アットホームな分析基盤の作り方 / Homemade Machine Learning Toolkits

7dc8611c26c3ca62c551109c65d04270?s=47 Yuichiro Someya
July 23, 2018
Programming
1
640

アットホームな分析基盤の作り方 / Homemade Machine Learning Toolkits

分析基盤トーク #1
https://daft.connpass.com/event/93036/

7dc8611c26c3ca62c551109c65d04270?s=128

Yuichiro Someya

July 23, 2018
Tweet

More Decks by Yuichiro Someya

See All by Yuichiro Someya
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
0
3.8k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
3
1.5k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
0
2.4k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
15
4.5k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
1
260
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
15
9.4k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
1
2.3k
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
0
300
7dc8611c26c3ca62c551109c65d04270?s=48 ayemos
0
72

Other Decks in Programming

See All in Programming
C6296f654b1cc6886aabe5b9703cbe67?s=48 ishiitetsuji
1
580
53de6e9d0d7bedce85c7b29256448274?s=48 forrep
12
3.1k
4ab3fec3e82ddb19bcadd93ef909a443?s=48 syucream
2
380
Dcbf2269af2483cabf43128ad1718c11?s=48 grapecity_dev
0
310
0b8291daeda1cd55e445af644d402bb0?s=48 aoiroaoino
2
290
1fe64ebb176498be5f73ab51986c6b7b?s=48 davidfowl
9
3k
15934fa2aa7b2ce21f091e9b7cffa856?s=48 manfredsteyer
1
210
5b8d20aa7d63c5d391b1c881e1764460?s=48 quasilyte
2
220
5bd19ec9172dcb6dd3cb8033d88b262e?s=48 rom1000
1
160
Defdf943f6e9bfc3d3cd856d9d9e0f9b?s=48 takutakahashi
1
440
89f41026db2c96853402524fca4a2ff8?s=48 brentchang
0
940
68ba97b6be908d58de5294261a6d8638?s=48 ca_dp
0
290

Featured

See All Featured
E837d2996f52008ace055a08fbfff992?s=48 frogandcode
122
20k
79acae287842acf29084005533a7fb3b?s=48 hursman
104
9.1k
96270e4c3e5e9806cf7245475c00b275?s=48 addyosmani
492
110k
3d4519fd34b76fb265fc0237f3792bd4?s=48 danielanewman
197
19k
E190023b66e2b8aa73a842b106920c93?s=48 zenorocha
296
39k
Ae14cc4491ac334f9cd23f9f93b4305e?s=48 orderedlist
PRO
326
35k
1177e050db6bafe62885362edf6e3537?s=48 lauravandoore
435
28k
B3ace07412a5178ea778e7eec161fc0a?s=48 jcasabona
7
460
465724d73fe3a92c0879fdfb43a3a6f3?s=48 philhawksworth
190
8.7k
20bfe76b3d6105641f879fe45cfc9272?s=48 bkeepers
319
53k
433acaea1012b25d97ae66da9b998dad?s=48 malarkey
118
15k
56c4053438af8e8b90d6f53cbb7573be?s=48 jakevdp
769
190k

Transcript

  1. ΞοτϗʔϜͳ
 ෼ੳج൫ͷ࡞Γํ BZFNPT

  2. ࣗݾ঺հ  છ୩༔Ұ࿠<:VJDIJSP4PNFZB>  ౦ژ޻ۀେֶେֶӃܭࢉ޻ֶम࢜  ΫοΫύουגࣜձࣾݚڀ։ൃ෦
 ϦαʔνΤϯδχΞ೥݄d  ػցֶशج൫

    Ϩγϐσʔλͷ෼ੳ౳ UXJUUFSDPN!BZFNPT@Z HJUIVCDPNBZFNPT IUUQTXXXBZFNPTNF

  3. ΫοΫύουݚڀ։ൃ෦  ೥݄ʹൃ଍  ౰࣌͸໊ͷϝϯόʔ  ೥݄ݱࡏ͸ࠃ಺ʹ໊ ւ֎ʹ໊

  4. ΫοΫύουͱػցֶश  .ZϑΥϧμ ͓ؾʹೖΓػೳ ಺ͷϨγϐΛࣗಈΧςΰϦ෼ྨ  ʮྉཧ͖Ζ͘ʯεϚʔτϑΥϯ಺ͷྉཧࣸਅΛΧϨϯμʔܗࣜ ͰৼΓฦΓ  ࡐྉ໊ͷਖ਼نԽ

