get-started-with-machine-learning-on-aws-20170401

by ryo nakamaru

Slide 1

Slide 1 text

AWS Ͱ࢝ΊΑ͏ʂ͸͡Ίͯͷػցֶश IoT ALGYAN @ 2017.04.01

Slide 2

Slide 2 text

@pottava (AWS Certiﬁed) SA, DevOps Engineer Pro ❤ Amazon ECS, AWS Batch, IAM

Slide 3

Slide 3 text

גࣜձࣾεϐϯϑ

Slide 4

Slide 4 text

ɹɹػցֶशͷ͓͞Β͍ • ػցֶशͱ͸ • ػցֶशͱਂ૚ֶश ɹɹֶश • σʔλͷ४උ • ࢼߦࡨޡ / POC • ֶश ɹɹਪ࿦ ɹɹAWS ར༻ Tips ࠓ೔͓࿩͢͠Δ͜ͱ 4

Slide 5

Slide 5 text

ػցֶशͷ͓͞Β͍

Slide 6

Slide 6 text

ػցֶश

Slide 7

Slide 7 text

Կ͔ಛఆͷ໰୊͕͋Δͱͯ͠ɻ ίϯϐϡʔλʹ ٬؍తࣄ࣮ͷΈ Λ༩͑Δ͜ͱͰ ۩ମతͳճ౴ΛಘΔɺ·ͨ͸ͦΕΛվળ͢Δ͜ͱɻ ػցֶशͱ͸ 7

Slide 8

Slide 8 text

y = ax + b a, b Λਓ͕ؒࣄલʹܾΊΔͷ͕Ұൠతϓϩάϥϛϯάɻ ࣮σʔλ͔Βίϯϐϡʔλʹܭࢉͤ͞Δͷ͕ػցֶशɻ ػցֶशͱ͸ 8

Slide 9

Slide 9 text

y = ax + b ྫʣ x : ࠷ۙ஌Γ߹ͬͨਓͷಛ௃ y : ͜ͷͻͱͱকདྷ݁ࠗͨ͠Β޾ͤʹͳΕΔ͔ ػցֶशͱ͸ 9

Slide 10

Slide 10 text

y = ax + b ྫʣ ਓؒʮ y = 0.7 * ੑ֨x + 0.2 * ֎ݟx + 0.1 * ऩೖx ͰΑΖʯ ػցʮաڈͷσʔλ͔Β͍͑͹ ɹɹɹy = 0.4 * ੑ֨x + 0.1 * ֎ݟx + 0.5 * ऩೖx ͕ద੾ʯ ػցֶशͱ͸ 10

Slide 11

Slide 11 text

y = ax + b a, b ΛܾΊΔͨΊͷ࡞ۀΛֶशɺ ܾ·ͬͨ a, b ͷ͜ͱΛֶशࡁΈϞσϧͱݴ͏ɻ ֶशࡁΈϞσϧΛ࢖ͬͯ ࣮ࡍʹ x Λ౤ೖ͠ y ΛಘΔͷ͕ਪ࿦ɻ ֶशͱਪ࿦ 11

Slide 12

Slide 12 text

y = ax + b Ϗδωε্ॏཁͳͷ͸ɺ༏Εͨਪ࿦͕Ͱ͖Δ͔ɻ ༏Εͨਪ࿦Λ͢ΔͨΊʹ͸ɺ༏ΕͨϞσϧ͕ඞཁɻ ༏Εͨ a, b ΛܾΊΔͨΊͷֶश͕ɺ࿹ͷݟͤͲ͜Ζɻ ֶशͱਪ࿦ 12

Slide 13

Slide 13 text

y = ax + b σʔλ͔Β a, b ΛٻΊΔํ๏͸ͨ͘͞Μ͋Δɻ ղ͖͍ͨ໰୊ʹΑͬͯɺద੾ͳํ๏ΛબͿඞཁ͕͋Δɻ ֶशΞϧΰϦζϜ 13

Slide 14

Slide 14 text

y = ax + b σʔλ͔Β a, b ΛؼೲతʹٻΊΔ۩ମతͳํ๏ͷ͜ͱɻ • A / B Ͳͬͪʁ໰୊ → ϩδεςΟοΫճؼ • ച্Λ༧ଌ͍ͨ͠ → ઢܗճؼ • ސ٬Ληάϝϯτ෼͚͍ͨ͠ → k ฏۉ๏ • ϨίϝϯυΛग़͍ͨ͠ → ڠௐϑΟϧλϦϯά • … ֶशΞϧΰϦζϜ 14

