ruby-dnn 0.10.4 → 0.12.4
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +4 -4
- data/.travis.yml +1 -2
- data/README.md +33 -6
- data/examples/cifar100_example.rb +3 -3
- data/examples/cifar10_example.rb +3 -3
- data/examples/dcgan/dcgan.rb +112 -0
- data/examples/dcgan/imgen.rb +20 -0
- data/examples/dcgan/train.rb +41 -0
- data/examples/iris_example.rb +3 -6
- data/examples/mnist_conv2d_example.rb +5 -5
- data/examples/mnist_define_by_run.rb +52 -0
- data/examples/mnist_example.rb +3 -3
- data/examples/mnist_lstm_example.rb +3 -3
- data/examples/xor_example.rb +4 -5
- data/ext/rb_stb_image/rb_stb_image.c +103 -0
- data/lib/dnn.rb +10 -10
- data/lib/dnn/cifar10.rb +1 -1
- data/lib/dnn/cifar100.rb +1 -1
- data/lib/dnn/core/activations.rb +21 -22
- data/lib/dnn/core/cnn_layers.rb +94 -111
- data/lib/dnn/core/embedding.rb +30 -9
- data/lib/dnn/core/initializers.rb +31 -21
- data/lib/dnn/core/iterator.rb +52 -0
- data/lib/dnn/core/layers.rb +99 -66
- data/lib/dnn/core/link.rb +24 -0
- data/lib/dnn/core/losses.rb +69 -59
- data/lib/dnn/core/merge_layers.rb +71 -0
- data/lib/dnn/core/models.rb +393 -0
- data/lib/dnn/core/normalizations.rb +27 -14
- data/lib/dnn/core/optimizers.rb +212 -134
- data/lib/dnn/core/param.rb +8 -6
- data/lib/dnn/core/regularizers.rb +10 -7
- data/lib/dnn/core/rnn_layers.rb +78 -85
- data/lib/dnn/core/utils.rb +6 -3
- data/lib/dnn/downloader.rb +3 -3
- data/lib/dnn/fashion-mnist.rb +89 -0
- data/lib/dnn/image.rb +57 -18
- data/lib/dnn/iris.rb +1 -3
- data/lib/dnn/mnist.rb +38 -34
- data/lib/dnn/version.rb +1 -1
- data/third_party/stb_image.h +16 -4
- data/third_party/stb_image_resize.h +2630 -0
- data/third_party/stb_image_write.h +4 -7
- metadata +12 -4
- data/lib/dnn/core/dataset.rb +0 -34
- data/lib/dnn/core/model.rb +0 -440
checksums.yaml
CHANGED
@@ -1,7 +1,7 @@
|
|
1
1
|
---
|
2
2
|
SHA256:
|
3
|
-
metadata.gz:
|
4
|
-
data.tar.gz:
|
3
|
+
metadata.gz: 2835cd619265328ffa05c41515915aaeb54924ca4ddd4958edba992ea4feda89
|
4
|
+
data.tar.gz: 5ab01dbcf1219cdfac7b3a0e20bc5783fa15baaf82225ea1a9a2d076569cb7f1
|
5
5
|
SHA512:
|
6
|
-
metadata.gz:
|
7
|
-
data.tar.gz:
|
6
|
+
metadata.gz: 1335b77cf701501930e32f07a9faa027706b908d79a24b1b369228b45c5ef15a38ac476c6951d14b7ce398b1eb7789bf0c96566427b84fffb1ed892af7e43629
|
7
|
+
data.tar.gz: baf8ceeec91699a75a14bde79e815561189e26c9317e34f09580f90f45562120ba12540a7a64e42e562d4d26162490b6c2170721459b8e635731e900812a357b
|
data/.travis.yml
CHANGED
data/README.md
CHANGED
@@ -26,7 +26,7 @@ Or install it yourself as:
|
|
26
26
|
### MNIST MLP example
|
27
27
|
|
28
28
|
```ruby
|
29
|
-
model =
|
29
|
+
model = Sequential.new
|
30
30
|
|
31
31
|
model << InputLayer.new(784)
|
32
32
|
|
@@ -38,10 +38,38 @@ model << ReLU.new
|
|
38
38
|
|
39
39
|
model << Dense.new(10)
|
40
40
|
|
41
|
-
model.
