Slide 1
Slide 1 text
Pythonॳ৺ऀ͕PyTorchΛ͍ͬͯ͡
ػցֶशͷܭࢉΛͯ͠Έͨ
Python-newbie’s machine learning learning with PyTorch
㝳๏
PyCon mini Hiroshima 2018
@ౡେֶ ະདྷੜηϯλʔ @201810݄6()
Slide 2
Slide 2 text
ࣗݾհ
Slide 3
Slide 3 text
No content
Slide 4
Slide 4 text
㝳๏
▸ ͖͍ͨͩΏ͖ͷΓ
▸ ʮ๏ʯΛԻಡΈ(༗৬ಡΈ)ͯ͠ňʹΐ΄͏ʼn
▸ ϋΠύɾϝσΟΞɾτϥϯεϨʔλ
▸ ༵ϓϩάϥϚ
Slide 5
Slide 5 text
iPad ౡห ݕࡧ
ৄ͘͠
Slide 6
Slide 6 text
ʮ͜Μͳײ͡ͷ͕ग़͖ͯ·͢ɻʯ
Slide 7
Slide 7 text
Pythonॳ৺ऀ͕PyTorchΛ
͍ͬͯ͡
ػցֶशͷܭࢉΛͯ͠Έͨ
݁Ռwwwwwwwwww
Slide 8
Slide 8 text
Pythonॳ৺ऀ͕PyTorchΛ
͍ͬͯ͡
ػցֶशͷܭࢉΛͯ͠Έͨ
݁ՌΛൃද͠·͢ɻ
Slide 9
Slide 9 text
ʢɾ㱼ɾʣŘŘ!!
Slide 10
Slide 10 text
͝ਗ਼ௌ͋Γ͕ͱ͏͍͟͝·ͨ͠ɻ
Slide 11
Slide 11 text
͏ͪΐͬͱ
ৄ͘͠
Slide 12
Slide 12 text
ຊͷϝχϡʔ
ػցֶश
1Z5PSDI
ͬͯΈͨྫ
Slide 13
Slide 13 text
Pythonॳ৺ऀ?
Slide 14
Slide 14 text
Pythonॳ৺ऀ߹͍
▸ for ͱ͔ if ͱ͔ with ͱ͔ϨϕϧͰຊɾυΩϡϝϯτɾωοτΛݟͳ͕ΒͰ
ͳ͍ͱॻ͚ͳ͍
▸ Ψνͷॳ৺ऀ
Slide 15
Slide 15 text
ػցֶश?
Slide 16
Slide 16 text
CourseraͷNgઌੜͷ
ػցֶशͷίʔε
Slide 17
Slide 17 text
मྃ! (20165݄͝Ζ)
Slide 18
Slide 18 text
͔͠मྃ࣌
ʹNgઌੜ͔Β
Slide 19
Slide 19 text
͔͠मྃ࣌ʹ Andrew Ng ઌੜ͔Β
▸ Those were the topics of this class and if you worked all the way through this course
you should now consider yourself an expert in machine learning.
▸ As you know, machine learning is a technology that's having huge impact on
science, technology and industry.
▸ And you're now well qualified to use these tools of machine learning to great effect.
▸ I hope that many of you in this class will find ways to use machine learning to build
cool systems and cool applications and cool products.
▸ And I hope that you find ways to use machine learning not only to make your life
better but maybe someday to use it to make many other people's life better as well.