Slide 15

Slide 15 text

IoT ͰूΊͨσʔλ͔Β༏ΕͨϞσϧΛ࡞Γਪ࿦͢Δɻ ͖ͬ͞ͷֶशΞϧΰϦζϜɺ࢖͑ͦ͏Ͱ͢ΑͶʁ • ΋͏͙͢ނো͢Δʁ͠ͳ͍ʁ • ऩ֭ྔ͸Ͳͷ͘Β͍ʹͳΔͩΖ͏ʁ • ࣅ௨ͬͨάϧʔϓʹ෼͚͍ͨ • ௥Ճ஫จͯ͘͠Εͦ͏ͳαΠυϝχϡʔ͸ԿͩΖ͏ʁ IoT × ػցֶश 15

Slide 16

Slide 16 text

Ұํɺࠓ೔΋࿩୊ͷ

Slide 17

Slide 17 text

ਂ૚ֶश

Slide 18

Slide 18 text

σΟʔϓϥʔχϯάɻ ଟ૚ߏ଄ͷωοτϫʔΫΛ༻͍ͨػցֶशͷ͜ͱɻ ྫʣ 4 ૚ͷωοτϫʔΫྫ ਂ૚ֶशͱ͸ 18

Slide 19

Slide 19 text

੨ؙʹ஋Λೖྗ͢Δͱɺ྘ؙʹ౴͕͑ग़ྗ͞ΕΔɻ ྫʣ͍҆ɺඒຯ͍͠ → ങ͏΂͖ 0.9ɺങΘͳ͍΂͖ 0.1 ਂ૚ֶशͱ͸ 19

Slide 20

Slide 20 text

2 ૚໨Ҏ߱ͷ֤ϊʔυ˓͕ɺલͷ૚͔ΒͷೖྗΛجʹ y = ax + b Λ࢖ͬͯࣗ෼ࣗ਎ͷ஋Λܭࢉɻ ਂ૚ֶशͱ͸ 20

Slide 21

Slide 21 text

ΑΓ΋ͬͱ΋Β͍͠౴͑Λฦͨ͢Ίʹ ֤ϊʔυͷ a, b ΛࣄલʹܾΊΔͷ͕ɺֶशɻ શϊʔυͷ a, b ͕ܾ·Ε͹ɺͦΕֶ͕शࡁΈϞσϧɻ ਂ૚ֶशͱ͸ 21

Slide 22

Slide 22 text

ػցֶशશൠɺղܾ͍ͨ͠໰୊͝ͱʹબ୒͢Δͷ͕ ֶशΞϧΰϦζϜɻਂ૚ֶश΋ྫ֎Ͱ͸ͳ͍ɾɾ ྫʣ • ͜ͷ͖Ύ͏Γ͸ S? M? L? → ৞ΈࠐΈχϡʔϥϧωοτϫʔΫ (CNN) • ࠓͷൃݴ͸ϙδςΟϒʁ → ࠶ؼܕχϡʔϥϧωοτϫʔΫ (RNN) • ਓ͕ඈͼग़͖ͯͨ͠ʂंΛݮ଎ʂʂ → ͋Ε͜Ε૊Έ߹Θͤ • … ਂ૚ֶशΞϧΰϦζϜ 22

