ruby-dnn 0.8.5 → 0.8.6

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 3b897e7f51a4bf649e3bc9b0207dd3dbea30fc35d50ca4cf5ff08c7f8daacb4c
-  data.tar.gz: '078b1fa114f3abe04cc40e2e01211f3a563c2796bc9da78b904478ad0a5e83ac'
+  metadata.gz: 63b7cd17090ec6fd4e1f23e6b40f3c41e8a31ad0fa887945bda962a6df305c1b
+  data.tar.gz: b7e173f392c73e8151bde633d61913e8ce1645aaf9acc3a48788a950890d6ca4
 SHA512:
-  metadata.gz: fe348f3964752cdc520d2f571da843a26073795de1033a823757e388f506e800fd310ffe6063d3cb415d55502c5c9576162cb1404a925088dced844b143aa129
-  data.tar.gz: d3347bb4ed8c6b2ad9945252947aa44c44eda188ec64cb5aa4ac371d7f3a426241f9c76c67eb12de42854ceebf4aa796a2762b43b7395a09a75e234e403b6619
+  metadata.gz: c35a8a90216f42fcd8f88b4d4ee6f98fa517c5c2c8aaa6e61eeb8c260b176966c979e0e25311c6c4d0237e89f5e9c0f39eb056f1fad37f5fbaeeff7d0a4e1739
+  data.tar.gz: d2905e0b21c2dd2911cd985e490425be1909fed26e66e63b357aff7d657c2c86099653b1cdff8910171a7f1302029b2aa1f5f63f850b2a3f3d398e2325982475

data/API-Reference.ja.md

@@ -2,7 +2,7 @@
 ruby-dnnのAPIリファレンスです。このリファレンスでは、APIを利用するうえで必要となるクラスとメソッドしか記載していません。
 そのため、プログラムの詳細が必要な場合は、ソースコードを参照してください。
 
-最終更新バージョン:0.8.5
+最終更新バージョン:0.8.6
 
 # module DNN
 ruby-dnnの名前空間をなすモジュールです。
@@ -706,6 +706,16 @@ Float alpha
 パラメータを0で初期化します。
 
 
+# class Const < Initializer
+パラメータを指定の定数で初期化します。
+
+## 【Instance methods】
+## def initialize(const)
+### arguments
+* Float const  
+初期化する定数
+
+
 # class RandomNormal < Initializer
 パラメータを正規分布による乱数で初期化します。
 

data/LIB-API-Reference.ja.md

@@ -1,6 +1,6 @@
 # LIB-APIリファレンス
 ruby-dnnの付属ライブラリのリファレンスです。  
-最終更新バージョン:0.5.10
+最終更新バージョン:0.8.6
 
 
 # dnn/lib/mnist
@@ -71,11 +71,11 @@ Array
   Numo::UInt8の[10000]の形式
 
 
-# dnn/lib/image_io
-画像のin/outを行うライブラリです。内部でstb_image.hとstb_image_write.hを使用しています。
+# dnn/lib/image
+画像のread/writeを行うライブラリです。内部でstb_image.hとstb_image_write.hを使用しています。
 
 
-# module ImageIO
+# module Image
 
 ## def self.read(file_name)
 画像をNumo::UInt8形式で読み込みます。

data/lib/dnn/core/cnn_layers.rb

@@ -271,12 +271,12 @@ module DNN
 
       def backward(dout)
         unpool_h, unpool_w = @unpool_size
-        dout = dout.reshape(dout.shape[0], @x_shape[0], unpool_h, @x_shape[1], unpool_w, @num_channel)
+        dout = dout.reshape(dout.shape[0], @x_shape[1], unpool_h, @x_shape[2], unpool_w, @num_channel)
         dout[true, true, 0, true, 0, true].clone
       end
 
       def shape
-        [@out_width, @out_height, @num_channel]
+        [*@out_size, @num_channel]
       end
 
       def to_hash

data/lib/dnn/core/initializers.rb

@@ -20,6 +20,25 @@ module DNN
         param.data = param.data.fill(0)
       end
     end
+
+
+    class Const < Initializer
+      def self.load_hash(hash)
+        self.new(hash[:const])
+      end
+
+      def initialize(const)
+        @const = const
+      end
+
+      def init_param(layer, param)
+        param.data = param.data.fill(@const)
+      end
+
+      def to_hash
+        super({const: @const})
+      end
+    end
     
