ruby_brain 0.1.0
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +7 -0
- data/.gitignore +10 -0
- data/.rspec +2 -0
- data/.travis.yml +5 -0
- data/Gemfile +4 -0
- data/LICENSE.txt +21 -0
- data/README.md +48 -0
- data/README.org +237 -0
- data/Rakefile +6 -0
- data/bin/console +14 -0
- data/bin/setup +8 -0
- data/examples/mnist.rb +79 -0
- data/examples/mnist2.rb +82 -0
- data/lib/ruby_brain.rb +19 -0
- data/lib/ruby_brain/dataset/mnist/data.rb +62 -0
- data/lib/ruby_brain/dataset/mnist/test_mnist.rb +28 -0
- data/lib/ruby_brain/exception.rb +17 -0
- data/lib/ruby_brain/layer.rb +37 -0
- data/lib/ruby_brain/network.rb +252 -0
- data/lib/ruby_brain/nodes.rb +58 -0
- data/lib/ruby_brain/trainer.rb +71 -0
- data/lib/ruby_brain/training_data_manipulator.rb +35 -0
- data/lib/ruby_brain/version.rb +3 -0
- data/lib/ruby_brain/weights.rb +82 -0
- data/ruby_brain.gemspec +34 -0
- metadata +125 -0
checksums.yaml
ADDED
@@ -0,0 +1,7 @@
|
|
1
|
+
---
|
2
|
+
SHA1:
|
3
|
+
metadata.gz: 243add4b6f64c99d4bc2de70f69ff74f8fe439fd
|
4
|
+
data.tar.gz: 979c86696fd11c0892031495e57e7f3746b8df2f
|
5
|
+
SHA512:
|
6
|
+
metadata.gz: a4d8118899133f660ebfb9a64dd1cd068bb7501dff0cf371b6a8937a51fb9d9c0949b45f76176e71404dc745aa766fd9d3853ef34b6015f1d9bc33025df00b48
|
7
|
+
data.tar.gz: e0aec12b42540e424555f50e3bd7784d136fd7d831ee01486967198461d2940d63f174cdcb4a6b7583bb0b5ffdd0aae8968879a45d137d03a07c02b725d13e49
|
data/.gitignore
ADDED
data/.rspec
ADDED
data/.travis.yml
ADDED
data/Gemfile
ADDED
data/LICENSE.txt
ADDED
@@ -0,0 +1,21 @@
|
|
1
|
+
The MIT License (MIT)
|
2
|
+
|
3
|
+
Copyright (c) 2016 TODO: Write your name
|
4
|
+
|
5
|
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
6
|
+
of this software and associated documentation files (the "Software"), to deal
|
7
|
+
in the Software without restriction, including without limitation the rights
|
8
|
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
9
|
+
copies of the Software, and to permit persons to whom the Software is
|
10
|
+
furnished to do so, subject to the following conditions:
|
11
|
+
|
12
|
+
The above copyright notice and this permission notice shall be included in
|
13
|
+
all copies or substantial portions of the Software.
|
14
|
+
|
15
|
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
16
|
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
17
|
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
18
|
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
19
|
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
20
|
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
21
|
+
THE SOFTWARE.
|
data/README.md
ADDED
@@ -0,0 +1,48 @@
|
|
1
|
+
# RubyBrain
|
2
|
+
|
3
|
+
RubyBrain is a library of neural net, deep learning for Ruby.
|
4
|
+
You can install/use this library easily because the core is created by using only Ruby standard library.
|
5
|
+
|
6
|
+
The code of RubyBrain is the neuron oriented style.
|
7
|
+
This means that a class which represents a neuraon exists and each neurons are instances of the class.
|
8
|
+
So, you can treat neurons flexibly in a network.
|
9
|
+
Instead, the speed is very slow and it might not be reasonable for applications to use this library in the core.
|
10
|
+
However this library may help you get more deep knowledge around neuralnet/deep learning.
|
11
|
+
|
12
|
+
## Installation
|
13
|
+
|
14
|
+
Add this line to your application's Gemfile:
|
15
|
+
|
16
|
+
```ruby
|
17
|
+
gem 'ruby_brain'
|
18
|
+
```
|
19
|
+
|
20
|
+
And then execute:
|
21
|
+
|
22
|
+
$ bundle
|
23
|
+
|
24
|
+
Or install it yourself as:
|
25
|
+
|
26
|
+
$ gem install ruby_brain
|
27
|
+
|
28
|
+
## Usage
|
29
|
+
|
30
|
+
Please refer to
|
31
|
+
[github.com/elgoog/ruby_brain/README.org](https://github.com/elgoog/ruby_brain/blob/master/README.org)
|
32
|
+
for detail.
