ruby-dnn 0.1.0

checksums.yaml

@@ -0,0 +1,7 @@
+---
+SHA256:
+  metadata.gz: f4d8dd255a75d99969d877a06faa150456210f131858a6f08b558dc070f5af9f
+  data.tar.gz: 72fe90225b3deb1a2957af33529ef4f1845749cd53ce0246d75ffd8ed36fded5
+SHA512:
+  metadata.gz: 96194040842f66d4ea3499fab0eefb7e5b47212ec8ba2981640879446ac72103c531bab05e55aab64ecc40d2e490fc8a5a96b20068167e017d58250f060230cf
+  data.tar.gz: 718a6f11e8a8e647dc0fa3867b72c3f289efa3d8a8efd689425d92ab5003e221f729a326011a46caa3c0d53be10cfe569a7e959197c0b818f44e948842a1b832

data/.gitignore

@@ -0,0 +1,8 @@
+/.bundle/
+/.yardoc
+/_yardoc/
+/coverage/
+/doc/
+/pkg/
+/spec/reports/
+/tmp/

data/.travis.yml

@@ -0,0 +1,5 @@
+sudo: false
+language: ruby
+rvm:
+  - 2.5.1
+before_install: gem install bundler -v 1.16.2

data/CODE_OF_CONDUCT.md

@@ -0,0 +1,74 @@
+# Contributor Covenant Code of Conduct
+
+## Our Pledge
+
+In the interest of fostering an open and welcoming environment, we as
+contributors and maintainers pledge to making participation in our project and
+our community a harassment-free experience for everyone, regardless of age, body
+size, disability, ethnicity, gender identity and expression, level of experience,
+nationality, personal appearance, race, religion, or sexual identity and
+orientation.
+
+## Our Standards
+
+Examples of behavior that contributes to creating a positive environment
+include:
+
+* Using welcoming and inclusive language
+* Being respectful of differing viewpoints and experiences
+* Gracefully accepting constructive criticism
+* Focusing on what is best for the community
+* Showing empathy towards other community members
+
+Examples of unacceptable behavior by participants include:
+
+* The use of sexualized language or imagery and unwelcome sexual attention or
+advances
+* Trolling, insulting/derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or electronic
+  address, without explicit permission
+* Other conduct which could reasonably be considered inappropriate in a
+  professional setting
+
+## Our Responsibilities
+
+Project maintainers are responsible for clarifying the standards of acceptable
+behavior and are expected to take appropriate and fair corrective action in
+response to any instances of unacceptable behavior.
+
+Project maintainers have the right and responsibility to remove, edit, or
+reject comments, commits, code, wiki edits, issues, and other contributions
+that are not aligned to this Code of Conduct, or to ban temporarily or
+permanently any contributor for other behaviors that they deem inappropriate,
+threatening, offensive, or harmful.
+
+## Scope
+
+This Code of Conduct applies both within project spaces and in public spaces
+when an individual is representing the project or its community. Examples of
+representing a project or community include using an official project e-mail
+address, posting via an official social media account, or acting as an appointed
+representative at an online or offline event. Representation of a project may be
+further defined and clarified by project maintainers.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be
+reported by contacting the project team at ootoro838861@outlook.jp. All
+complaints will be reviewed and investigated and will result in a response that
+is deemed necessary and appropriate to the circumstances. The project team is
+obligated to maintain confidentiality with regard to the reporter of an incident.
+Further details of specific enforcement policies may be posted separately.
+
+Project maintainers who do not follow or enforce the Code of Conduct in good
+faith may face temporary or permanent repercussions as determined by other
+members of the project's leadership.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage], version 1.4,
+available at [http://contributor-covenant.org/version/1/4][version]
+
+[homepage]: http://contributor-covenant.org
+[version]: http://contributor-covenant.org/version/1/4/

data/Gemfile

@@ -0,0 +1,6 @@
+source "https://rubygems.org"
+
+git_source(:github) {|repo_name| "https://github.com/#{repo_name}" }
+
+# Specify your gem's dependencies in ruby-dnn.gemspec
+gemspec

data/LICENSE.txt

@@ -0,0 +1,21 @@
+The MIT License (MIT)
+
+Copyright (c) 2018 unagiootoro
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+THE SOFTWARE.