     
     class RandomNormal < Initializer

data/lib/dnn/core/rnn_layers.rb

@@ -121,25 +121,27 @@ module DNN
 
 
     class SimpleRNN_Dense
-      def initialize(rnn)
-        @rnn = rnn
-        @activation = rnn.activation.clone
+      def initialize(weight, weight2, bias, activation)
+        @weight = weight
+        @weight2 = weight2
+        @bias = bias
+        @activation = activation.clone
       end
 
       def forward(x, h)
         @x = x
         @h = h
-        h2 = x.dot(@rnn.weight.data) + h.dot(@rnn.weight2.data) + @rnn.bias.data
+        h2 = x.dot(@weight.data) + h.dot(@weight2.data) + @bias.data
         @activation.forward(h2)
       end
 
       def backward(dh2)
         dh2 = @activation.backward(dh2)
-        @rnn.weight.grad += @x.transpose.dot(dh2)
-        @rnn.weight2.grad += @h.transpose.dot(dh2)
-        @rnn.bias.grad += dh2.sum(0)
-        dx = dh2.dot(@rnn.weight.data.transpose)
-        dh = dh2.dot(@rnn.weight2.data.transpose)
+        @weight.grad += @x.transpose.dot(dh2)
+        @weight2.grad += @h.transpose.dot(dh2)
+        @bias.grad += dh2.sum(0)
+        dx = dh2.dot(@weight.data.transpose)
+        dh = dh2.dot(@weight2.data.transpose)
         [dx, dh]
       end
     end
@@ -194,15 +196,17 @@ module DNN
         @weight_initializer.init_param(self, @weight2)
         @bias_initializer.init_param(self, @bias)
         @time_length.times do |t|
-          @layers << SimpleRNN_Dense.new(self)
+          @layers << SimpleRNN_Dense.new(@weight, @weight2, @bias)
         end
       end
     end
 
 
     class LSTM_Dense
-      def initialize(rnn)
-        @rnn = rnn
+      def initialize(weight, weight2, bias)
+        @weight = weight
+        @weight2 = weight2
+        @bias = bias
         @tanh = Tanh.new
         @g_tanh = Tanh.new
         @forget_sigmoid = Sigmoid.new
@@ -215,7 +219,7 @@ module DNN
         @h = h
         @c = c
         num_nodes = h.shape[1]
-        a = x.dot(@rnn.weight.data) + h.dot(@rnn.weight2.data) + @rnn.bias.data
+        a = x.dot(@weight.data) + h.dot(@weight2.data) + @bias.data
 
         @forget = @forget_sigmoid.forward(a[true, 0...num_nodes])
         @g = @g_tanh.forward(a[true, num_nodes...(num_nodes * 2)])
@@ -239,11 +243,11 @@ module DNN
 
         da = Xumo::SFloat.hstack([dforget, dg, din, dout])
 
-        @rnn.weight.grad += @x.transpose.dot(da)
-        @rnn.weight2.grad += @h.transpose.dot(da)
-        @rnn.bias.grad += da.sum(0)
-        dx = da.dot(@rnn.weight.data.transpose)
-        dh = da.dot(@rnn.weight2.data.transpose)
+        @weight.grad += @x.transpose.dot(da)
+        @weight2.grad += @h.transpose.dot(da)
+        @bias.grad += da.sum(0)
+        dx = da.dot(@weight.data.transpose)
+        dh = da.dot(@weight2.data.transpose)
         dc = dc2_tmp * @forget
         [dx, dh, dc]
       end
@@ -331,15 +335,17 @@ module DNN
         @weight_initializer.init_param(self, @weight2)
         @bias_initializer.init_param(self, @bias)
         @time_length.times do |t|
-          @layers << LSTM_Dense.new(self)
+          @layers << LSTM_Dense.new(@weight, @weight2, @bias)
         end
       end
     end
 
 
     class GRU_Dense
-      def initialize(rnn)
-        @rnn = rnn
+      def initialize(weight, weight2, bias)
+        @weight = weight
+        @weight2 = weight2
+        @bias = bias
         @update_sigmoid = Sigmoid.new
         @reset_sigmoid = Sigmoid.new
         @tanh = Tanh.new
@@ -349,16 +355,16 @@ module DNN
         @x = x
         @h = h
         num_nodes = h.shape[1]
-        @weight_a = @rnn.weight.data[true, 0...(num_nodes * 2)]
-        @weight2_a = @rnn.weight2.data[true, 0...(num_nodes * 2)]
-        bias_a = @rnn.bias.data[0...(num_nodes * 2)]
+        @weight_a = @weight.data[true, 0...(num_nodes * 2)]
+        @weight2_a = @weight2.data[true, 0...(num_nodes * 2)]
+        bias_a = @bias.data[0...(num_nodes * 2)]
         a = x.dot(@weight_a) + h.dot(@weight2_a) + bias_a
         @update = @update_sigmoid.forward(a[true, 0...num_nodes])
         @reset = @reset_sigmoid.forward(a[true, num_nodes..-1])
 