     ౳ʑ  ৄ͘͠͸IUUQTUFDIMJGFDPPLQBEDPN

  5. BHFOEB  ΫοΫύουͷ෼ੳج൫ͷ঺հ  ෼ੳج൫ͷ͜Ε͔Β

  6. ج൫঺հͦͷ;PPFZ

  7. None

  8. ෼ੳͱ࣮ݧΛ֤͕ࣗࣗ༝ʹߦ͏ͨΊͷ࠷௿ݶͷج൫

  9. ج൫঺հͦͷ̍;PPFZ  ܭࢉػ؀ڥߏங༻$IBU#PU  ܭࢉػ؀ڥͷ࡞੒ʗ্ཱͪ͛ʗఀࢭʗ࡟আ  ΞΠυϧঢ়ଶͷܭࢉػΛࣗಈఀࢭ

  10. None

  11.  $6%" DV%//  $6%" DV%//  %FFQ-FBSOJOH".*+BO
 FUD 0VUHPJOH8FCIPPLT

    TMBDL@VJE6  DPNNBOEDSFBUFXPSLCFODI MBNCEB*OWPLF  71$ 4VCOFU /BNFλά౳
 ؀ڥಛ༗ͷઃఆ΋͜͜Ͱ΍Δ  ΠϯελϯεʹATMBDL@VJEAΛ
 λά෇͚ͯ͠ॴ༗ऀΛ؅ཧ͢Δ DSFBUFXPSLCFODI ;PPFZ

  12. ATTIBZFNPTXPSLCFODIEOTDPNA ;PPFZ *ODPNJOH8FCIPPLT TTI

  13. ;PPFZ 4UPQ  ΞΠυϧͳΠϯελϯεΛࣗಈఀࢭ

  14. ج൫঺հͦͷ̍;PPFZ  ಺੡ͷԾ૝*NBHF %FFQ-FBSOJOH".*  Λར༻Մೳ  Πϯελϯε࡞੒ͷೖΓޱΛҰຊԽ  ωοτϫʔΫɺηΩϡϦςΟपΓͷࡉ͔͍ઃఆΛٵऩ


    ར༻ଆ͸*NBHFΛ൪߸ͰબͿ͚ͩ  ࣾ಺ΠϯϑϥͷมԽʹରͯ͠͸πʔϧͷΞοϓσʔτͰରԠ  ւ֎ͷϝϯόʔ΋ར༻  IUUQTBXTBNB[PODPNKQNBDIJOFMFBSOJOHBNJT

  15. ج൫঺հͦͷ̍;PPFZ  ֤͕ࣗSPPUΛऔΕΔΠϯελϯεΛ෷͍ग़͢  গʑલ࣌୅తʁ  ॊೈੑ͸ߴ͘ɺ࣮ݧஈ֊ͷ؀ڥͱͯ͠͸࢖͍৺஍͕Α͍  ࣗಈఀࢭ΋͋ΔͷͰ֤ࣗ҆৺ͯ͠ඞཁͳ্ཱ͚ͩͪ͛ΒΕΔ 

    ʮ࣮ݧʯΛؚΊαϙʔτ͢Δڞ௨ج൫Λ࣋ͨͳ͍ͷͰ
 ʮ࣮ݧʯʮຊ൪ʯͷ1PSUBCJMJUZ͸ผ్ิڧ͢Δඞཁ͕͋Δ

  16. ج൫঺հͦͷσʔλج൫

  17. None

  18. ͍ΘΏΔ%8) 42-ϕʔεͷ&5-ج൫

  19. σʔλج൫  ͜Ε͸ओʹผνʔϜ σʔλج൫άϧʔϓ ʹΑΔ؅ཧ  "NB[PO3FETIJGUʹΑΔ%8)ߏங  .Z42-3FETIJGU΁ͷΠϯϙʔτδϣϒΛίʔυ؅ཧ 

    ৄ͘͠͸IUUQTUFDIMJGFDPPLQBEDPNFOUSZ 

  20. σʔλج൫ͷར༻ঢ়گ  %8)͕ਆ  &5-  ࣮ݧͷ౓ʹ֤͕ࣗ౎౓ߦ͍ɺεΫϦϓτԽ͢Δࣄ͕ଟ͍  ڞ௨&5-ج൫ͷར༻ɺதؒσʔλͷڞ༗ͳͲʹର͢Δχʔζ ͸ͦ͜·Ͱେ͖͘ͳ͍

