ruby-dnn 0.9.0 → 0.9.1

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: c11d1f0246f81a299edfaa8be69ef8ba6ccdb940a2b185e3b0ab77b56e9ec8ad
-  data.tar.gz: 47441a8cef420dcb120c955cf9e30ef32848763e3301ddbe75abf08a657f5777
+  metadata.gz: '02929b358bbd4ff8c3107c54be0ce37ae86c20a3d345c0090f4be5bcd2ad8b32'
+  data.tar.gz: 7641e5072f9bcdd4eb1bd93d173f78fe0fa11769ecca614ce44155dc7e310b96
 SHA512:
-  metadata.gz: e643a3daaafebedb496d21084db402a5701787cbced23735b224b040556a7835e889d4458018b47351efce991606906bd89f3c878d8ab88fd9a99a98f89441e5
-  data.tar.gz: 9387a6e092ec219e11deb884164fac8364d2a25794791a7a086760dda920910c9157c9207a4c76e4a1cfd74eb3564897b9380654635b211651eee8611499be64
+  metadata.gz: df47d323eda15b0f11dcf2153083bf05a8bd6c158227c3c93c0c9e0ab1f4679769fa8ad031a8bcbb886e5610880b2c95e26ad425a49ccdfe1b79b0ee280628de
+  data.tar.gz: 86766ea8873229cb665d3e93e70142c246ad8b3e197543093d06ce674d66319a2f9a94a78cf089a0677df4a4966fdbfcfb0ffc95337e6465eabd2ab4d9ba7a1c

data/API-Reference.ja.md

@@ -939,25 +939,25 @@ Numo::SFloat y
 ### return
 損失関数の値。
 
-## abstruct def loss
-損失関数の値を取得します。
+## abstruct def backward(y)
+損失関数の逆伝搬を行います。全ての損失関数のクラスは、このメソッドを実装する必要があります。
 ### arguments
 Numo::SFloat y  
 出力データ。
 ### return
 損失関数の値。
 
-# class MeanSquaredError < OutputLayer
+# class MeanSquaredError < Loss
 二乗誤差の損失関数です。
 
 
-# class IdentityMAE < OutputLayer
+# class MeanAbsoluteError < Loss
 平均絶対誤差の損失関数です。
 
 
-# class SoftmaxCrossEntropy < OutputLayer
+# class SoftmaxCrossEntropy < Loss
 ソフトマックス関数とクロスエントロピー誤差を合わせた損失関数です。
 
 
-# class SigmoidCrossEntropy < OutputLayer
+# class SigmoidCrossEntropy < Loss
 シグモイド関数とクロスエントロピー誤差を合わせた損失関数です。

data/lib/dnn.rb

@@ -5,7 +5,9 @@ else
   Xumo = Numo
 end
 
-module DNN; end
+module DNN
+  NMath = Xumo::NMath
+end
 
 require_relative "dnn/version"
 require_relative "dnn/core/error"

data/lib/dnn/core/activations.rb

@@ -13,8 +13,6 @@ module DNN
 
 
     class Tanh < Layers::Layer
-      NMath = Xumo::NMath
-
       def forward(x)
         @out = NMath.tanh(x)
       end
@@ -38,8 +36,6 @@ module DNN
 
 
     class Softplus < Layers::Layer
-      NMath = Xumo::NMath
-
       def forward(x)
         @x = x
         NMath.log(1 + NMath.exp(x))
@@ -52,8 +48,6 @@ module DNN
 
 
     class Swish < Layers::Layer
-      NMath = Xumo::NMath
-
       def forward(x)
         @x = x
         @out = x * (1 / (1 + NMath.exp(-x)))
@@ -111,8 +105,6 @@ module DNN
 
 
     class ELU < Layers::Layer
-      NMath = Xumo::NMath
-
       attr_reader :alpha
 
       def self.load_hash(hash)

data/lib/dnn/core/layers.rb

@@ -318,14 +318,14 @@ module DNN
           mean = x.mean(0)
           @xc = x - mean
           var = (@xc**2).mean(0)
-          @std = Xumo::NMath.sqrt(var + 1e-7)
+          @std = NMath.sqrt(var + 1e-7)
           xn = @xc / @std
           @xn = xn
           @running_mean.data = @momentum * @running_mean.data + (1 - @momentum) * mean
           @running_var.data = @momentum * @running_var.data + (1 - @momentum) * var
         else
           xc = x - @running_mean.data
-          xn = xc / Xumo::NMath.sqrt(@running_var.data + 1e-7)
+          xn = xc / NMath.sqrt(@running_var.data + 1e-7)
         end
         @gamma.data * xn + @beta.data
       end

data/lib/dnn/core/losses.rb

@@ -57,11 +57,12 @@ module DNN
 
 
     class HuberLoss < Loss
-      def forward(out, y)
+      def forward(out, y, layers)
         @out = out
         loss = loss_l1(y)
         loss = loss > 1 ? loss : loss_l2(y)
-        @loss = loss + regularize
+        #@loss = loss + regularize(layers)
+        @loss = loss
       end
 
       def backward(y)
@@ -88,8 +89,6 @@ module DNN
 
 
     class SoftmaxCrossEntropy < Loss
-      NMath = Xumo::NMath
-
       def forward(x, y)
         @out = Utils.softmax(x)
         batch_size = y.shape[0]
@@ -103,8 +102,6 @@ module DNN
 