|
33
|
+
|
34
|
+
## Development
|
35
|
+
|
36
|
+
After checking out the repo, run `bin/setup` to install dependencies. Then, run `rake spec` to run the tests. You can also run `bin/console` for an interactive prompt that will allow you to experiment.
|
37
|
+
|
38
|
+
To install this gem onto your local machine, run `bundle exec rake install`. To release a new version, update the version number in `version.rb`, and then run `bundle exec rake release`, which will create a git tag for the version, push git commits and tags, and push the `.gem` file to [rubygems.org](https://rubygems.org).
|
39
|
+
|
40
|
+
## Contributing
|
41
|
+
|
42
|
+
Bug reports and pull requests are welcome on GitHub at https://github.com/elgoog/ruby_brain.
|
43
|
+
|
44
|
+
|
45
|
+
## License
|
46
|
+
|
47
|
+
The gem is available as open source under the terms of the [MIT License](http://opensource.org/licenses/MIT).
|
48
|
+
|
data/README.org
ADDED
@@ -0,0 +1,237 @@
|
|
1
|
+
* Introduction
|
2
|
+
|
3
|
+
RubyBrain is a library of neural net, deep learning for Ruby.
|
4
|
+
You can install/use this library easily because the core is created by using only Ruby standard library.
|
5
|
+
|
6
|
+
The code of RubyBrain is the neuron oriented style.
|
7
|
+
This means that a class which represents a neuraon exists and each neurons are instances of the class.
|
8
|
+
So, you can treat neurons flexibly in a network.
|
9
|
+
Instead, the speed is very slow and it might not be reasonable for applications to use this library in the core.
|
10
|
+
However this library may help you get more deep knowledge around neuralnet/deep learning.
|
11
|
+
|
12
|
+
* Installation
|
13
|
+
|
14
|
+
Add this line to your application's Gemfile:
|
15
|
+
|
16
|
+
#+BEGIN_SRC ruby
|
17
|
+
gem 'ruby_brain'
|
18
|
+
#+END_SRC
|
19
|
+
|
20
|
+
And then execute:
|
21
|
+
|
22
|
+
#+BEGIN_SRC shell
|
23
|
+
$ bundle
|
24
|
+
#+END_SRC
|
25
|
+
|
26
|
+
Or install it yourself as:
|
27
|
+
|
28
|
+
#+BEGIN_SRC shell
|
29
|
+
$ gem install ruby_brain
|
30
|
+
#+END_SRC
|
31
|
+
|
32
|
+
* Usage
|
33
|
+
|
34
|
+
** dataset
|
35
|
+
|
36
|
+
All dataset used for training/test/predicate must be 2 dimension array structure.
|
37
|
+
1st dimension indicates samples and 2nd dimension is used for features.
|
38
|
+
|
39
|
+
*** example "AND"
|
40
|
+
Now we assume that we train a network to operate as "AND operator"
|
41
|
+
The true table of "AND operator" is as below.
|
42
|
+
In short, when both "in 1" and "in 2" are 1, the "out" should be 1
|
43
|
+
and 0 should be output for other input combinations.
|
44
|
+
|
45
|
+
| in 1 | in 2 | out |
|
46
|
+
|------+------+-----|
|
47
|
+
| 0 | 0 | 0 |
|
48
|
+
| 0 | 1 | 0 |
|
49
|
+
| 1 | 0 | 0 |
|
50
|
+
| 1 | 1 | 1 |
|
51
|
+
|
52
|
+
In this situation, you can prepare the dataset lake following Ruby array.
|
53
|
+
#+BEGIN_SRC ruby
|
54
|
+
training_input_set = [
|
55
|
+
[0, 0],
|
56
|
+
[0, 1],
|
57
|
+
[1, 0],
|
58
|
+
[1, 1],
|
59
|
+
]
|
60
|
+
|
61
|
+
training_supervisor_set = [
|
62
|
+
[0],
|
63
|
+
[0],
|
64
|
+
[0],
|
65
|
+
[1],
|
66
|
+
]
|
67
|
+
#+END_SRC
|
68
|
+
|
69
|
+
** constructing a network
|
70
|
+
|
71
|
+
RubyBrain::Network class represents a network. The constructor accepts an array which specifies the network structure.
|
72
|
+
If we use 1 hidden layer which has 3 neurons for above "AND operator" example, following array indicates the structure.