Slide 23

Slide 23 text

ඒ͍͠૊Έ߹ΘͤͷྫɺAlphaGoɻ ʢ࿦จ: http://www.nature.com/nature/journal/v529/n7587/abs/nature16961.html ʣ • 4 ͭͷσΟʔϓχϡʔϥϧωοτϫʔΫΛར༻ • ͦΕͧΕ ॳڃऀ pπ / தڃऀ pσ / ্ڃऀ pρ / ৘੎൑அ vθ ͱ͍ͬͨҐஔ෇͚ • pπ ͸ਓؒͷ 800 ສͷ൫໘σʔλΛݩʹֶशɻਫ਼౓͸௿͍͕ߴ଎ʹղΛಘΔɻ • pσ ͸ 13 ૚ͷ CNNɻ3,000 ສ൫໘Λ 50 GPU Ͱ 3.4 ԯεςοϓɺ3 िֶؒशɻ ϓϩͷࢦ͠खΛ 57.0% ͷਫ਼౓Ͱ༧૝Ͱ͖Δɻ • pρ ͸ 50 GPU Ͱ 1 ೔͔͚128 ສճࣗݾରઓɻطଘιϑτʹ 85% ͷѹ౗తউ཰ɻ • vθ ͸ pσ ͰϥϯμϜʹ 3,000 ສ൫໘Λੜ੒͠ɺpρ Ͱ 1 ԯ 6,000 ສճϩʔϧΞ΢τͨ͠উ཰Λ ڭࢣσʔλʹɺ50 GPU ͰҰिؒ 5,000 ສճ֬཰ޯ഑߱Լ๏Λ࣮ࢪɻ • ࣮ରઓͰ͸1,202 CPU + 176 GPU͕࢖ΘΕɺpσ Ͱ࣍ͷखબ୒ɺvθ Ͱ൫໘ධՁɻ • উ཰͸͍͍͕ཧ٧Ίͷ pρ ΑΓɺਓؒͷบΛֶΜͩ pσ Λ࢖͔ͬͨɻ • উҼ͸ pρ ͷపఈతͳڧԽֶशʹՃ͑ɺϞϯςΧϧϩ໦୳ࡧͱ CNN ͷ૊߹͔ͤɻ ਂ૚ֶशΞϧΰϦζϜ 23

Slide 24

Slide 24 text

࣮ફతͳωοτϫʔΫɺ ࣮ࡍʹ͸ͲΕ͘Β͍σΟʔϓͳϨΠϠʔͳͷʁ ྫʣ ৞ΈࠐΈχϡʔϥϧωοτϫʔΫͷҰछɺResNet ͸ 152 ૚ ਂ૚ֶशΞϧΰϦζϜ https://research.googleblog.com/2016/08/improving-inception-and-image.html 24

Slide 25

Slide 25 text

͓ɺ͓͏ɾɾ Ϟσϧͱܾͯ͠ΊΔม਺ͲΕ͚ͩ͋Δͷɾɾ   ֶशʹͲΕ͚͔͔ͩ࣌ؒΔͷɾɾ ͱ͍͏͔ɺϨΠϠʔఆٛ͢Δ͚ͩͰ৺͕ંΕͦ͏ɻ ࣗલͰ࣮૷ʁ·͋ແཧͰ͢ΑͶɾɾ ਂ૚ֶशΞϧΰϦζϜ 25

Slide 26

Slide 26 text

ͦ͜Ͱ

Slide 27

Slide 27 text

ਂ૚ֶशϑϨʔϜϫʔΫ

Slide 28

Slide 28 text

ར༻ऀ͸ΞϧΰϦζϜͷ࣮૷Λ͢Δ͜ͱͳ͘ ֤छύϥϝλͷࢦఆ͚ͩͰֶशɾਪ࿦͕Ͱ͖Δɻ ྫʣTensorFlow ʹΑΔ৞ΈࠐΈχϡʔϥϧωοτϫʔΫ (CNN) ఆٛ ɹ ͲΜͳॱংͰͲΜͳ૚Λܦ༝͢Δ͔ɺ௚ײతʹΘ͔Δ ਂ૚ֶशϑϨʔϜϫʔΫ 28

Slide 29

Slide 29 text

ΞϧΰϦζϜͷ࣮૷͸ͦͷಓͷϓϩʹ͓೚ͤͭͭ͠ɾɾ ࢲͨͪ͸΍Γ͍ͨ͜ͱ͚ͩͰ͖Δ࣌୅ɺ౸དྷɻ ྫʣը૾ʹ͍ࣸͬͯΔਓ͕஌Γ͍ͨ → TensorFlow Ͱ CNN Λ࢖͑͹ֶशɾਪ࿦Ͱ͖Δʂ ≒ Golang Ͱ HTTP/2 Λ࢖͑͹ηΩϡΞͳ௨৴΋؆୯ʂ ਂ૚ֶशϑϨʔϜϫʔΫ 29