 
     class SigmoidCrossEntropy < Loss
-      NMath = Xumo::NMath
-
       def forward(x, y)
         @out = Utils.sigmoid(x)
         batch_size = y.shape[0]

data/lib/dnn/core/model.rb

@@ -43,11 +43,7 @@ module DNN
         hash_params.each do |key, (shape, base64_param)|
           bin = Base64.decode64(base64_param)
           data = Xumo::SFloat.from_binary(bin).reshape(*shape)
-          if layer.params[key].is_a?(Param)
-            layer.params[key].data = data
-          else
-            layer.params[key] = data
-          end
+          layer.params[key].data = data
         end
         has_param_layers_index += 1
       end
@@ -154,9 +150,7 @@ module DNN
               verbose: true,
               batch_proc: nil,
               &epoch_proc)
-      unless compiled?
-        raise DNN_Error.new("The model is not compiled.")
-      end
+      raise DNN_Error.new("The model is not compiled.") unless compiled?
       check_xy_type(x, y)
       dataset = Dataset.new(x, y)
       num_train_datas = x.shape[0]
@@ -194,11 +188,16 @@ module DNN
     end
   
     def train_on_batch(x, y, &batch_proc)
+      raise DNN_Error.new("The model is not compiled.") unless compiled?
       check_xy_type(x, y)
       input_data_shape_check(x, y)
       x, y = batch_proc.call(x, y) if batch_proc
       out = forward(x, true)
-      loss_value = @loss.forward(out, y) + @loss.regularize(get_all_layers)
+      loss_value = if @loss.is_a?(HuberLoss)
+        @loss.forward(out, y, get_all_layers)
+      else
+        @loss.forward(out, y) + @loss.regularize(get_all_layers)
+      end
       dout = @loss.backward(y)
       backward(dout, true)
       @loss.d_regularize(get_all_layers)

data/lib/dnn/core/optimizers.rb

@@ -88,7 +88,7 @@ module DNN
         params.select { |key, param| param.grad }.each_value do |param|
           @g[param] ||= 0
           @g[param] += param.grad**2
-          param.data -= (@learning_rate / Xumo::NMath.sqrt(@g[param] + 1e-7)) * param.grad
+          param.data -= (@learning_rate / NMath.sqrt(@g[param] + 1e-7)) * param.grad
         end
       end
     end
@@ -111,7 +111,7 @@ module DNN
         params.select { |key, param| param.grad }.each_value do |param|
           @g[param] ||= 0
           @g[param] = @alpha * @g[param] + (1 - @alpha) * param.grad**2
-          param.data -= (@learning_rate / Xumo::NMath.sqrt(@g[param] + 1e-7)) * param.grad
+          param.data -= (@learning_rate / NMath.sqrt(@g[param] + 1e-7)) * param.grad
         end
       end
 
@@ -140,7 +140,7 @@ module DNN
           @h[param] ||= Xumo::SFloat.zeros(*param.data.shape)
           @s[param] ||= Xumo::SFloat.zeros(*param.data.shape)
           @h[param] = @rho * @h[param] + (1 - @rho) * param.grad**2
-          v = (Xumo::NMath.sqrt(@s[param] + 1e-6) / Xumo::NMath.sqrt(@h[param] + 1e-6)) * param.grad
+          v = (NMath.sqrt(@s[param] + 1e-6) / NMath.sqrt(@h[param] + 1e-6)) * param.grad
           @s[param] = @rho * @s[param] + (1 - @rho) * v**2
           param.data -= v
         end
@@ -177,7 +177,7 @@ module DNN
           @v[param] ||= 0
           @m[param] += (1 - @beta1) * (param.grad - @m[param])
           @v[param] += (1 - @beta2) * (param.grad**2 - @v[param])
-          param.data -= lr * @m[param] / Xumo::NMath.sqrt(@v[param] + 1e-7)
+          param.data -= lr * @m[param] / NMath.sqrt(@v[param] + 1e-7)
         end
       end
 

data/lib/dnn/core/utils.rb

@@ -1,8 +1,6 @@
 module DNN
   # This module provides utility functions.
   module Utils
-    NMath = Xumo::NMath
-
     # Categorize labels into "num_classes" classes.
     def self.to_categorical(y, num_classes, narray_type = nil)
       narray_type ||= y.class

data/lib/dnn/lib/image.rb

@@ -14,7 +14,7 @@ module DNN
       if img.shape.length == 2
         img = Numo::UInt8[img, img, img].transpose(1, 2, 0).clone
       elsif img.shape[2] == 1
-        img = img.shape(img.shape[0], img.shape[1])
+        img = img.reshape(img.shape[0], img.shape[1])
         img = Numo::UInt8[img, img, img].transpose(1, 2, 0).clone
       end
       h, w, ch = img.shape

data/lib/dnn/version.rb

@@ -1,3 +1,3 @@
 module DNN
-  VERSION = "0.9.0"
+  VERSION = "0.9.1"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: ruby-dnn
 version: !ruby/object:Gem::Version
-  version: 0.9.0
+  version: 0.9.1
 platform: ruby
 authors:
 - unagiootoro
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2019-05-03 00:00:00.000000000 Z
+date: 2019-05-04 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: numo-narray