data/README.md

@@ -0,0 +1,42 @@
+# ruby-dnn
+
+ruby-dnn is a ruby deep learning library. This library supports full connected neural network and convolution neural network.
+Currently, you can get 99% accuracy with MNIST and 60% with CIFAR 10.
+
+## Installation
+
+Add this line to your application's Gemfile:
+
+```ruby
+gem 'dnn'
+```
+
+And then execute:
+
+    $ bundle
+
+Or install it yourself as:
+
+    $ gem install dnn
+
+## Usage
+
+TODO: Write usage instructions here
+
+## Development
+
+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.
+
+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).
+
+## Contributing
+
+Bug reports and pull requests are welcome on GitHub at https://github.com/[USERNAME]/dnn. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the [Contributor Covenant](http://contributor-covenant.org) code of conduct.
+
+## License
+
+The gem is available as open source under the terms of the [MIT License](https://opensource.org/licenses/MIT).
+
+## Code of Conduct
+
+Everyone interacting in the Dnn project’s codebases, issue trackers, chat rooms and mailing lists is expected to follow the [code of conduct](https://github.com/[USERNAME]/dnn/blob/master/CODE_OF_CONDUCT.md).

data/Rakefile

@@ -0,0 +1,10 @@
+require "bundler/gem_tasks"
+require "rake/testtask"
+
+Rake::TestTask.new(:test) do |t|
+  t.libs << "test"
+  t.libs << "lib"
+  t.test_files = FileList["test/**/*_test.rb"]
+end
+
+task :default => :test

data/bin/console

@@ -0,0 +1,14 @@
+#!/usr/bin/env ruby
+
+require "bundler/setup"
+require "ruby/dnn"
+
+# You can add fixtures and/or initialization code here to make experimenting
+# with your gem easier. You can also use a different console, if you like.
+
+# (If you use this, don't forget to add pry to your Gemfile!)
+# require "pry"
+# Pry.start
+
+require "irb"
+IRB.start(__FILE__)

data/bin/setup

@@ -0,0 +1,8 @@
+#!/usr/bin/env bash
+set -euo pipefail
+IFS=$'\n\t'
+set -vx
+
+bundle install
+
+# Do any other automated setup that you need to do here

data/lib/dnn.rb

@@ -0,0 +1,14 @@
+require "numo/narray"
+
+Numo::SFloat.srand(rand(2**64))
+
+module DNN; end
+
+require "dnn/core/version"
+require "dnn/core/error"
+require "dnn/core/model"
+require "dnn/core/initializers"
+require "dnn/core/layers"
+require "dnn/core/activations"
+require "dnn/core/optimizers"
+require "dnn/core/util"

data/lib/dnn/core/activations.rb

@@ -0,0 +1,116 @@
+module DNN
+  module Activations
+    Layer = Layers::Layer
+    OutputLayer = Layers::OutputLayer
+
+
+    module SigmoidFunction
+      def forward(x)
+        @out = 1.0 / (1 + NMath.exp(-x))
+      end
+    end
+
+
+    class Sigmoid < Layer
+      include SigmoidFunction
+    
+      def backward(dout)
+        dout * (1.0 - @out) * @out
+      end
+    end
+
+
+    class Tanh < Layer
+      include Numo
+
+      def forward(x)
+        @x = x
+        NMath.tanh(x)
+      end
+      
+      def backward(dout)
+        dout * (1.0 / NMath.cosh(@x)**2)
+      end
+    end
+    
+    
+    class ReLU < Layer
+      def forward(x)
+        @x = x.clone
+        x[x < 0] = 0
+        x
+      end
+    
+      def backward(dout)
+        @x[@x > 0] = 1
+        @x[@x <= 0] = 0
+        dout * @x
+      end
+    end
+
+
+    class LeakyReLU < Layer
+      def initialize(alpha = 0.3)
+        @alpha = alpha
+      end
+
+      def forward(x)
+        @x = x.clone
+        a = Numo::SFloat.ones(x.shape)
+        a[x <= 0] = @alpha
+        x * a
+      end
+
+      def backward(dout)
+        @x[@x > 0] = 1
+        @x[@x <= 0] = @alpha
+        dout * @x
+      end
+    end
+
+
+    class IdentityWithLoss < OutputLayer
+      def forward(x)
+        @out = x
+      end
+    
+      def backward(y)
+        @out - y
+      end
+    
+      def loss(y)
+        0.5 * ((@out - y) ** 2).sum / @model.batch_size + ridge
+      end
+    end
+    
+    
+    class SoftmaxWithLoss < OutputLayer
+      def forward(x)
+        @out = NMath.exp(x) / NMath.exp(x).sum(1).reshape(x.shape[0], 1)
+      end
+    
+      def backward(y)
+        @out - y
+      end
+    
+      def loss(y)
+        -(y * NMath.log(@out + 1e-7)).sum / @model.batch_size + ridge
+      end
+    end
+
+
+    class SigmoidWithLoss < OutputLayer
+      include Numo
+      include SigmoidFunction
+
+      def backward(y)
+        @out - y
+      end
+
+      def loss(y)
+        -(y * NMath.log(@out + 1e-7) + (1 - y) * NMath.log(1 - @out + 1e-7)).sum / @model.batch_size + ridge
+      end
+    end
+
+  end
+end