|
41
|
+
model.setup(RMSProp.new, SoftmaxCrossEntropy.new)
|
42
42
|
|
43
43
|
model.train(x_train, y_train, 10, batch_size: 100, test: [x_test, y_test])
|
44
|
+
```
|
45
|
+
|
46
|
+
When create a model with 'define by run' style:
|
44
47
|
|
48
|
+
```ruby
|
49
|
+
class MLP < Model
|
50
|
+
def initialize
|
51
|
+
super
|
52
|
+
@l1 = Dense.new(256)
|
53
|
+
@l2 = Dense.new(256)
|
54
|
+
@l3 = Dense.new(10)
|
55
|
+
end
|
56
|
+
|
57
|
+
def call(x)
|
58
|
+
x = InputLayer.(x)
|
59
|
+
x = @l1.(x)
|
60
|
+
x = ReLU.(x)
|
61
|
+
x = @l2.(x)
|
62
|
+
x = ReLU.(x)
|
63
|
+
x = @l3.(x)
|
64
|
+
x
|
65
|
+
end
|
66
|
+
end
|
67
|
+
|
68
|
+
model = MLP.new
|
69
|
+
|
70
|
+
model.setup(Adam.new, SoftmaxCrossEntropy.new)
|
71
|
+
|
72
|
+
model.train(x_train, y_train, 10, batch_size: 100, test: [x_test, y_test])
|
45
73
|
```
|
46
74
|
|
47
75
|
Please refer to examples for basic usage.
|
@@ -50,14 +78,13 @@ If you want to know more detailed information, please refer to the source code.
|
|
50
78
|
## Implemented
|
51
79
|
|| Implemented classes |
|
52
80
|
|:-----------|------------:|
|
53
|
-
| Connections | Dense, Conv2D,
|
81
|
+
| Connections | Dense, Conv2D, Conv2DTranspose, Embedding, SimpleRNN, LSTM, GRU |
|
54
82
|
| Layers | Flatten, Reshape, Dropout, BatchNormalization, MaxPool2D, AvgPool2D, UnPool2D |
|
55
83
|
| Activations | Sigmoid, Tanh, Softsign, Softplus, Swish, ReLU, LeakyReLU, ELU |
|
56
|
-
| Optimizers | SGD, Nesterov, AdaGrad, RMSProp, AdaDelta, Adam,
|
57
|
-
| Losses | MeanSquaredError, MeanAbsoluteError, HuberLoss, SoftmaxCrossEntropy, SigmoidCrossEntropy |
|
84
|
+
| Optimizers | SGD, Nesterov, AdaGrad, RMSProp, AdaDelta, RMSPropGraves, Adam, AdaBound |
|
85
|
+
| Losses | MeanSquaredError, MeanAbsoluteError, Hinge, HuberLoss, SoftmaxCrossEntropy, SigmoidCrossEntropy |
|
58
86
|
|
59
87
|
## TODO
|
60
|
-
● Support to define by run model.
|
61
88
|
● Write a test.
|
62
89
|
● Write a document.
|
63
90
|
● Support to GPU.
|
@@ -7,7 +7,7 @@ include DNN::Layers
|
|
7
7
|
include DNN::Activations
|
8
8
|
include DNN::Optimizers
|
9
9
|
include DNN::Losses
|
10
|
-
|
10
|
+
include DNN::Models
|
11
11
|
CIFAR100 = DNN::CIFAR100
|
12
12
|
|
13
13
|
x_train, y_train = CIFAR100.load_train
|
@@ -25,7 +25,7 @@ y_test = y_test[true, 1]
|
|
25
25
|
y_train = DNN::Utils.to_categorical(y_train, 100, Numo::SFloat)
|
26
26
|
y_test = DNN::Utils.to_categorical(y_test, 100, Numo::SFloat)
|
27
27
|
|
28
|
-
model =
|
28
|
+
model = Sequential.new
|
29
29
|
|
30
30
|
model << InputLayer.new([32, 32, 3])
|
31
31
|
|
@@ -66,6 +66,6 @@ model << Dropout.new(0.5)
|
|
66
66
|
|
67
67
|
model << Dense.new(100)
|
68
68
|
|
69
|
-
model.