|
73
|
+
#+BEGIN_SRC ruby
|
74
|
+
# 2 inputs
|
75
|
+
# 3 units in a hidden layer
|
76
|
+
# 1 output
|
77
|
+
[2, 3, 1]
|
78
|
+
#+END_SRC
|
79
|
+
|
80
|
+
You can use 2 hidden layers with following code.
|
81
|
+
#+BEGIN_SRC ruby
|
82
|
+
# 2 inputs
|
83
|
+
# 4 units in 1st hidden layer
|
84
|
+
# 3 units in 2nd hidden layer
|
85
|
+
# 1 output
|
86
|
+
[2, 4, 2, 1]
|
87
|
+
#+END_SRC
|
88
|
+
|
89
|
+
So, a netowrk is created by
|
90
|
+
#+BEGIN_SRC ruby
|
91
|
+
a_network = RubyBrain::Network.new([2, 3, 1])
|
92
|
+
|
93
|
+
# learning_rate can be set
|
94
|
+
a_network.learning_rate = 0.5
|
95
|
+
|
96
|
+
# the networks must be initialized before it is used
|
97
|
+
a_network.init_network
|
98
|
+
#+END_SRC
|
99
|
+
|
100
|
+
There are other options for the constructor.
|
101
|
+
Please refer to the code. Sorry for missing document.
|
102
|
+
|
103
|
+
** training
|
104
|
+
|
105
|
+
An instance method =learn= is used for training the network.
|
106
|
+
You can specify not only dataset but also other options for training.
|
107
|
+
#+BEGIN_SRC ruby
|
108
|
+
# max_training_cout : max epoch
|
109
|
+
# tolerance : stop training if RMS error become smaller than this value.
|
110
|
+
a_network.learn(training_input_set, training_supervisor_set, max_training_count=100, tolerance=0.0004, monitoring_channels=[:best_params_training])
|
111
|
+
#+END_SRC
|
112
|
+
|
113
|
+
** predicate
|
114
|
+
|
115
|
+
Use =get_forward_outputs= with input data for predicating something.
|
116
|
+
Input data should be 1 sample.
|
117
|
+
#+BEGIN_SRC ruby
|
118
|
+
a_network.get_forward_outputs([1, 0])
|
119
|
+
#+END_SRC
|
120
|
+
|
121
|
+
** save weights to a file
|
122
|
+
|
123
|
+
You can save optimized weights into a file.
|
124
|
+
Weights are saved as YAML format.
|
125
|
+
|
126
|
+
#+BEGIN_SRC ruby
|
127
|
+
a_network.dump_weights_to_yaml('/path/to/saved/weights/file.yml')
|
128
|
+
#+END_SRC
|
129
|
+
|
130
|
+
** restore weights from a file
|
131
|
+
|
132
|
+
Optimized weights can be saved into a YAML file and you can use it for initializing weights when you create a new network.
|
133
|
+
#+BEGIN_SRC ruby
|
134
|
+
a_network = RubyBrain::Network.new([2, 3, 1])
|
135
|
+
a_network.init_network
|
136
|
+
a_network.load_weights_from_yaml_file('/path/to/saved/weights/file.yml')
|
137
|
+
#+END_SRC
|
138
|
+
|
139
|
+
* Examples
|
140
|
+
|
141
|
+
** MNIST
|
142
|
+
Following code is included in [[https://github.com/elgoog/ruby_brain/blob/master/examples/mnist.rb][examples/mnist.rb]]
|
143
|
+
|
144
|
+
This module dependos on [[https://rubygems.org/gems/mnist][mnist]] gem to load mnist data into ruby array.
|
145
|
+
|
146
|
+
#+BEGIN_SRC ruby
|
147
|
+
require 'ruby_brain'
|
148
|
+
require 'ruby_brain/dataset/mnist/data'
|
149
|
+
#+END_SRC
|
150
|
+
|
151
|
+
Get MNIST dataset from [[http://yann.lecun.com/exdb/mnist/][THE MNIST DATABASE of handwritten digits]] if the dataset files don't exist in the working directory.
|
152
|
+
And load them into Ruby array =dataset=.
|
153
|
+
|
154
|
+
#+BEGIN_SRC ruby
|
155
|
+
dataset = RubyBrain::DataSet::Mnist::data
|
156
|
+
#+END_SRC
|
157
|
+
|
158
|
+
Divide =dataset= into training and test data.
|
159
|
+
NUM_TRAIN_DATA means how many first images are used as training data.
|
160
|
+
We use first 5000 images for training here.