Slide 30

Slide 30 text

TensorFlowɺMXNetɺCaffeɺChainerɺTheanoɾɾ ͦΕͧΕͷಛ௃ΛؑΈͯͲΕΛ࢖͏ͷ͔ɻ • ରԠΞϧΰϦζϜ • ಈ࡞୺຤ɾ؀ڥ • ܭࢉ଎౓ / Ϧιʔεར༻ޮ཰ • ར༻Մೳͳݴޠ / खଓతɾએݴత • εέʔϥϏϦςΟ / ෳ਺ GPUɺฒྻαʔόରԠ • ৘ใͷ๛෋͞ / ΤίγεςϜ / ঎༻αϙʔτ • … ਂ૚ֶशϑϨʔϜϫʔΫ 30

Slide 31

Slide 31 text

ΫΠζͰ͢

Slide 32

Slide 32 text

ʮ2 ͔݄લ͔ΒूΊͨσʔλ͔Β ໨ඪମॏ·Ͱ͋ͱԿϲ݄ ͔͔Δ͔༧ଌ͍ͨ͠ʯ

Slide 33

Slide 33 text

࣍ͷͲΕΛ࢖͏ɾɾʁ

Slide 34

Slide 34 text

ϩδεςΟοΫճؼ ઢܗճؼ ৞ΈࠐΈχϡʔϥϧωοτϫʔΫ Ҩ఻తϓϩάϥϛϯά

Slide 35

Slide 35 text

࣮ફ͢Δલͷɺେ੾ͳϙΠϯτ

Slide 36

Slide 36 text

ʮ͋ͳͨͷۀ຿ʹػցֶशΛ׆༻͢Δ 5 ͭͷϙΠϯτʯ https://www.slideshare.net/shoheihido/5-38372284 גࣜձࣾ Preferred Infrastructure ൺށ কฏ͞Μ ͱͯ΋͍͍εϥΠυͰͨ͠ɻ 36

Slide 37

Slide 37 text

ͯ͞

Slide 38

Slide 38 text

ֶशͷྲྀΕΛ࠶֬ೝ

Slide 39

Slide 39 text

ػցֶशͷྲྀΕ 2. σʔλલॲཧ 3. ֶश 4. ਪ࿦ 1. σʔλऩू 39

Slide 40

Slide 40 text

Ҏ߱ɺ͜ͷྲྀΕʹԊ͍ɺAWS ΛͲ͏࢖͑͹͍͍ͷ͔ ར༻λΠϛϯάͱ໨తผʹ͝঺հ͠·͢ɻ ػցֶशͷྲྀΕ 2. σʔλલॲཧ 3. ֶश 4. ਪ࿦ 1. σʔλऩू 40

Slide 41

Slide 41 text

ֶश

Slide 42

Slide 42 text

σʔλͷ४උ

Slide 43

Slide 43 text

1ɹσʔλͷऩू ΋ͪΖΜɺIoT ͔ΒಘΒΕΔηϯασʔλ͸༗༻ʂʂ ͱ͸͍͑ɺ·ͣػցֶशΛࢼͯ͠ΈΔ͚ͩͳΒ Ұൠެ։͞ΕͨσʔλΛ׆༻͢Δͷ͕؆୯Ͱ͢ɻ 43

Slide 44

Slide 44 text

Ұൠެ։͞Εͨը૾ू ݚڀ΋ίϯςετ΋੝ΜͰɺͨ͘͞Μ͋Γ·͢ɻ • MNIST ɹ http://yann.lecun.com/exdb/mnist/ • CIFAR-10 & CIFAR-100 ɹ https://www.cs.toronto.edu/~kriz/cifar.html • ImageNet ɹ http://www.image-net.org/ • … 44