data/lib/dnn/core/error.rb

@@ -0,0 +1,13 @@
+module DNN
+  class DNN_Error < StandardError; end
+
+  class DNN_TypeError < DNN_Error; end
+
+  class DNN_SharpError < DNN_Error; end
+
+  class DNN_GradUnfairError < DNN_Error
+    def initialize(grad, n_grad)
+      super("gradient is #{grad}, but numerical gradient is #{n_grad}")
+    end
+  end
+end

data/lib/dnn/core/initializers.rb

@@ -0,0 +1,46 @@
+module DNN
+  module Initializers
+
+    class Initializer
+      def init_param(layer, param_key, param)
+        layer.params[param_key] = param
+      end
+    end
+
+
+    class Zeros < Initializer
+      def init_param(layer, param_key)
+        super(layer, param_key, layer.params[param_key].fill(0))
+      end
+    end
+    
+    
+    class RandomNormal < Initializer
+      def initialize(mean = 0, std = 0.05)
+        @mean = mean
+        @std = std
+      end
+    
+      def init_param(layer, param_key)
+        super(layer, param_key, layer.params[param_key].rand_norm(@mean, @std))
+      end
+    end
+    
+    
+    class Xavier < Initializer
+      def init_param(layer, param_key)
+        prev_dim = layer.prev_layer.shape.reduce(:*)
+        super(layer, param_key, layer.params[param_key].rand_norm / Math.sqrt(prev_dim))
+      end
+    end
+    
+    
+    class He < Initializer
+      def init_param(layer, param_key)
+        prev_dim = layer.prev_layer.shape.reduce(:*)
+        super(layer, param_key, layer.params[param_key].rand_norm / Math.sqrt(prev_dim) * Math.sqrt(2))
+      end
+    end
+
+  end
+end

