿݪຊɺӈɿ(PPHMF%SJWF0$3

0$3ͷਫ਼౓ʹ͍ͭͯ • ࠓճɺ0$3ͷਫ਼౓ʹ͍ͭͯ͸ͩ͜ΘΒͳ͍ ◦ ಉ͡໰୊Ͱ͋Ε͹ɺಉ͡ਫ਼౓ͰςΩετԽ͞ΕΔ͸ͣ ▪ ยํ͕ޡͬͯղੳ͞Εͯ΋ɺ΋͏ยํ΋ޡͬͯղੳ͞ΕΔ͸ͣ • ͱΓ͋͑ͣ0$3ॲཧͰ͖͍ͯΕ͹0,

σʔλͷલॲཧ • 0$3ޙͷσʔλΛݟͨͱ͜Ζɺͦͷ··Ͱ͸ѻ͍ͮΒ͍͜ͱʹؾͮ͘ ◦ վߦ͕ଟ͍ ◦ ʮ໰ʯ͚͕ͩʮ໰ʯͱͯ͠0$3ॲཧ͞ΕͯΔ ▪ ଞͷ໰୊͸େৎ෉ͦ͏ͳͷʹ •
ͱ͸͍͑ɺ͋·ΓྗೖΕͳ͍

ػցֶशͰͷࣗવݴޠॲཧ • ػցֶशͰͷࣗવݴޠॲཧ͸ɺ͍Ζ͍Ζ͋Δ • ࠓճ͸%PD7FDΛબ୒ ◦ จষͷྨࣅੑΛݟ͚ͭ΍ͦ͢͏ͩͬͨͷͰ • %PD7FDͱ͸ ◦
จষΛϕΫτϧԽ ▪ ϕΫτϧԽ͢ΔͱίϯϐϡʔλͰѻ͍΍͘͢ͳΔ ▪ ෼ྨͱ͔

%PD7FDϞσϧ • 5FOTPS'MPX ◦ ͦΕͬΆ͍࣮૷͸͋ͬͨ ▪ IUUQTHJUIVCDPN1BDLU1VCMJTIJOH5FOTPS'MPX.BDIJOF-FBSOJOH $PPLCPPLCMPCNBTUFS$IBQUFSEPDWFDQZ ◦ ࣗྗ࣮૷͢Δʹ͸஌͕ࣝ଍Γͳ͍

%PD7FDϞσϧ • (FOTJN ◦ (FOTJNJTB1ZUIPOMJCSBSZGPSUPQJDNPEFMMJOH EPDVNFOU JOEFYJOHBOETJNJMBSJUZSFUSJFWBMXJUIMBSHFDPSQPSB5BSHFU BVEJFODFJTUIFOBUVSBMMBOHVBHFQSPDFTTJOH /-1 BOE
JOGPSNBUJPOSFUSJFWBM *3 DPNNVOJUZ ▪ IUUQTHJUIVCDPN3B3F5FDIOPMPHJFTHFOTJN ◦ %PD7FDϞσϧ͕ఏڙ͞Ε͍ͯΔ ▪ IUUQTSBEJNSFIVSFLDPNHFOTJNNPEFMTEPDWFDIUNM

%PD7FD͢ΔͨΊʹ • จॻΛܗଶૉղੳΛ͢Δඞཁ͕͋Δ ◦ ܗଶૉղੳɿจॻΛɺҙຯΛ࣋ͭ࠷খ୯Ґʹ෼ղ͢Δ͜ͱ ◦ ͓଴͓ͪͯ͠Γ·͢ ͓଴ͪ ͠ ͯ
͓Γ ·͢ 8JLJQFEJB • ϥΠϒϥϦ ◦ .F$BC΍KBOPNFɺ+6."/ ͳͲ ▪ ࠓճ͸ɺखݩʹ͋ͬͨ+6."/ Λ࢖༻ • +6."/ ͷΠϯετʔϧ ◦ .BDͷ৔߹ɺ)PNFCSFXͰΠϯετʔϧ ▪ IUUQTDIF[PVIBUFOBCMPHDPNFOUSZ • 1ZUIPO͔Βѻ͏ͨΊʹɺ1Z,/1ΛΠϯετʔϧ ◦ IUUQOMQJTUJLZPUPVBDKQJOEFYQIQ 1Z,/1