|
69
|
+
model.setup(Adam.new, SoftmaxCrossEntropy.new)
|
70
70
|
|
71
71
|
model.train(x_train, y_train, 10, batch_size: 100, test: [x_test, y_test])
|
data/examples/cifar10_example.rb
CHANGED
@@ -7,7 +7,7 @@ include DNN::Layers
|
|
7
7
|
include DNN::Activations
|
8
8
|
include DNN::Optimizers
|
9
9
|
include DNN::Losses
|
10
|
-
|
10
|
+
include DNN::Models
|
11
11
|
CIFAR10 = DNN::CIFAR10
|
12
12
|
|
13
13
|
x_train, y_train = CIFAR10.load_train
|
@@ -22,7 +22,7 @@ x_test /= 255
|
|
22
22
|
y_train = DNN::Utils.to_categorical(y_train, 10, Numo::SFloat)
|
23
23
|
y_test = DNN::Utils.to_categorical(y_test, 10, Numo::SFloat)
|
24
24
|
|
25
|
-
model =
|
25
|
+
model = Sequential.new
|
26
26
|
|
27
27
|
model << InputLayer.new([32, 32, 3])
|
28
28
|
|
@@ -63,6 +63,6 @@ model << Dropout.new(0.5)
|
|
63
63
|
|
64
64
|
model << Dense.new(10)
|
65
65
|
|
66
|
-
model.
|
66
|
+
model.setup(Adam.new, SoftmaxCrossEntropy.new)
|
67
67
|
|
68
68
|
model.train(x_train, y_train, 10, batch_size: 100, test: [x_test, y_test])
|
@@ -0,0 +1,112 @@
|
|
1
|
+
include DNN::Layers
|
2
|
+
include DNN::Activations
|
3
|
+
include DNN::Optimizers
|
4
|
+
include DNN::Losses
|
5
|
+
include DNN::Models
|
6
|
+
|
7
|
+
class Generator < Model
|
8
|
+
def initialize
|
9
|
+
super
|
10
|
+
@l1 = Dense.new(1024)
|
11
|
+
@l2 = Dense.new(7 * 7 * 64)
|
12
|
+
@l3 = Conv2DTranspose.new(64, 4, strides: 2, padding: true)
|
13
|
+
@l4 = Conv2D.new(64, 4, padding: true)
|
14
|
+
@l5 = Conv2DTranspose.new(32, 4, strides: 2, padding: true)
|
15
|
+
@l6 = Conv2D.new(32, 4, padding: true)
|
16
|
+
@l7 = Conv2D.new(1, 4, padding: true)
|
17
|
+
@bn1 = BatchNormalization.new
|
18
|
+
@bn2 = BatchNormalization.new
|
19
|
+
@bn3 = BatchNormalization.new
|
20
|
+
@bn4 = BatchNormalization.new
|
21
|
+
@bn5 = BatchNormalization.new
|
22
|
+
@bn6 = BatchNormalization.new
|
23
|
+
end
|
24
|
+
|
25
|
+
def call(x)
|
26
|
+
x = InputLayer.new(20).(x)
|
27
|
+
x = @l1.(x)
|
28
|
+
x = @bn1.(x)
|
29
|
+
x = ReLU.(x)
|
30
|
+
|
31
|
+
x = @l2.(x)
|
32
|
+
x = @bn2.(x)
|
33
|
+
x = ReLU.(x)
|
34
|
+
|
35
|
+
x = Reshape.(x, [7, 7, 64])
|
36
|
+
x = @l3.(x)
|
37
|
+
x = @bn3.(x)
|
38
|
+
x = ReLU.(x)
|
39
|
+
|
40
|
+
x = @l4.(x)
|
41
|
+
x = @bn4.(x)
|
42
|
+
x = ReLU.(x)
|
43
|
+
|
44
|
+
x = @l5.(x)
|
45
|
+
x = @bn5.(x)
|
46
|
+
x = ReLU.(x)
|
47
|
+
|
48
|
+
x = @l6.(x)
|
49
|
+
x = @bn6.(x)
|
50
|
+
x = ReLU.(x)
|
51
|
+
|
52
|
+
x = @l7.(x)
|
53
|
+
x = Tanh.(x)
|
54
|
+
x
|
55
|
+
end
|
56
|
+
end
|
57
|
+
|
58
|
+
|
59
|
+
class Discriminator < Model
|
60
|
+
def initialize
|
61
|
+
super
|
62
|
+
@l1 = Conv2D.new(32, 4, strides: 2, padding: true)
|
63