data/lib/dnn/core/layers.rb

@@ -0,0 +1,366 @@
+module DNN
+  module Layers
+
+    #Super class of all optimizer classes.
+    class Layer
+      include Numo
+
+      #Initialize layer when model is compiled.
+      def init(model)
+        @model = model
+      end
+
+      #Forward propagation.
+      def forward() end
+
+      #Backward propagation.
+      def backward() end
+    
+      #Get the shape of the layer.
+      def shape
+        prev_layer.shape
+      end
+    
+      #Get the previous layer.
+      def prev_layer
+        @model.layers[@model.layers.index(self) - 1]
+      end
+    end
+    
+    
+    class HasParamLayer < Layer
+      attr_reader :params #The parameters of the layer.
+      attr_reader :grads  #Differential value of parameter of layer.
+    
+      def initialize
+        @params = {}
+        @grads = {}
+      end
+    
+      def init(model)
+        super
+        init_params
+      end
+    
+      #Update the parameters.
+      def update
+        @model.optimizer.update(self)
+      end
+    
+      private
+      
+      #Initialize of the parameters.
+      def init_params() end
+    end
+    
+    
+    class InputLayer < Layer
+      attr_reader :shape
+    
+      def initialize(dim_or_shape)
+        @shape = dim_or_shape.is_a?(Array) ? dim_or_shape : [dim_or_shape]
+      end
+
+      def forward(x)
+        x
+      end
+    
+      def backward(dout)
+        dout
+      end
+    end
+    
+    
+    class Dense < HasParamLayer
+      include Initializers
+
+      attr_reader :num_nodes
+      attr_reader :weight_decay
+    
+      def initialize(num_nodes,
+                     weight_initializer: nil,
+                     bias_initializer: nil,
+                     weight_decay: 0)
+        super()
+        @num_nodes = num_nodes
+        @weight_initializer = (weight_initializer || RandomNormal.new)
+        @bias_initializer = (bias_initializer || Zeros.new)
+        @weight_decay = weight_decay
+      end
+    
+      def forward(x)
+        @x = x
+        @x.dot(@params[:weight]) + @params[:bias]
+      end
+    
+      def backward(dout)
+        @grads[:weight] = @x.transpose.dot(dout)
+        if @weight_decay > 0
+          dridge = @weight_decay * @params[:weight]
+          @grads[:weight] += dridge
+        end
+        @grads[:bias] = dout.sum(0)
+        dout.dot(@params[:weight].transpose)
+      end
+    
+      def shape
+        [@num_nodes]
+      end
+    
+      private
+    
+      def init_params
+        num_prev_nodes = prev_layer.shape[0]
+        @params[:weight] = SFloat.new(num_prev_nodes, @num_nodes)
+        @params[:bias] = SFloat.new(@num_nodes)
+        @weight_initializer.init_param(self, :weight)
+        @bias_initializer.init_param(self, :bias)
+      end
+    end
+    
+    
+    module Convert
+      def im2col(img, out_h, out_w, fh, fw, strides)
+        bs, fn = img.shape[0..1]
+        col = SFloat.zeros(bs, fn, fh, fw, out_h, out_w)
+        (0...fh).each do |i|
+          i_range = (i...(i + strides[0] * out_h)).step(strides[0]).to_a
+          (0...fw).each do |j|
+            j_range = (j...(j + strides[1] * out_w)).step(strides[1]).to_a
+            col[true, true, i, j, true, true] = img[true, true, i_range, j_range]
+          end
+        end
+        col.transpose(0, 4, 5, 1, 2, 3).reshape(bs * out_h * out_w, fn * fh * fw)
+      end
+      
+      def col2im(col, img_shape, out_h, out_w, fh, fw, strides)
+        bs, fn, ih, iw = img_shape
+        col = col.reshape(bs, out_h, out_w, fn, fh, fw).transpose(0, 3, 4, 5, 1, 2)
+        img = SFloat.zeros(bs, fn, ih, iw)
+        (0...fh).each do |i|
+          i_range = (i...(i + strides[0] * out_h)).step(strides[0]).to_a
+          (0...fw).each do |j|
+            j_range = (j...(j + strides[1] * out_w)).step(strides[1]).to_a
+            img[true, true, i_range, j_range] += col[true, true, i, j, true, true]
+          end
+        end
+        img
+      end
+
+      def padding(img, pad)
+        bs, c, ih, iw = img.shape
+        ih2 = ih + pad * 2
+        iw2 = iw + pad * 2
+        img2 = SFloat.zeros(bs, c, ih2, iw2)
+        img2[true, true, pad...(ih + pad), pad...(iw + pad)] = img
+        img2
+      end
+    end
+    
+    
+    class Conv2D < HasParamLayer
+      include Initializers
+      include Convert
+    
+      def initialize(num_filters, filter_height, filter_width,
+                     weight_initializer: nil,
+                     bias_initializer: nil,
+                     strides: [1, 1],
+                     padding: 0,
+                     weight_decay: 0)
+        super()
+        @num_filters = num_filters
+        @filter_height = filter_height
+        @filter_width = filter_width
+        @weight_initializer = (weight_initializer || RandomNormal.new)
+        @bias_initializer = (bias_initializer || Zeros.new)
+        @strides = strides
+        @weight_decay = weight_decay
+        @padding = padding
+      end
+    
+      def init(model)
+        super
+        prev_height, prev_width = prev_layer.shape[1], prev_layer.shape[2]
+        @out_height = (prev_height + @padding * 2 - @filter_height) / @strides[0] + 1