    ݱঢ়

  21. ج൫঺հͦͷίʔυ؅ཧ

  22. ج൫঺հͦͷίʔυ؅ཧ  ج൫ʁ   DPPLJFDVUUFSEBUBTDJFODF ࣾ಺GPSL Λར༻  ϓϩδΣΫτߏ੒ͷςϯϓϨʔτ

     ATSDEBUBA ATSDNPEFMAσʔλੜ੒ɺֶशεΫϦϓτ  AEBUBA ANPEFMTAσʔλɺϞσϧ  HJUʹ͸DIFDLJOͤͣɺ4ͱTZOD͢Δ ANBLFTZOD@EBUB@UP@TA

  23. ج൫঺հͦͷίʔυ؅ཧ  IUUQTHJUIVCDPNEPDLFSTDJFODFDPPLJFDVUUFSEPDLFS TDJFODF  DPPLJFDVUUFSEBUBTDJFODFΛࢀߟʹͨ͠ςϯϓϨʔτ  EPDLFSΛར༻͠ɺϓϩδΣΫτͷ࣮ݱੑΛ͞ΒʹߴΊΔ  OPUFCPPLͷ্ཱͪ͛

    1PSUGPSXBSE౳ߦ͏UBSHFU΋

  24. ج൫·ͱΊ  ݱঢ়͸౷ҰԽ͞ΕͨػցֶशϓϥοτϑΥʔϜΛར༻͍ͯ͠ͳ͍  ࣗ੡πʔϧͷ૊Έ߹Θͤ %8)Ͱ͓͓ΉͶճ͍ͬͯΔ  ج൫ͦͷ΋ͷͷ؅ཧίετͳͲߟྀͭͭ͠ਐΊͨ݁Ռ
 ݱঢ়͸ ൺֱతΧδϡΞϧͳӡ༻ͱͳ͍ͬͯΔ

  25. ෼ੳج൫ͷࠓޙ ʙΞοτϗʔϜͳ෼ੳج൫͔Βେ౷Ұج൫΁ͷಓͷΓʙ ͔͜͜Βߟ࡯

  26. ෼ੳج൫ͷࠜຊత՝୊  (FOFSJDBOE&YUFOTJCMF

  27. ෼ੳج൫ͷࠜຊత՝୊ 5'9"5FOTPS'MPX#BTFE1SPEVDUJPO4DBMF.BDIJOF-FBSOJOH1MBUGPSN ΑΓ l5IFNBDIJOFMFBSOJOHQMBUGPSNNVTUCFHFOFSJD FOPVHIUPIBOEMFUIFNPTUDPNNPOTFUPGMFBSOJOH UBTLTBTXFMMBTCFFYUFOTJCMFUPTVQQPSUPOFP⒎ BUZQJDBMVTFDBTFTz ҙ༁ ػցֶशج൫͸แׅత͔֦ͭுՄೳͰ͋Δ΂͖

  28. (FOFSJDBOE&YUFOTJWF  ʰ࣮ݧ͔Βຊ൪ӡ༻·Ͱɺ౷߹తͳ؀ڥͰػցֶशʱ
 ͠Α͏ͱ͍ͯ͠Δ  (FOFSJDͳΔ΂͘ଟ͘ͷϢʔεέʔεʹ ಁաతʹ ରԠ͍ͨ͠  ྑ͍ந৅Խͱ"1*֦ॆͷؤுΓ͖ͬͱେม

     &YUFOTJWFྫ֎తͳέʔεʹରͯ͠΋ গͳ͍࿑ྗͰ ରԠ͍ͨ͠  ೚ҙͷίʔυΛࠩ͠ࠐΊΔ৔ॴΛ༻ҙ͢Δɺͱ͔ʜ

  29. (FOFSJDBOE&YUFOTJWF  αϙʔτ͢Δ໰୊ͷόϦΤʔγϣϯͱͷઓ͍͸ଓ͘  ෼ྨʗճؼ͘Β͍ͷந৅Խ͔Β࿙Εͯ͠·͏෦෼͕ແࢹͰ͖ͳ͍ ྔʹͳ͍ͬͯΔ ཁग़య   ج൫ͱ͍͏ܕʹ͸ΊΔ͜ͱͰࣦΘΕΔॊೈੑͷ૯ྔΛ༧ଌ͢Δ͜