Slide 45

Slide 45 text

AWS Public Datasets https://aws.amazon.com/jp/public-datasets/ 45

Slide 46

Slide 46 text

Public Datasets | Google Cloud Platform https://cloud.google.com/public-datasets/ 46

Slide 47

Slide 47 text

Public data sets for Azure analytics https://docs.microsoft.com/en-us/azure/sql-database/sql-database-public-data-sets 47

Slide 48

Slide 48 text

Datasets « Deep Learning http://deeplearning.net/datasets/ 48

Slide 49

Slide 49 text

http://www.data.go.jp/data/dataset 49 ͜Μͳͷ΋͋Δ

Slide 50

Slide 50 text

AWS

Slide 51

Slide 51 text

ؔ࿈αʔϏε܈

Slide 52

Slide 52 text

σʔλऩूʹศརͳαʔϏε • AWS IoT • Amazon Kinesis Streams • Amazon CloudWatch Logs • Amazon S3 • Amazon DynamoDB • Amazon Cognito + AWS SDK • Amazon API Gateway 52

Slide 53

Slide 53 text

2ɹσʔλͷલॲཧ Python ͚ͩͰࡁΉͳΒͦΕͰ͍͍΋ͷͷɺ ෳ਺ͷσʔλιʔε͔ΒϝλσʔλΛऔಘͨ͠Γ େن໛ͳϑΝΠϧ͔ΒσʔλΛൈ͖ग़͢ͳΒ ઐ༻ͷιϑτ΢ΣΞ΍αʔϏε͕ศརɻ 53

Slide 54

Slide 54 text

AWS ͷؔ࿈αʔϏε܈

Slide 55

Slide 55 text

σʔλલॲཧʹศརͳαʔϏε • Amazon Mechanical Turk • Amazon Athena • AWS Lambda / Step Functions • AWS CloudWatch Events • Amazon EMR / Batch / EC2 55

Slide 56

Slide 56 text

Amazon Mechanical Turk 56 ΞϝϦΧͰ͸ͦͷར༻͕ ͱͯ΋ྲྀߦ͍ͬͯΔ ͱͷ͜ͱɾɾ

Slide 57

Slide 57 text

ࢼߦࡨޡ / POC

Slide 58

Slide 58 text

ࢼߦࡨޡʹศརͳ΋ͷͨͪ

Slide 59

Slide 59 text

ओʹՊֶٕज़ܭࢉ΍ػցֶशͷۀքͰ ͋Ε͜Εࢼߦࡨޡͨ͠ΓɺͦΕΛ୭͔ͱڞ༗͢ΔͨΊͷ   πʔϧɻଟ͘ͷݚڀऀ΍ΤϯδχΞʹѪ༻͞Ε͍ͯΔɻ git ͳͲͰόʔδϣϯ؅ཧ͢Δͷ΋༰қʂ Jupyter notebook 59

Slide 60

Slide 60 text

ֶशʹ͸ͱͯ΋͕͔͔࣌ؒΔ΋ͷɻ ߦྻܭࢉ͕ಘҙͳ GPU Λ࢖͑͹͕࣌ؒઅ໿Ͱ͖·͢ʂ ࣗ୐ͷ PC ʹ͍ࢗͬͯ͞Δ GPU ͕࢖͑Δ͔΋ɾɾʁ (NVIDIA) GPU 60 ʢ͜Ε͕ࢗͬͯ͞Δਓ͸͍ͳ͍ͱࢥ͏͚Ͳ..ʣ

Slide 61

Slide 61 text

(NVIDIA) GPU 61 ݱࡏ AWS Ͱ GPU Λ࢖͏ͱ͖ͷ Tips Λ·ͱΊ·ͨ͠ https://speakerdeck.com/pottava/tesorﬂow-v1-dot-0-on-ec2

Slide 62

Slide 62 text

ࢼߦࡨޡ͢Δʹ͸͜Ε΋ͱͯ΋ศརͰ͢ɻ ϥΠϒϥϦ͕ͲΜͲΜόʔδϣϯߋ৽ͯ͠΋େৎ෉ʂ Ϋϥ΢υ্ʹֶशɾਪ࿦Λ࣋ͬͯߦ͘ͱ͖ʹ΋༗༻ʂ docker run -it --rm -p 8888:8888 jupyter/tensorﬂow-notebook Docker 62