|
161
|
+
|
162
|
+
#+BEGIN_SRC ruby
|
163
|
+
NUM_TRAIN_DATA = 5000
|
164
|
+
training_input = dataset[:input][0..(NUM_TRAIN_DATA-1)]
|
165
|
+
training_supervisor = dataset[:output][0..(NUM_TRAIN_DATA-1)]
|
166
|
+
#+END_SRC
|
167
|
+
|
168
|
+
Then construct the network and initialize.
|
169
|
+
In this case, an image has 784(28x28) pixcels and 10 classes(0..9).
|
170
|
+
So, the network structure should be [784, 50, 10] with 1 hidden layer which has 50 units.
|
171
|
+
You can construct the structure with following code.
|
172
|
+
|
173
|
+
#+BEGIN_SRC ruby
|
174
|
+
# network structure [784, 50, 10]
|
175
|
+
network = RubyBrain::Network.new([dataset[:input].first.size, 50, dataset[:output].first.size])
|
176
|
+
# learning rate is 0.7
|
177
|
+
network.learning_rate = 0.7
|
178
|
+
# initialize network
|
179
|
+
network.init_network
|
180
|
+
#+END_SRC
|
181
|
+
|
182
|
+
Run training.
|
183
|
+
#+BEGIN_SRC ruby
|
184
|
+
network.learn(training_input, training_supervisor, max_training_count=100, tolerance=0.0004, monitoring_channels=[:best_params_training])
|
185
|
+
#+END_SRC
|
186
|
+
|
187
|
+
Now, An optimized network was completed.
|
188
|
+
You can check it.
|
189
|
+
|
190
|
+
First, add =argmax= function into Array class.
|
191
|
+
This method finds the index of the array position the max value exists.
|
192
|
+
We use this method for finding the class(label 0~9) whose probability is the highest.
|
193
|
+
|
194
|
+
#+BEGIN_SRC ruby
|
195
|
+
class Array
|
196
|
+
def argmax
|
197
|
+
max_i, max_val = 0, self.first
|
198
|
+
self.each_with_index do |v, i|
|
199
|
+
max_val, max_i = v, i if v > max_val
|
200
|
+
end
|
201
|
+
max_i
|
202
|
+
end
|
203
|
+
end
|
204
|
+
#+END_SRC
|
205
|
+
|
206
|
+
Then, you can review each classes(labels) predicated by the model with following code.
|
207
|
+
|
208
|
+
#+BEGIN_SRC ruby
|
209
|
+
results = []
|
210
|
+
test_input.each_with_index do |input, i|
|
211
|
+
input.each_with_index do |e, j|
|
212
|
+
print(e > 0.3 ? 'x' : ' ')
|
213
|
+
puts if (j % 28) == 0
|
214
|
+
end
|
215
|
+
puts
|
216
|
+
supervisor_label = test_supervisor[i].argmax
|
217
|
+
predicated_label = network.get_forward_outputs(test_input[i]).argmax
|
218
|
+
puts "test_supervisor: #{supervisor_label}"
|
219
|
+
puts "predicate: #{predicated_label}"
|
220
|
+
results << (supervisor_label == predicated_label)
|
221
|
+
puts "------------------------------------------------------------"
|
222
|
+
end
|
223
|
+
|
224
|
+
puts "accuracy: #{results.count(true).to_f/results.size}"
|
225
|
+
#+END_SRC
|
226
|
+
|
227
|
+
I tried to train wioth above conditions.
|
228
|
+
The accuracy of trained model was 92.3%.
|
229
|
+
The weights file is [[https://github.com/elgoog/weights_ruby_brain/blob/master/weights_782_50_10_1.yml][here]].
|
230
|
+
|
231
|
+
* Contributing
|
232
|
+
|
233
|
+
1. Fork it ( https://github.com/elgoog/ruby_brain/fork )
|
234
|
+
2. Create your feature branch (`git checkout -b my-new-feature`)
|
235
|
+
3. Commit your changes (`git commit -am 'Add some feature'`)
|
236
|
+
4. Push to the branch (`git push origin my-new-feature`)
|
237
|
+
5. Create a new Pull Request
|
data/Rakefile
ADDED
data/bin/console
ADDED
@@ -0,0 +1,14 @@
|
|
1
|
+
#!/usr/bin/env ruby
|
2
|
+
|
3
|
+
require "bundler/setup"
|
4
|
+
require "ruby_brain"