|
+
@l2 = Conv2D.new(32, 4, padding: true)
|
64
|
+
@l3 = Conv2D.new(64, 4, strides: 2, padding: true)
|
65
|
+
@l4 = Conv2D.new(64, 4, padding: true)
|
66
|
+
@l5 = Dense.new(1024)
|
67
|
+
@l6 = Dense.new(1)
|
68
|
+
end
|
69
|
+
|
70
|
+
def call(x, trainable = true)
|
71
|
+
has_param_layers.each do |layer|
|
72
|
+
layer.trainable = trainable
|
73
|
+
end
|
74
|
+
x = InputLayer.new([28, 28, 1]).(x)
|
75
|
+
x = @l1.(x)
|
76
|
+
x = LeakyReLU.(x, 0.2)
|
77
|
+
|
78
|
+
x = @l2.(x)
|
79
|
+
x = LeakyReLU.(x, 0.2)
|
80
|
+
|
81
|
+
x = @l3.(x)
|
82
|
+
x = LeakyReLU.(x, 0.2)
|
83
|
+
|
84
|
+
x = @l4.(x)
|
85
|
+
x = LeakyReLU.(x, 0.2)
|
86
|
+
|
87
|
+
x = Flatten.(x)
|
88
|
+
x = @l5.(x)
|
89
|
+
x = LeakyReLU.(x, 0.2)
|
90
|
+
|
91
|
+
x = @l6.(x)
|
92
|
+
x
|
93
|
+
end
|
94
|
+
end
|
95
|
+
|
96
|
+
|
97
|
+
class DCGAN < Model
|
98
|
+
attr_reader :gen
|
99
|
+
attr_reader :dis
|
100
|
+
|
101
|
+
def initialize(gen, dis)
|
102
|
+
super()
|
103
|
+
@gen = gen
|
104
|
+
@dis = dis
|
105
|
+
end
|
106
|
+
|
107
|
+
def call(x)
|
108
|
+
x = @gen.(x)
|
109
|
+
x = @dis.(x, false)
|
110
|
+
x
|
111
|
+
end
|
112
|
+
end
|
@@ -0,0 +1,20 @@
|
|
1
|
+
require "dnn"
|
2
|
+
require "dnn/image"
|
3
|
+
require "numo/linalg/autoloader"
|
4
|
+
require_relative "dcgan"
|
5
|
+
|
6
|
+
Image = DNN::Image
|
7
|
+
|
8
|
+
batch_size = 100
|
9
|
+
|
10
|
+
dcgan = Model.load("trained/dcgan_model_epoch20.marshal")
|
11
|
+
gen = dcgan.gen
|
12
|
+
|
13
|
+
Numo::SFloat.srand(rand(1 << 31))
|
14
|
+
noise = Numo::SFloat.new(batch_size, 20).rand(-1, 1)
|
15
|
+
images = gen.predict(noise)
|
16
|
+
|
17
|
+
batch_size.times do |i|
|
18
|
+
img = Numo::UInt8.cast(((images[i, false] + 1) * 127.5).round)
|
19
|
+
Image.write("img/img_#{i}.jpg", img)
|
20
|
+
end
|
@@ -0,0 +1,41 @@
|
|
1
|
+
require "dnn"
|
2
|
+
require "dnn/mnist"
|
3
|
+
require "numo/linalg/autoloader"
|
4
|
+
require_relative "dcgan"
|
5
|
+
|
6
|
+
MNIST = DNN::MNIST
|
7
|
+
|
8
|
+
Numo::SFloat.srand(rand(1 << 31))
|
9
|
+
|
10
|
+
epochs = 20
|
11
|
+
batch_size = 128
|
12
|
+
|
13
|
+
gen = Generator.new
|
14
|
+
dis = Discriminator.new
|
15
|
+
dcgan = DCGAN.new(gen, dis)
|
16
|
+
|
17
|
+
dis.setup(Adam.new(alpha: 0.00001, beta1: 0.1), SigmoidCrossEntropy.new)
|
18
|
+
dcgan.setup(Adam.new(alpha: 0.0002, beta1: 0.5), SigmoidCrossEntropy.new)
|
19
|
+
|
20
|
+
x_train, y_train = MNIST.load_train
|
21
|
+
x_train = Numo::SFloat.cast(x_train)
|
22
|
+
x_train = x_train / 127.5 - 1
|
23
|
+
|
24
|
+
iter = DNN::Iterator.new(x_train, y_train)
|
25
|
+
(1..epochs).each do |epoch|
|
26
|
+
puts "epoch: #{epoch}"
|
27
|
+
iter.foreach(batch_size) do |x_batch, y_batch, index|