+        @out_width = (prev_width + @padding * 2 - @filter_width) / @strides[1] + 1
+      end
+    
+      def forward(x)
+        x = padding(x, 2) if @padding > 0
+        @x_shape = x.shape
+        @col = im2col(x, @out_height, @out_width, @filter_height, @filter_width, @strides)
+        out = @col.dot(@params[:weight])
+        out.reshape(@model.batch_size, @out_height, @out_width, out.shape[3]).transpose(0, 3, 1, 2)
+      end
+    
+      def backward(dout)
+        dout = dout.transpose(0, 2, 3, 1)
+        dout = dout.reshape(dout.shape[0..2].reduce(:*), dout.shape[3])
+        @grads[:weight] = @col.transpose.dot(dout)
+        if @weight_decay > 0
+          dridge = @weight_decay * @params[:weight]
+          @grads[:weight] += dridge
+        end
+        @grads[:bias] = dout.sum(0)
+        dcol = dout.dot(@params[:weight].transpose)
+        col2im(dcol, @x_shape, @out_height, @out_width, @filter_height, @filter_width, @strides)
+      end
+    
+      def shape
+        [@num_filters, @out_height, @out_width]
+      end
+    
+      private
+    
+      def init_params
+        num_prev_filter = prev_layer.shape[0]
+        @params[:weight] = SFloat.new(num_prev_filter * @filter_height * @filter_height, @num_filters)
+        @params[:bias] = SFloat.new(@num_filters)
+        @weight_initializer.init_param(self, :weight)
+        @bias_initializer.init_param(self, :bias)
+      end
+    end
+    
+    
+    class MaxPool2D < Layer
+      include Convert
+
+      def initialize(pool_height, pool_width, strides: nil, padding: 0)
+        @pool_height = pool_height
+        @pool_width = pool_width
+        @strides = strides ? strides : [@pool_height, @pool_width]
+        @padding = padding
+      end
+    
+      def init(model)
+        super
+        prev_height, prev_width = prev_layer.shape[1], prev_layer.shape[2]
+        @num_channel = prev_layer.shape[0]
+        @out_height = (prev_height - @pool_height) / @strides[0] + 1
+        @out_width = (prev_width - @pool_width) / @strides[1] + 1
+      end
+    
+      def forward(x)
+        @x_shape = x.shape
+        col = im2col(x, @out_height, @out_width, @pool_height, @pool_width, @strides)
+        col = col.reshape(x.shape[0] * @out_height * @out_width * x.shape[1], @pool_height * @pool_width)
+        @max_index = col.max_index(1)
+        col.max(1).reshape(x.shape[0], @out_height, @out_width, x.shape[1]).transpose(0, 3, 1, 2)
+      end
+    
+      def backward(dout)
+        dout = dout.transpose(0, 2, 3, 1)
+        pool_size = @pool_height * @pool_width
+        dmax = SFloat.zeros(dout.size * pool_size)
+        dmax[@max_index] = dout.flatten
+        dcol = dmax.reshape(dout.shape[0..2].reduce(:*), dout.shape[3] * pool_size)
+        col2im(dcol, @x_shape, @out_height, @out_width, @pool_height, @pool_width, @strides)
+      end
+    
+      def shape
+        [@num_channel, @out_height, @out_width]
+      end
+    end
+    
+    
+    class Flatten < Layer
+      def forward(x)
+        @shape = x.shape
+        x.reshape(x.shape[0], x.shape[1..-1].reduce(:*))
+      end
+    
+      def backward(dout)
+        dout.reshape(*@shape)
+      end
+    
+      def shape
+        [prev_layer.shape.reduce(:*)]
+      end
+    end
+
+
+    class Reshape < Layer
+      attr_reader :shape
+      
+      def initialize(shape)
+        @shape = shape
+        @x_shape = nil
+      end
+
+      def forward(x)
+        @x_shape = x.shape
+        x.reshape(*@shape)
+      end
+
+      def backward(dout)
+        dout.reshape(@x_shape)
+      end
+    end
+
+
+    class OutputLayer < Layer
+      private
+    
+      def ridge
+        @model.layers.select { |layer| layer.is_a?(Dense) }
+                     .reduce(0) { |sum, layer| layer.weight_decay * (layer.params[:weight]**2).sum }
+      end
+    end
+    
+    
+    class Dropout < Layer
+      def initialize(dropout_ratio)
+        @dropout_ratio = dropout_ratio
+        @mask = nil
+      end
+    
+      def forward(x)
+        if @model.training
+          @mask = SFloat.ones(*x.shape).rand < @dropout_ratio
+          x[@mask] = 0
+        else
+          x *= (1 - @dropout_ratio)
+        end
+        x
+      end
+    
+      def backward(dout)
+        dout[@mask] = 0 if @model.training
+        dout
+      end
+    end
+    
+    
+    class BatchNormalization < HasParamLayer
+      def forward(x)
+        @mean = x.mean(0)
+        @xc = x - @mean
+        @var = (@xc**2).mean(0)
+        @std = NMath.sqrt(@var + 1e-7)
+        @xn = @xc / @std
+        @params[:gamma] * @xn + @params[:beta]
+      end
+    
+      def backward(dout)
+        @grads[:beta] = dout.sum(0)
+        @grads[:gamma] = (@xn * dout).sum(0)
+        dxn = @params[:gamma] * dout
+        dxc = dxn / @std
+        dstd = -((dxn * @xc) / (@std**2)).sum(0)
+        dvar = 0.5 * dstd / @std
+        dxc += (2.0 / @model.batch_size) * @xc * dvar
+        dmean = dxc.sum(0)
+        dxc - dmean / @model.batch_size
+      end
+    
+      private
+    
+      def init_params
+        @params[:gamma] = 1
+        @params[:beta] = 0
+      end
+    end
+  end
+  
+end