|
5
|
+
|
6
|
+
# You can add fixtures and/or initialization code here to make experimenting
|
7
|
+
# with your gem easier. You can also use a different console, if you like.
|
8
|
+
|
9
|
+
# (If you use this, don't forget to add pry to your Gemfile!)
|
10
|
+
# require "pry"
|
11
|
+
# Pry.start
|
12
|
+
|
13
|
+
require "irb"
|
14
|
+
IRB.start
|
data/bin/setup
ADDED
data/examples/mnist.rb
ADDED
@@ -0,0 +1,79 @@
|
|
1
|
+
require 'ruby_brain'
|
2
|
+
require 'ruby_brain/dataset/mnist/data'
|
3
|
+
|
4
|
+
# NUM_TEST_DATA = 50000
|
5
|
+
|
6
|
+
dataset = RubyBrain::DataSet::Mnist::data
|
7
|
+
|
8
|
+
NUM_TRAIN_DATA = 5000
|
9
|
+
|
10
|
+
training_input = dataset[:input][0..(NUM_TRAIN_DATA-1)]
|
11
|
+
training_supervisor = dataset[:output][0..(NUM_TRAIN_DATA-1)]
|
12
|
+
|
13
|
+
# test_input = dataset[:input][NUM_TRAIN_DATA..(NUM_TRAIN_DATA+NUM_TEST_DATA-1)]
|
14
|
+
# test_supervisor = dataset[:output][NUM_TRAIN_DATA..(NUM_TRAIN_DATA+NUM_TEST_DATA-1)]
|
15
|
+
test_input = dataset[:input][NUM_TRAIN_DATA..-1]
|
16
|
+
test_supervisor = dataset[:output][NUM_TRAIN_DATA..-1]
|
17
|
+
|
18
|
+
network = RubyBrain::Network.new([dataset[:input].first.size, 50, dataset[:output].first.size])
|
19
|
+
network.learning_rate = 0.7
|
20
|
+
network.init_network
|
21
|
+
### You can load weights from file in this timing if you want.
|
22
|
+
# network.load_weights_from_yaml_file(File.dirname(__FILE__) + '/../best_weights_1469044985.yml')
|
23
|
+
|
24
|
+
### You can initializes weights by loading weights from file if you want.
|
25
|
+
# network.load_weights_from_yaml_file('path/to/weights.yml.file')
|
26
|
+
|
27
|
+
network.learn(training_input, training_supervisor, max_training_count=100, tolerance=0.0004, monitoring_channels=[:best_params_training])
|
28
|
+
|
29
|
+
### You can save weights into a yml file if you want.
|
30
|
+
# network.dump_weights_to_yaml('path/to/weights.yml.file')
|
31
|
+
|
32
|
+
|
33
|
+
class Array
|
34
|
+
def argmax
|
35
|
+
max_i, max_val = 0, self.first
|
36
|
+
self.each_with_index do |v, i|
|
37
|
+
max_val, max_i = v, i if v > max_val
|
38
|
+
end
|
39
|
+
max_i
|
40
|
+
end
|
41
|
+
end
|
42
|
+
|
43
|
+
results = []
|
44
|
+
test_input.each_with_index do |input, i|
|
45
|
+
### You can see test input, label and predicated lable in standard out if you uncomment in this block
|
46
|
+
# input.each_with_index do |e, j|
|
47
|
+
# print(e > 0.3 ? 'x' : ' ')
|
48
|
+
# puts if (j % 28) == 0
|
49
|
+
# end
|
50
|
+
# puts
|
51
|
+
supervisor_label = test_supervisor[i].argmax
|
52
|
+
predicated_label = network.get_forward_outputs(test_input[i]).argmax
|
53
|
+
# puts "test_supervisor: #{supervisor_label}"
|
54
|
+
# puts "predicate: #{predicated_label}"
|
55
|
+
results << (supervisor_label == predicated_label)
|
56
|
+
# puts "------------------------------------------------------------"
|
57
|
+
end
|
58
|
+
|
59
|
+
puts "accuracy: #{results.count(true).to_f/results.size}"
|
60
|
+
|
61
|
+
|
62
|
+
|
63
|
+
### you can do above procedure simply by using Trainer
|
64
|
+
|
65
|
+
# training_option = {
|
66
|
+
# learning_rate: 0.5,
|
67
|
+
# max_training_count: 50,
|
68
|
+
# tolerance: 0.0004,
|
69
|
+
# # initial_weights_file: 'weights_3_30_10_1429166740.yml',
|
70
|
+
# # initial_weights_file: 'best_weights_1429544001.yml',
|
71
|
+
# monitoring_channels: [:best_params_training]
|
72
|
+
# }
|
73
|
+
|
74
|
+
# RubyBrain::Trainer.normal_learning([dataset[:input].first.size, 50, dataset[:output].first.size],
|
75
|
+
# training_input, training_supervisor,
|
76
|
+
# training_option)
|
77
|
+
|
78
|
+
|
79
|
+
|