|
28
|
+
noise = Numo::SFloat.new(batch_size, 20).rand(-1, 1)
|
29
|
+
images = gen.predict(noise)
|
30
|
+
x = x_batch.concatenate(images)
|
31
|
+
y = Numo::SFloat.cast([1] * batch_size + [0] * batch_size).reshape(batch_size * 2, 1)
|
32
|
+
dis_loss = dis.train_on_batch(x, y)
|
33
|
+
|
34
|
+
noise = Numo::SFloat.new(batch_size, 20).rand(-1, 1)
|
35
|
+
label = Numo::SFloat.cast([1] * batch_size).reshape(batch_size, 1)
|
36
|
+
dcgan_loss = dcgan.train_on_batch(noise, label)
|
37
|
+
|
38
|
+
puts "index: #{index}, dis_loss: #{dis_loss.mean}, dcgan_loss: #{dcgan_loss.mean}"
|
39
|
+
end
|
40
|
+
dcgan.save("trained/dcgan_model_epoch#{epoch}.marshal")
|
41
|
+
end
|
data/examples/iris_example.rb
CHANGED
@@ -7,20 +7,17 @@ include DNN::Layers
|
|
7
7
|
include DNN::Activations
|
8
8
|
include DNN::Optimizers
|
9
9
|
include DNN::Losses
|
10
|
-
|
10
|
+
include DNN::Models
|
11
11
|
Iris = DNN::Iris
|
12
12
|
|
13
13
|
x, y = Iris.load(true)
|
14
14
|
x_train, y_train = x[0...100, true], y[0...100]
|
15
15
|
x_test, y_test = x[100...150, true], y[100...150]
|
16
16
|
|
17
|
-
x_train /= 255
|
18
|
-
x_test /= 255
|
19
|
-
|
20
17
|
y_train = DNN::Utils.to_categorical(y_train, 3, Numo::SFloat)
|
21
18
|
y_test = DNN::Utils.to_categorical(y_test, 3, Numo::SFloat)
|
22
19
|
|
23
|
-
model =
|
20
|
+
model = Sequential.new
|
24
21
|
|
25
22
|
model << InputLayer.new(4)
|
26
23
|
|
@@ -29,6 +26,6 @@ model << ReLU.new
|
|
29
26
|
|
30
27
|
model << Dense.new(3)
|
31
28
|
|
32
|
-
model.
|
29
|
+
model.setup(Adam.new, SoftmaxCrossEntropy.new)
|
33
30
|
|
34
31
|
model.train(x_train, y_train, 1000, batch_size: 10, test: [x_test, y_test])
|
@@ -7,14 +7,14 @@ include DNN::Layers
|
|
7
7
|
include DNN::Activations
|
8
8
|
include DNN::Optimizers
|
9
9
|
include DNN::Losses
|
10
|
-
|
10
|
+
include DNN::Models
|
11
11
|
MNIST = DNN::MNIST
|
12
12
|
|
13
13
|
x_train, y_train = MNIST.load_train
|
14
14
|
x_test, y_test = MNIST.load_test
|
15
15
|
|
16
|
-
x_train = Numo::SFloat.cast(x_train)
|
17
|
-
x_test = Numo::SFloat.cast(x_test)
|
16
|
+
x_train = Numo::SFloat.cast(x_train)
|
17
|
+
x_test = Numo::SFloat.cast(x_test)
|
18
18
|
|
19
19
|
x_train /= 255
|
20
20
|
x_test /= 255
|
@@ -22,7 +22,7 @@ x_test /= 255
|
|
22
22
|
y_train = DNN::Utils.to_categorical(y_train, 10, Numo::SFloat)
|
23
23
|
y_test = DNN::Utils.to_categorical(y_test, 10, Numo::SFloat)
|
24
24
|
|
25
|
-
model =
|
25
|
+
model = Sequential.new
|
26
26
|
|
27
27
|
model << InputLayer.new([28, 28, 1])
|
28
28
|
|
@@ -45,6 +45,6 @@ model << Dropout.new(0.5)
|
|
45
45
|
|
46
46
|
model << Dense.new(10)
|
47
47
|
|
48
|
-
model.