Slide 20
Slide 20 text
Ngઌੜ͔ΒઐՈ
ೝఆ͞ΕͱΔ
Slide 21
Slide 21 text
͜Γ͏ؒҧ͍ͳ͘Θ͠
ػցֶशͷઐՈ
Ͱ͢(ތுදݱ)
Slide 22
Slide 22 text
1. ػցֶश
Slide 23
Slide 23 text
ػցֶशͱ
▸ ͬ͘͟Γ
▸ ڭࢣ͋Γֶश (supervised learning)
▸ ڭࢣͳֶ͠श (unsupervised learning)
Slide 24
Slide 24 text
ػցֶशͱ
▸ ͍ΖΜͳσʔλͷഎޙʹજΉ๏ଇͱ͔ߏΛɺऔΓग़ͦ͏ͱ͢Δܭࢉ
▸ ͬ͘͟Γ
▸ ڭࢣ͋Γֶश (supervised learning)
▸ ڭࢣͳֶ͠श (unsupervised learning)
▸ ͷ2छྨ͕͋Δ
Slide 25
Slide 25 text
ڭࢣ͋Γֶश — ػցֶश
▸ σʔλʹରԠ͢Δσʔλ͕͋Δ
▸ ͱ͍͍ͩͨͳΓͦ͏ͳؔ f Λ୳Γ͍ͯͨ
▸ ͔͜͠Ε͔ΒಘΔະͷσʔλʹ͍ͭͯͦ͏ͳͬͯཉ͍͠
x1
, x2
, …, xN
(Nී௨େ͖͍)
y1
, y2
, …, yN xn
↦ yn
yn
= f(xn
)
Slide 26
Slide 26 text
ྫ͑
Slide 27
Slide 27 text
ը૾Λೖྗͨ͠Βɺ
͍ࣸͬͯΔͷ(ೣɾ
ص…)Λ͑Δؔ
Slide 28
Slide 28 text
Γํ
Slide 29
Slide 29 text
શવҧ͏͔ؔΒ࢝Ί
ͯɺ
͡Θ͡Θ͚͍ۙͮͯ͘
Slide 30
Slide 30 text
தֶ2ֶՊ
Slide 31
Slide 31 text
தֶߍ2ੜΛࢥ͍ग़͢
σʔλʹҰ൪߹͏ઢ
0.2 0.4 0.6 0.8 1.0
1.1
1.2
1.3
1.4
1.5
1.6
1.7
y = ax + b
a, bΛܾΊΔ
Slide 32
Slide 32 text
͜͏͍͏ͷ
ཱͳػցֶश
Slide 33
Slide 33 text
͞Βʹผͷ
ؔΛ߹
Slide 34
Slide 34 text
▸ Կ͔ʹԠ͢Δײ͡ͷ͕ؔ࡞ΕΔ
͞ΒʹผͷؔΛ߹
-1.0 -0.5 0.5 1.0 1.5 2.0
0.5
1.0
1.5
2.0
-10 -5 5 10
0.2
0.4
0.6
0.8
1.0
s(x) =
1
1 + e−x
,
s(f(x))
Slide 35
Slide 35 text
ࡢࠓྲྀߦͬͯ
͍Δͷ
Slide 36
Slide 36 text
ਂֶश
Deep learning
Slide 37
Slide 37 text
͖ͬ͞ͷؔΛ
ͨ͘͞Μ༻ҙͯ͠
ͭͳ͛·͘Δ
Slide 38
Slide 38 text
ͭͳ͛·ͬͨ͘ͷ = χϡʔϥϧɾωοτϫʔΫ
▸ ○1ͭ1͕͖ͭͬ͞ͷؔ
▸ ͡Δ͠ͷΑ͏ʹ
▸ ग़ྗΛ࣍ͷೖྗʹ
Slide 39
Slide 39 text
2. PyTorch
Slide 40
Slide 40 text
No content
Slide 41
Slide 41 text
PyTorch
▸ ਂֶश༻ͷϑϨʔϜϫʔΫ
▸ theano, Caffe, TensorFlow, mxnet, Chainer
▸ ͍ͬͺ͍͋Δ͏ͪͷҰͭ
▸ Torch ΑΓ (Torch LuaJIT)
▸ ChainerΛେ͍ʹࢀߟʹ͍ͯ͠Δ
▸ (ʮforkͱݺͼ͚ͨΕͲ͏ͧʯhttps://www.reddit.com/r/MachineLearning/comments/
74md00/n_how_to_use_chainer_for_theano_users/dnzkba1/ )
Slide 42
Slide 42 text
PyTorch
▸ ಛ
▸ υΩϡϝϯτɾνϡʔτϦΞϧ͕݁ߏৄ͍͠
▸ ಈతʹܭࢉάϥϑ࡞
▸ ࣗಈඍ
▸ χϡʔϥϧɾωοτϫʔΫ͕ॻ͖͍͢ (Chainerͱࣅ͍ͯΔ)
Slide 43
Slide 43 text
ॻ͖ͦ͢͏ͳงғؾΛݟΔ
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
self.conv2 = nn.Conv2d(10, 20, kernel_size=5)