+6."/ ͰܗଶૉղੳɺτϨʔχϯάσʔλ༻ҙ • doc2vec_runner.py def morphological_analysis(doc): r = Jumanpp().analysis(doc) return
[mrph.midasi for mrph in r.mrph_list()] def get_trainings(issue_type): trainings = [] paths = get_file_paths(issue_type) for p in paths: doc = read_file(p) words = morphological_analysis(doc) trainings.append(TaggedDocument(words=words, tags=[p.stem])) return trainings

HFOTJNͷ%PD7FDͰֶश • doc2vec_runner.py model = Doc2Vec( documents=all_trainings, dm=1, # PV-DMモデルを使う
min_count=1, # これ以下の出現数の単語は無視 workers=4, # スレッドのワーカー数 epochs=EPOCHS, # エポック数 <= これが重要だった )

%PD7FDͰͷֶश ΤϙοΫ਺ͷҧ͍ • ΤϙοΫ਺ FQPDI ◦ ༻ҙֶͨ͠शσʔλͰɺԿճ܇࿅͢Δ͔ • ॏཁͳͷ͸ɺద੾ͳ਺Ͱ܇࿅͢Δ͜ͱ
◦ IUUQTTUBDLPWFSGMPXDPNRVFTUJPOTXIBUBSFEPDWFDUSBJOJOH JUFSBUJPOT

%PD7FDͰͷֶश ΤϙοΫ਺ͷҧ͍ • ݁Ռͷҧ͍ ࠨɿΤϙοΫɺӈɿΤϙοΫ

ֶश݁ՌΛදࣔ • TJNJMBSJUZ@SVOOFSQZ ◦ Ұ൪ྨࣅͯ͠ΔͷΛग़͢ TJNJMBSJUZ

ֶश݁ՌΛදࣔ • TJNJMBSJUZ@SVOOFSQZ ◦ ࣅ͍ͯΔ্ҐO݅ NPTU@TJNJMBS

·ͱΊ • ҙ֎ͱͦΕͳΓͷਫ਼౓͕ग़ͨ ◦ ΄͍͠σʔλ͕ಘΒΕͨ • πʔϧΛ૊Έ߹ΘͤΕ͹ɺࣗ෼ͷͨΊͷػցֶश͕Ͱ͖ͦ͏ ◦ ָ͍͠ •
͜ΕΛ࡞Δͷָ͕͔ͬͨ͠ͷͰɺࢼݧରࡦͰ͖ͯͳ͍ɻɻɻ

$. • /4&( ◦ ౔ dʮ044ϥΠηϯεʯษڧձ ▪ ߨࢣɿςΫχΧϧϥΠλʔʗ*5ϥΠλʔͷՄ஌๛ࢯ ▪
৔ॴɿΪʔΫϥϘ௕໺ ▪ IUUQTOTFHDPOOQBTTDPNFWFOU • *P5-5 ◦ ౔ d௕໺൛*P5റΓͷษڧձ *P5-5௕໺ WPM ▪ ৔ॴɿΪʔΫϥϘ௕໺ ▪ ୈ෦లࣔ IUUQTJPUMUDPOOQBTTDPNFWFOU ▪ ୈ෦-5େձ IUUQTJPUMUDPOOQBTTDPNFWFOU