-        @weight_h = @rnn.weight.data[true, (num_nodes * 2)..-1]
-        @weight2_h = @rnn.weight2.data[true, (num_nodes * 2)..-1]
-        bias_h = @rnn.bias.data[(num_nodes * 2)..-1]
+        @weight_h = @weight.data[true, (num_nodes * 2)..-1]
+        @weight2_h = @weight2.data[true, (num_nodes * 2)..-1]
+        bias_h = @bias.data[(num_nodes * 2)..-1]
         @tanh_h = @tanh.forward(x.dot(@weight_h) + (h * @reset).dot(@weight2_h) + bias_h)
         h2 = (1 - @update) * h + @update * @tanh_h
         h2
@@ -383,9 +389,9 @@ module DNN
         dh += da.dot(@weight2_a.transpose)
         dbias_a = da.sum(0)
 
-        @rnn.weight.grad += Xumo::SFloat.hstack([dweight_a, dweight_h])
-        @rnn.weight2.grad += Xumo::SFloat.hstack([dweight2_a, dweight2_h])
-        @rnn.bias.grad += Xumo::SFloat.hstack([dbias_a, dbias_h])
+        @weight.grad += Xumo::SFloat.hstack([dweight_a, dweight_h])
+        @weight2.grad += Xumo::SFloat.hstack([dweight2_a, dweight2_h])
+        @bias.grad += Xumo::SFloat.hstack([dbias_a, dbias_h])
         [dx, dh]
       end
     end
@@ -425,7 +431,7 @@ module DNN
         @weight_initializer.init_param(self, @weight2)
         @bias_initializer.init_param(self, @bias)
         @time_length.times do |t|
-          @layers << GRU_Dense.new(self)
+          @layers << GRU_Dense.new(@weight, @weight2, @bias)
         end
       end
     end

data/lib/dnn/lib/{image_io.rb → image.rb}

@@ -2,9 +2,9 @@ require "numo/narray"
 require_relative "../ext/rb_stb_image/rb_stb_image"
 
 module DNN
-  module ImageIO
+  module Image
     def self.read(file_name)
-      raise ImageIO::ReadError.new("#{file_name} is not found.") unless File.exist?(file_name)
+      raise Image::ReadError.new("#{file_name} is not found.") unless File.exist?(file_name)
       bin, w, h, n = Stb.stbi_load(file_name, 3)
       img = Numo::UInt8.from_binary(bin)
       img.reshape(h, w, 3)
@@ -29,13 +29,13 @@ module DNN
         Stb.stbi_write_jpg(file_name, w, h, ch, bin, quality)
       end
     rescue => ex
-      raise ImageIO::WriteError.new(ex.message)
+      raise Image::WriteError.new(ex.message)
     end
   end
 
-  class ImageIO::Error < StandardError; end
+  class Image::Error < StandardError; end
 
-  class ImageIO::ReadError < ImageIO::Error; end
+  class Image::ReadError < Image::Error; end
 
-  class ImageIO::WriteError < ImageIO::Error; end
+  class Image::WriteError < Image::Error; end
 end

data/lib/dnn/version.rb

@@ -1,3 +1,3 @@
 module DNN
-  VERSION = "0.8.5"
+  VERSION = "0.8.6"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: ruby-dnn
 version: !ruby/object:Gem::Version
-  version: 0.8.5
+  version: 0.8.6
 platform: ruby
 authors:
 - unagiootoro
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2019-02-22 00:00:00.000000000 Z
+date: 2019-02-27 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: numo-narray
@@ -123,7 +123,7 @@ files:
 - lib/dnn/ext/rb_stb_image/stb_image.h
 - lib/dnn/ext/rb_stb_image/stb_image_write.h
 - lib/dnn/lib/cifar10.rb
-- lib/dnn/lib/image_io.rb
+- lib/dnn/lib/image.rb
 - lib/dnn/lib/mnist.rb
 - lib/dnn/version.rb
 - ruby-dnn.gemspec