    ͱ͸ग़དྷͳ͍ ཁग़య

  30. ߟ࡯  ج൫͸େมͳ͸ͣͳͷͰ ࡞Δਓͱ࢖͏ਓ͕૿͑ͯίϛϡχςΟ͕੒௕ ͢Δඞཁ͕͋Δ  ҰํɺνʔϜͷεέʔϧʹ࠷దԽ͞ΕͨπʔϧΛར༻ʗࣗ੡ͯ͠ࡁ·ͤ Δͱ͍͏બ୒ࢶ΋͋Δ  ͦͷΑ͏ͳݱ৔͔Β͸౷Ұ͞Εͨ෼ੳج൫ʹΑͬͯ΋ͨΒ͞ΕΔϝϦο

    τ͕Πϝʔδ͠ʹ͍͘

  31. ߟ࡯  ͷͰɺ લड़ͨ͠Α͏ͳ՝୊Λ্๊͑ͨͰ ෼ੳج൫͕ଧͪग़͢ϝϦοτ͸Կ͔ɺΛߟ͑Δ ͷ͕େࣄͩͱࢥ͏  ػցֶशΛج൫ͳ͠Ͱεέʔϧͤ͞Δͷ͕ ෆՄೳʹۙ͘ ೉͍͔͠Βɺ


    ಋೖʹ൐͏ίετ΍՝୊Λड͚ೖΕͯͰ΋ج൫ʹैͬͯ։ൃ͢Δͱ͍͏બ୒Λ͢΂͖ɺ ͳͷ͔  εέʔϧ͕೉͍͠ͱ͍͏࣮ײΛ ۀքશମ͕ ࠓͷ࣌఺Ͱ ಘ͍ͯΔͷ͔  ڞ௨ج൫Խ͢Δ͜ͱʹΑͬͯಘΒΕΔߴڃͳػೳ "#ςετɺ؂ࢹͳͲ 
 ͷັྗΛԡ͠ग़͢ͷ͔  ౳ʑ

  32. ߟ࡯  ʮσʔλͷྲྀΕ͕͋ΓɺγεςϜ͕σʔλͱڞʹ੒௕͢ΔΑ͏ͳγε ςϜʯ·Ͱࢹ໺Λ޿͛Δͱʜ  ػցֶशΛ༻͍ͳ͍ ݕࡧΤϯδϯ΋ٞ࿦ͷର৅
 ͱͳΔͷ͕ඞવ  ݕࡧΤϯδϯࣗମ͸ಠࣗਐԽͯ͠ΔͷͰ৽ͨͳ


    ιϦϡʔγϣϯΛଧͪग़͢༨஍͸ͳ͍͔΋͠Εͳ͍͕ʜ  ͱ͸͍͑χϡʔϥϧݕࡧ౳૊Έ߹ΘͤͨγεςϜʹର͢Δ
 χʔζ͸ࠓޙ૿͍͑ͯ͘͸ͣ  ͜͜·Ͱ&YUFOTJCMFʹͳΕΔ͔ ͳΔ΂͖͔ IUUQTXXXBNB[PODPKQ#VJMEJOH*OUFMMJHFOU4ZTUFNT-FBSOJOH&OHJOFFSJOHFCPPLEQ##82)3

  33. ·ͱΊ  ΫοΫύουݚڀ։ൃ෦͕ར༻͍ͯ͠Δ෼ੳج൫Λ঺հ  ෦಺ʹج൫୲౰͕͍ͯɺϢʔεέʔεʹরΒ͠߹Θͤͳ͕Βۀ຿վળ  ݱঢ়͸ࣗ੡πʔϧͷ૊Έ߹Θͤ %8)Ͱ͓͓ΉͶຬ଍  ෼ੳج൫ͷ͜Ε͔Β

     ݱ৔ͱͯ͠͸ͨ͘͞ΜͷϢʔεέʔεΛൃ৴͍ͯ͘͠  ػցֶशͷ๊͑Δ՝୊Λ۩ମԽ͠ɺϝοηʔδͱͯ͠఻͍͑ͯ͘ͷ͕
 ॏཁͰେมͦ͏