自分のための機械学習をしてみた話

自分のための機械学習をしてみた話

thinkAmi

More Decks by thinkAmi

Other Decks in Programming

Featured

Transcript

ࣗ෼ͷͨΊͷػցֶशΛ ͯ͠Έͨ࿩ !UIJOL"NJ 3FDBQPG5FOTPS'MPX%FW4VNNJUJO৴भ -5

ࣗݾ঺հ • UIJOL"NJ ◦ ͸ͯͳ 5XJUUFS (JU)VC • झຯɿΓΜͨ͝΂͋Δ͖ ◦

ࣗ෼ͷͨΊͷػցֶशΛͯ͠Έͨ࿩ ◦ ΄͍݁͠ՌΛಘΔͨΊʹɺ͍Ζ͍ΖͱࢼߦࡨޡΛͯ͠Έͨ࿩ • ͋ͱͰεϥΠυ͸ެ։͠·͢

࣭໰ ೔ ͸Կͷ೔ʁ

౴͑ *1"ͷ৘ใॲཧࢼݧ

ߴ౓ࢼݧͷޕલ͸ ಉ೔ͷԠ༻৘ใٕज़ऀࢼݧ ͷޕલͱಉ͡໰୊͕ग़Δ Β͍͠

͔ͤͬ͘ͳͷͰ ػցֶशͰௐ΂ͯΈͨ ͱ͍͏-5

࣭໰ • ػցֶशɺಛʹࣗવݴޠॲཧΛ΍ͬͨ͜ͱ͕͋Δํ • 1ZUIPOΛॻ͍ͨ͜ͱ͕͋Δํ

ௐ΂ͯΈͨྲྀΕ աڈ໰Λೖख 0$3ॲཧ ղੳ݁ՌΛલॲཧͯ͠ɺσʔλΛΩϨΠʹ͢Δ ػցֶशͰࣗવݴޠॲཧ

աڈ໰Λೖख • ͪ͜Β͔Β ◦ *1"ಠཱߦ੓๏ਓ ৘ใॲཧਪਐػߏɿաڈ໰୊ ▪ IUUQTXXXKJUFDJQBHPKQ@IBOOJ@TVLJSV@JOEFY@NP OEBJIUNM •

0$3ॲཧ • 1ZUIPOͷ৔߹ɺQZPDS 5FTTFSBDUͰ͍͚ͦ͏ • ͱ͸͍͑ɺ͜͜͸(%(৴भ ◦ (%((PPHMF%FWFMPQFST(SPVQT • (PPHMFΛ࢖Θͳ͍ͱʂ

(PPHMFͰ0$3ॲཧͰ͖ͦ͏ͳ΋ͷ • (PPHMF$MPVE7JTJPO"1* • (PPHMF%SJWFͷ0$3ॲཧ ◦ ࠓճ͸ͪ͜ΒΛ࢖͏

(PPHMF%SJWF"1*Ͱ0$3ॲཧͰ͖Δ͔ • ύϥϝʔλ͕͋ͬͨ ◦ 1045IUUQTXXXHPPHMFBQJTDPNVQMPBEESJWFWGJMFTͷ PDS-BOHVBHF ▪ IUUQTEFWFMPQFSTHPPHMFDPNESJWFWSFGFSFODFGJMFTDSFBUF • 1ZUIPOͰॻ͍ͯΈͨ

0$3ྫ ࠨɿݪຊɺӈɿ(PPHMF%SJWF0$3

0$3ͷਫ਼౓ʹ͍ͭͯ • ࠓճɺ0$3ͷਫ਼౓ʹ͍ͭͯ͸ͩ͜ΘΒͳ͍ ◦ ಉ͡໰୊Ͱ͋Ε͹ɺಉ͡ਫ਼౓ͰςΩετԽ͞ΕΔ͸ͣ ▪ ยํ͕ޡͬͯղੳ͞Εͯ΋ɺ΋͏ยํ΋ޡͬͯղੳ͞ΕΔ͸ͣ • ͱΓ͋͑ͣ0$3ॲཧͰ͖͍ͯΕ͹0,