self.conv2_drop = nn.Dropout2d()
self.fc1 = nn.Linear(320, 50)
self.fc2 = nn.Linear(50, 10)
def forward(self, x):
x = F.relu(F.max_pool2d(self.conv1(x), 2))
x =
F.relu(F.max_pool2d(self.conv2_drop(self.conv2(x)), 2))
x = x.view(-1, 320)
x = F.relu(self.fc1(x))
x = F.dropout(x, training=self.training)
x = self.fc2(x)
return F.log_softmax(x, dim=1)
͖ͬ͞ͷ˓Λ
Կݸ༻ҙͯ͠
ͭͳ͙͔
ͭͳ͛Δͱ͖ʹ߹͢Δؔ
Slide 44
Slide 44 text
ॻ͖ͦ͢͏ͳงғؾΛݟΔ
▸ random_split
▸ σʔλΛγϟοϑϧׂͭͭͯ͘͠͠ΕΔศརϝιου
train_size = int(0.8 * len(dataset))
test_size = len(dataset) - train_size
trainset, testset = torch.utils.data.random_split(dataset,
[train_size, test_size])
NEW 0.4.1
Slide 45
Slide 45 text
PyTorchདྷͱΔ
▸ Frameworks mentioned @ICLR (International Conference on Learning Representations)
2018➡2019
▸ TensorFlow 228 ➡ 266
▸ Keras: 42 ➡ 56
▸ PyTorch 87 ➡ 252
▸ https://www.reddit.com/r/MachineLearning/comments/9kys38/
r_frameworks_mentioned_iclr_20182019_tensorflow/
▸ ݁ߏݚڀͷίʔυ͕PyTorchͰॻ͔Ε͍ͯΔ
▸ ΦϯϥΠϯߨٛPyTorch͕૿͖͑ͯͨ
Slide 46
Slide 46 text
PyTorch 1.0.0.dev20181002
(1.0 developer preview!)
͏ͪΐͬͱͰ ver.1.0
https://code.fb.com/ai-research/facebook-accelerates-ai-development-with-new-partners-and-production-capabilities-for-pytorch-1-0/
Slide 47
Slide 47 text
3. ͬͯΈͨྫ
Slide 48
Slide 48 text
ᶃΣϒϖʔδྨ
Slide 49
Slide 49 text
Σϒϖʔδྨ?
Slide 50
Slide 50 text
Σϒϖʔδྨ?
▸ ΣϒϖʔδͷςΩετσʔλΛೖྗͱ͢Δ
▸ ͦͷϖʔδͷ༰ͷछྨΛਪఆ͢Δ
▸ ·ͣ࿅शͷͨΊʹʮ2ྨʯ
Slide 51
Slide 51 text
Γํ
▸ ࣗͷʮͯͳϒοΫϚʔΫʯ4ສҎ্ϒοΫϚʔΫ͍ͯ͠Δ
▸ ϒοΫϚʔΫλάΛͤͬͤͱ͚͍ͯΔ
▸ ʮػցֶशʯλάΛ͚͍ͯΔ͔Ͳ͏͔Ͱྨ
▸ 1000Ҏ্͋Δ
▸ ࣗͷϒοΫϚʔΫ͔ΒσʔλΛूΊ͖ͯͯɺֶशɻ
▸ text ➡ MeCab(ܗଶૉղੳ) ➡ fasttext ➡ จॻϕΫτϧʹ͢Δ
Slide 52
Slide 52 text
Γํ
࢈ۀٕज़૯߹ݚڀॴ
20185݄29ɺ౦ژେ
ֶͱڞಉͰɺػցֶशͷֶ
शॲཧʹඞཁͳԋࢉྔΛ
ݮ͢Δܭࢉํࣜͱԋࢉճ࿏
ΛߟҊͨ͠ͱൃදͨ͠ɻ
ࠓقͷΞχϝͷ
͋ͦͼ͋ͦͤͬͯ
໘ന͍Ͷɻ
ػցֶशͷ͜ͱ͕ॻ͍ͯ͋Δϖʔδ͡Ό!
ΣϒϖʔδB
ΣϒϖʔδA
ػցֶश͡Όͳ͍ϖʔδ͡Ό!
Slide 53
Slide 53 text
Demo
Slide 54
Slide 54 text
No content
Slide 55
Slide 55 text
No content
Slide 56
Slide 56 text
ͨͿΜ͏·͍ͬͨ͘!