|
48
|
+
model.setup(Adam.new, SoftmaxCrossEntropy.new)
|
49
49
|
|
50
50
|
model.train(x_train, y_train, 10, batch_size: 100, test: [x_test, y_test])
|
@@ -0,0 +1,52 @@
|
|
1
|
+
require "dnn"
|
2
|
+
require "dnn/mnist"
|
3
|
+
# If you use numo/linalg then please uncomment out.
|
4
|
+
# require "numo/linalg/autoloader"
|
5
|
+
|
6
|
+
include DNN::Layers
|
7
|
+
include DNN::Activations
|
8
|
+
include DNN::Optimizers
|
9
|
+
include DNN::Losses
|
10
|
+
include DNN::Models
|
11
|
+
MNIST = DNN::MNIST
|
12
|
+
|
13
|
+
x_train, y_train = MNIST.load_train
|
14
|
+
x_test, y_test = MNIST.load_test
|
15
|
+
|
16
|
+
x_train = Numo::SFloat.cast(x_train).reshape(x_train.shape[0], 784)
|
17
|
+
x_test = Numo::SFloat.cast(x_test).reshape(x_test.shape[0], 784)
|
18
|
+
|
19
|
+
x_train /= 255
|
20
|
+
x_test /= 255
|
21
|
+
|
22
|
+
y_train = DNN::Utils.to_categorical(y_train, 10, Numo::SFloat)
|
23
|
+
y_test = DNN::Utils.to_categorical(y_test, 10, Numo::SFloat)
|
24
|
+
|
25
|
+
class MLP < Model
|
26
|
+
def initialize
|
27
|
+
super
|
28
|
+
@l1 = Dense.new(256)
|
29
|
+
@l2 = Dense.new(256)
|
30
|
+
@l3 = Dense.new(10)
|
31
|
+
@bn1 = BatchNormalization.new
|
32
|
+
@bn2 = BatchNormalization.new
|
33
|
+
end
|
34
|
+
|
35
|
+
def call(x)
|
36
|
+
x = InputLayer.(x)
|
37
|
+
x = @l1.(x)
|
38
|
+
x = @bn1.(x)
|
39
|
+
x = ReLU.(x)
|
40
|
+
x = @l2.(x)
|
41
|
+
x = @bn2.(x)
|
42
|
+
x = ReLU.(x)
|
43
|
+
x = @l3.(x)
|
44
|
+
x
|
45
|
+
end
|
46
|
+
end
|
47
|
+
|
48
|
+
model = MLP.new
|
49
|
+
|
50
|
+
model.setup(Adam.new, SoftmaxCrossEntropy.new)
|
51
|
+
|
52
|
+
model.train(x_train, y_train, 10, batch_size: 100, test: [x_test, y_test])
|
data/examples/mnist_example.rb
CHANGED
@@ -7,7 +7,7 @@ include DNN::Layers
|
|
7
7
|
include DNN::Activations
|
8
8
|
include DNN::Optimizers
|
9
9
|
include DNN::Losses
|
10
|
-
|
10
|
+
include DNN::Models
|
11
11
|
MNIST = DNN::MNIST
|
12
12
|
|
13
13
|
x_train, y_train = MNIST.load_train
|
@@ -22,7 +22,7 @@ x_test /= 255
|
|
22
22
|
y_train = DNN::Utils.to_categorical(y_train, 10, Numo::SFloat)
|
23
23
|
y_test = DNN::Utils.to_categorical(y_test, 10, Numo::SFloat)
|
24
24
|
|
25
|
-
model =
|
25
|
+
model = Sequential.new
|
26
26
|
|
27
27
|
model << InputLayer.new(784)
|
28
28
|
|
@@ -34,6 +34,6 @@ model << ReLU.new
|
|
34
34
|
|
35
35
|
model << Dense.new(10)
|
36
36
|
|
37
|
-
model.
|
37
|
+
model.setup(RMSProp.new, SoftmaxCrossEntropy.new)
|
38
38
|
|
39
39
|
model.train(x_train, y_train, 10, batch_size: 100, test: [x_test, y_test])
|