σʔλͷલॲཧ • 0$3ޙͷσʔλΛݟͨͱ͜Ζɺͦͷ··Ͱ͸ѻ͍ͮΒ͍͜ͱʹؾͮ͘ ◦ վߦ͕ଟ͍ ◦ ʮ໰ʯ͚͕ͩʮ໰ʯͱͯ͠0$3ॲཧ͞ΕͯΔ ▪ ଞͷ໰୊͸େৎ෉ͦ͏ͳͷʹ •

ػցֶशͰͷࣗવݴޠॲཧ • ػցֶशͰͷࣗવݴޠॲཧ͸ɺ͍Ζ͍Ζ͋Δ • ࠓճ͸%PD7FDΛબ୒ ◦ จষͷྨࣅੑΛݟ͚ͭ΍ͦ͢͏ͩͬͨͷͰ • %PD7FDͱ͸ ◦

%PD7FDϞσϧ • 5FOTPS'MPX ◦ ͦΕͬΆ͍࣮૷͸͋ͬͨ ▪ IUUQTHJUIVCDPN1BDLU1VCMJTIJOH5FOTPS'MPX.BDIJOF-FBSOJOH $PPLCPPLCMPCNBTUFS$IBQUFSEPDWFDQZ ◦ ࣗྗ࣮૷͢Δʹ͸஌͕ࣝ଍Γͳ͍

%PD7FDϞσϧ • (FOTJN ◦ (FOTJNJTB1ZUIPOMJCSBSZGPSUPQJDNPEFMMJOH EPDVNFOU JOEFYJOHBOETJNJMBSJUZSFUSJFWBMXJUIMBSHFDPSQPSB5BSHFU BVEJFODFJTUIFOBUVSBMMBOHVBHFQSPDFTTJOH /-1 BOE

%PD7FD͢ΔͨΊʹ • จॻΛܗଶૉղੳΛ͢Δඞཁ͕͋Δ ◦ ܗଶૉղੳɿจॻΛɺҙຯΛ࣋ͭ࠷খ୯Ґʹ෼ղ͢Δ͜ͱ ◦ ͓଴͓ͪͯ͠Γ·͢ ͓଴ͪ ͠ ͯ

+6."/ ͰܗଶૉղੳɺτϨʔχϯάσʔλ༻ҙ • doc2vec_runner.py def morphological_analysis(doc): r = Jumanpp().analysis(doc) return

HFOTJNͷ%PD7FDͰֶश • doc2vec_runner.py model = Doc2Vec( documents=all_trainings, dm=1, # PV-DMモデルを使う

%PD7FDͰͷֶश ΤϙοΫ਺ͷҧ͍ • ΤϙοΫ਺ FQPDI ◦ ༻ҙֶͨ͠शσʔλͰɺԿճ܇࿅͢Δ͔ • ॏཁͳͷ͸ɺద੾ͳ਺Ͱ܇࿅͢Δ͜ͱ

%PD7FDͰͷֶश ΤϙοΫ਺ͷҧ͍ • ݁Ռͷҧ͍ ࠨɿΤϙοΫɺӈɿΤϙοΫ

ֶश݁ՌΛදࣔ • TJNJMBSJUZ@SVOOFSQZ ◦ Ұ൪ྨࣅͯ͠ΔͷΛग़͢ TJNJMBSJUZ

ֶश݁ՌΛදࣔ • TJNJMBSJUZ@SVOOFSQZ ◦ ࣅ͍ͯΔ্ҐO݅ NPTU@TJNJMBS

·ͱΊ • ҙ֎ͱͦΕͳΓͷਫ਼౓͕ग़ͨ ◦ ΄͍͠σʔλ͕ಘΒΕͨ • πʔϧΛ૊Έ߹ΘͤΕ͹ɺࣗ෼ͷͨΊͷػցֶश͕Ͱ͖ͦ͏ ◦ ָ͍͠ •

$. • /4&( ◦ ౔ dʮ044ϥΠηϯεʯษڧձ ▪ ߨࢣɿςΫχΧϧϥΠλʔʗ*5ϥΠλʔͷՄ஌๛ࢯ ▪