Slide 63

Slide 63 text

Docker 63 ݱࡏ AWS Ͱ Docker Λ࢖͏ͱ͖ͷ·ͱΊ https://speakerdeck.com/pottava/containers-on-aws

Slide 64

Slide 64 text

NVIDIA ͷ Docker Πϝʔδ 64 ҎԼͷΑ͏ͳܧঝؔ܎ͷΠϝʔδ͕ެ։͞Ε͍ͯ·͢ɻ ػցֶशͷಈ࡞ཁ݅ʹదͨ͠ΠϝʔδΛϕʔεʹɻ ಠࣗ Docker ΠϝʔδͷϏϧυ΋Ͱ͖·͢ɻ cuda:7.x-runtime ubuntu:14.04 cuda:7.x-devel cuda:7.x-cudax-runtime cuda:7.x-cudax-devel caffe (v0.14) digits (v4.0)

Slide 65

Slide 65 text

AWS Ϣʔβ Retty ͞Μ͕ GPUʢ෺ཧʣΛങͬͨ࿩ɻ ͱͯ΋ڵຯਂ͍ɻ Ϋϥ΢υ͸ඞཁͳͷʁ 65 http://qiita.com/taru0216/items/dda1f9f11397f811e98a

Slide 66

Slide 66 text

࣮ࡍɺνʔϜͰݚڀɾ։ൃΛ͢ΔͱͳΔͱ ؀ڥͷ෼཭ / ෳ੡ / ڞ༗ / ݖݶ؅ཧͳͲ՝୊͸ͨ͘͞Μɻ ޙड़ͷ EMR or ECS + IAM ͳͲͷ૊Έ߹ΘͤΕ͹ ࠷৽ͷ GPU ؀ڥΛࣗಈ഑෍͢Δͱ͍ͬͨ͜ͱ΋ʂ AWS Ͱͷ෼ੳ؀ڥ 66

Slide 67

Slide 67 text

AWS ͷؔ࿈αʔϏε܈

Slide 68

Slide 68 text

ࢼߦࡨޡϑΣʔζʹศརͳαʔϏε • Amazon Machine Learning • Amazon EMR / EC2 ‣ p2 / g2 (GPU) instances ‣ Deep Learning AMI • Amazon EBS / S3 / ECR 68

Slide 69

Slide 69 text

ೋ߲෼ྨɺෳ਺Ϋϥε෼ྨɺઢܗճؼͷϚωʔδυαʔ ϏεɻσʔλͷऔΓࠐΈ͔Βɺֶशɾਪ࿦͕ߦ͑·͢ɻ αʔόͷ؅ཧ͕ෆཁͳͨΊɺεέʔϥϏϦςΟ΍ਪ࿦αʔ ϏεͷՄ༻ੑ͸ؾʹͤͣ OKʂ Amazon Machine Learning 69

Slide 70

Slide 70 text

GPU Πϯελϯε 70 AWS ʹ͸ 2 छྨ͋Γ·͢ʢݱߦੈ୅ʣ g2 ܥ: NVIDIA GRID K520 ɹɹɹɹ1,536 CUDA cores / GPU ͕ 2 ͭͰ 1 ͭͷ K520 ɹɹɹɹg2 Ͱ࢖͑Δ GPU ͸ຊདྷάϥϑΟοΫɾήʔϛϯά༻్ p2 ܥ: NVIDIA Tesla K80 ɹɹɹɹഒਫ਼౓ԋࢉ࠷େ 2.91 TFLOPSɺ୯ਫ਼౓ԋࢉ࠷େ 8.74 TFLOPS ɹɹɹɹ2,496 CUDA cores / GPU ͕ 2 ͭͰ 1 ͭͷ K80 ɹɹɹɹp2 ͷ GPU ͸൚༻ίϯϐϡʔςΟϯά༻్

Slide 71

Slide 71 text

Amazon DeepLearning AMI 71 શ෦ೖΓ AMIɺ͋Γ·͢ʂʂ TensorFlow 1.0, MXNet, Caffe, CNTK, Theano, Torchɻ CUDA 7.5, cuDNN 5.0, Anaconda ΋ɻ