Slide 57
Slide 57 text
࣍
Slide 58
Slide 58 text
ᶄϒοΫϚʔΫਪఆ
Slide 59
Slide 59 text
ϒοΫϚʔΫਪఆ?
Slide 60
Slide 60 text
ϒοΫϚʔΫਪఆ?
▸ ࣗͷʮͯͳϒοΫϚʔΫʯ4ສҎ্ϒοΫϚʔΫ͍ͯ͠Δ
▸ ֤ϖʔδͷϒοΫϚʔΫ…ଞͷਓ߹ΘͤͯԿਓϒοΫϚʔΫ͔ͨ͠
▸ ·ͨࣗͷϒοΫϚʔΫ͔ΒσʔλΛूΊֶͯश
▸ text ➡ MeCab(ܗଶૉղੳ) ➡ fasttext ➡ จॻϕΫτϧʹ͢Δ
▸ దʹωοτϫʔΫΛֶͭͬͯ͘शͯ͠ΈΔ
Slide 61
Slide 61 text
Demo
Slide 62
Slide 62 text
܇࿅σʔλଛࣦ⤵
ࢼݧσʔλଛࣦ⤴
ʮաֶश͠·͘ΔΈ͍ͨͰ͢ɻʯ
Slide 63
Slide 63 text
(´ɾωɾ`)
શવ͏·ֶ͘श͠ͳ͍ (աֶश͔Γ͢Δ)
Slide 64
Slide 64 text
ϒοΫϚʔΫਪఆจॻͰແཧͰ?
▸ ബʑؾ͍͍ͯͨ
▸ ͯͳϒοΫϚʔΫͷΛϖʔδͷ༰͚͔ͩΒਪଌ͢ΔͷແཧͰɻ
▸ ༗໊ਓ͕֦ࢄͨ͠ͱ͔ɺଞͷӡͷཁૉ͕ڧ͍
▸ Ͱ७ਮʹจষ͚͔ͩΒજࡏతͳʮϒοΫϚʔΫ͞ΕྗʯΈ͍ͨͳͷ͕Ͱ
ͳ͍͔ͳ͋ͱ୶͍ظ͋ͬͨ
▸ ݚڀଓߦ͢Δ͖! ͯϒΛՔ͍͗ͨ!
Slide 65
Slide 65 text
ᶅGAN
Slide 66
Slide 66 text
GAN?
Slide 67
Slide 67 text
GAN?
▸ GAN = generative adversarial network (ఢରతੜωοτϫʔΫ)
▸ ੜऀ generator (G) v.s. ࣝผऀ discriminator (D) ͕ఢର͠ͳ͕Βֶशͯ͠
͍͘
▸ G ͱ D ͕ఢର͠ͳ͕Βֶश͍ͯ͘͠
▸ Կ͔Λʮੜ͢Δʯ
Slide 68
Slide 68 text
ը૾ੜͷ࡞ྫ
Slide 69
Slide 69 text
͜͏͍͏ͷ͕
࡞Γ͍ͨ
▸ ҴѪ߳ͪΌΜͱΘ͠ͷը૾Λ
ֶशͯ͠ɺը૾Λੜ
Slide 70
Slide 70 text
epoch 25
▸ ΄΅ϊΠζ
Slide 71
Slide 71 text
epoch 37
▸ إΒ͖͠ͷ͕ݟ͖͑ͯͨ
Slide 72
Slide 72 text
epoch 149
▸ ੜͰ͖͖ͯͨ!
Slide 73
Slide 73 text
epoch 192
Slide 74
Slide 74 text
ൺֱ
ਖ਼ ϑΣΠΫ
Slide 75
Slide 75 text
ָ͍͠
✌('ω'✌ )ࡾ✌('ω')✌ࡾ( ✌'ω')✌
Slide 76
Slide 76 text
·ͱΊ
▸ Pythonॳ৺ऀ͕PyTorchΛͬͯػցֶशΛͬͯΈͨ
▸ ͔֬ʹॻ͖͍͢ؾ͕͢Δ
▸ ͲΜͲΜ࣮ݧͯ͠༡Δ
▸ ৽͍͠ઃܭ͕࣍ʑग़͍ͯΔͷͰ͍͖͍ͬͯͨ
͋Γ͕ͱ͏͍͟͝·ͨ͠ɻ