Slide 72

Slide 72 text

ֶश

Slide 73

Slide 73 text

3ɹֶश ΋ͬͱ΋ॏཁɺ͔ͭ΋ͷ͕͔͔࣌ؒ͘͢͝Δͱ͜Ζɻ ࣗࣾͷϞσϧΛ࡞Δͷ͸ͱͯ΋େม͕ͩ ΦϦδφϧͷ΋ͷ͕Ͱ͖Ε͹وॏͳ஌ࡒʹɻ Ϋϥ΢υͷ༷ʑͳαʔϏε͕αϙʔτͯ͘͠Ε·͢ɻ 73

Slide 74

Slide 74 text

AWS ͷؔ࿈αʔϏε܈

Slide 75

Slide 75 text

ֶशϑΣʔζʹศརͳαʔϏε • Amazon EMR / Batch / EC2 ‣ p2 / g2 (GPU) instances ‣ Deep Learning AMI ‣ Spot Fleet / AutoScaling Group • Amazon Machine Learning • Amazon EFS / EBS / S3 / ECR • Amazon SQS 75

Slide 76

Slide 76 text

AWS Batch 76 https://www.youtube.com/watch?v=UR8BI2Exkbc Պֶٕज़ܭࢉɾϋΠύϑΥʔϚϯείϯϐϡʔςΟϯά ༻్ͰਅՁΛൃش͢Δɺେن໛ͳεέʔϧɺδϣϒͷґ ଘఆ͕ٛՄೳͳϚωʔδυฒྻ෼ࢄόονॲཧج൫ɻ

Slide 77

Slide 77 text

͢Ͱʹ Black Belt ͷࢿྉ͕ެ։͞Ε͍ͯ·͢ɻ AWS Batch http://aws.typepad.com/sajp/2017/02/aws-black-belt-online-seminar-aws-batch.html 77

Slide 78

Slide 78 text

ࢲ΋Ϣʔβࢹ఺Ͱݱঢ়Λ·ͱΊ·ͨ͠ɻ AWS Batch http://qiita.com/pottava/items/d9886b2e8835c5c0d30f 78

Slide 79

Slide 79 text

ਪ࿦

Slide 80

Slide 80 text

4ɹਪ࿦ ֶशࡁΈͷϞσϧΛ࢖͍ɺਪ࿦͢Δɻ Ϗδωεͱ௚݁͢Δ͜ͱ͕ଟ͘ɺՔಇ͸ 24 / 365ɻ Մ༻ੑͱϨΠςϯγ͕ॏཁͳͷ͸ҰൠαʔϏεಉ༷ɻ ΋͔ͯ͠͠αʔόϨεͰ΋ɾɾ͍͚Δɾɾɾʁ 80

Slide 81

Slide 81 text

AWS ͷؔ࿈αʔϏε܈

Slide 82

Slide 82 text

ਪ࿦ϑΣʔζʹศརͳαʔϏε • Amazon ECS / EC2 ‣ p2 / g2 (GPU) instances ‣ Deep Learning AMI ‣ Spot Fleet / AutoScaling Group • AWS Lambda / Amazon API Gateway • AWS ElasticBeanstalk • Amazon EFS / EBS / S3 / ECR 82

Slide 83

Slide 83 text

ϑϧϚωʔδυͳ Docker ίϯςφΫϥελ؀ڥɻ GPU ϕʔεͷਪ࿦ΞϓϦέʔγϣϯͩͬͯಈ͖·͢ʂ Amazon ECS 83 https://speakerdeck.com/ayemos/build-image-classiﬁcation-service-with-amazon-ecs-and-gpu-instances ΫοΫύουגࣜձࣾ છ୩ ༔Ұ࿠͞Μ

Slide 84

Slide 84 text

ͪ͜Β΋͢Ͱʹ Black Belt ͷࢿྉ͕ެ։͞Ε͍ͯ·͢ɻ Amazon ECS 84 http://aws.typepad.com/sajp/2017/02/aws-black-belt-online-seminar-aws-batch.html

Slide 85

Slide 85 text

AWS Lambda 85 αʔόϨεͰ MXNet ʹΑΔਪ࿦Λ͢Δ࣮૷ྫ http://aws.typepad.com/sajp/2017/01/ seamlessly-scale-predictions-with-aws-lambda- and-mxnet.html

Slide 86

Slide 86 text

AWS ར༻ Tips

Slide 87

Slide 87 text

ػցֶशΛ AWS Ͱ΍ΔͳΒɺ஌ͬͯಘ͢Δػೳ No. 1ʂ Ծ૝αʔόΛ҆͘࢖͑Δىಈํ๏ɻ AWS ͷσʔληϯλͷʮ༨৒෼ʯΛ ͜ͷֹۚͳΒ࢖͍·͢ʂͱʮೖࡳʯͯ͠ىಈɻ Spot Fleet / Spot Πϯελϯε 87

Slide 88

Slide 88 text

͜Ε·Ͱ঺հ͖ͯͨ͠ AWS ͷ֤αʔϏεΛ Ͳ͏࢖͍͍͔ͨΛ yaml / json Ͱએݴతʹهड़͓͖ͯ͠ Ұؾʹੜ੒ɾഁغ͢Δ͜ͱ͕Ͱ͖Δɻ ؆୯ਝ଎ʹɺ҆৺ͯ͠؀ڥ͕ߏஙͰ͖Δɻ ӡ༻ෛՙԼ͕Γ·͢ɻΠϯϑϥΛόʔδϣϯ؅ཧՄೳʹɻ CloudFormation 88

Slide 89

Slide 89 text

Let’s try, anyway!

Slide 90

Slide 90 text

ؼͬͨΒࣗ෼ͰҰ࿈ͷྲྀΕΛܦݧʂʂ࣮ફେࣄɻ ʮ਺ࣈը૾൑ఆ with TensorFlow on AWSʯ ࠓ೔ͷ॓୊ http://qiita.com/pottava/items/2fb2572f7099d432ebd9 90

Slide 91

Slide 91 text

ػցֶशʹͲͷΫϥ΢υΛ࢖͏͔ʁͱߟ͑ΔΑΓ ৄ͍͠ਓΛั·͑ͯɺͬ͞ͱ৘ใΛूΊͯ ͍͍ͱ͜ͲΓͰ࢖͏ͷ͕Α͍ͱࢥ͍·͢ɻ   ͱ͍͏͔ɺࢼ͚ͩ͢ͳΒ ϩʔΧϧʹ؀ڥΛ੔͑Ε͹े෼Ͱ͢ɾɾ ·ͱΊ͡Όͳ͍ɺ·ͱΊ 91

Slide 92

Slide 92 text

JAWS-UG AI ࢧ෦

Slide 93

Slide 93 text

ίϯςϯπ • AWS Ͱ AI αʔϏεΛ࣮૷ɾӡ༻͢ΔͨΊͷ ɹҰൠతͳٕज़৘ใɺ஌ݟɺࣄྫڞ༗ͷ৔ • ͢Ͱʹ׆༻͍ͯ͠Δํ • ಋೖΛݕ౼͍ͯ͠Δํ • ԿͦΕ͓͍͍͠ͷʁͳํʢ։࠵͝ͱʹ೉қ౓͕ଟগҧ͍·͢ʣ

Slide 94

Slide 94 text

͋Γ͕ͱ͏͍͟͝·ͨ͠ ࢀߟจݙ: • AWS Batch – ؆୯ʹ࢖͑ͯޮ཰తͳόονίϯϐϡʔςΟϯάػೳ – AWS https://aws.amazon.com/jp/batch/ • AWS Black Belt Online SeminarʮAWS Batchʯͷࢿྉ͓ΑͼQAެ։ http://aws.typepad.com/sajp/2017/02/aws-black-belt-online-seminar-aws- batch.html#QCPzBdn.twitter_tweet_count_m • re:Invent 2016: AWS Big Data & Machine Learning Sessionsɻ https://aws.amazon.com/blogs/big-data/reinvent-2016-aws-big-data- machine-learning-sessions/