ruby-dnn 0.6.2 → 0.6.3

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 530a8532898c2b8f934c257ff0d986407763fb3df432df8f6d061d31147b97d8
-  data.tar.gz: 333fb2f4e0cfd6c6aa78d3bea8f9159433acec613d8017218d24688ab1384d24
+  metadata.gz: 83bf68ae55435acfd33ccebf5798252ad670c7047f31de52c5b1951e0d0064f0
+  data.tar.gz: 63edb99dcafd2abb6b66f09a5fe44ada8e50f2674b56af2d3b3100980ba6e841
 SHA512:
-  metadata.gz: 9979622493c8056c75ea3c9a5561ffa6cb2a63147d7c230a43870125ae8a023d44ed9615b0fa48f169c29d19a3979d109570f46baf9122046be01080d10baf01
-  data.tar.gz: f9730237fcc631dab2109909508d73559dcbbcdad996578be7f1252776ab048cab0510b17e58a6662358dde6e8bd81356348c2576347b73aa8aaa0c0e4cd689b
+  metadata.gz: 78baa0705d6062f066d132f46c500277daf03957ddaa4e7530c9f75b8bfcc08d0d4ac6ffd3130d8cbf7e01b32ce6a1e60ab1b046af9ed187c2b658d6ee71bb8e
+  data.tar.gz: ae2930b1fcccd676f3e3273e0031c63888fe264d069c2e6c16aeb9c7def9a8bf5e05e2de2ee9265d2ce1e633c964e0b91843b3e2c0fa845428c77db09c9a05cd

data/README.md

@@ -21,7 +21,9 @@ Or install it yourself as:
 
 ## Usage
 
-TODO: Write usage instructions here
+Currently, only simple Japanese documents are available.
+Please refer to examples for basic usage.
+If you want to know more detailed information, please refer to the source code.
 
 ## Development
 

data/examples/mnist_lstm_example.rb

@@ -0,0 +1,36 @@
+require "dnn"
+require "dnn/lib/mnist"
+#require "numo/linalg/autoloader"
+
+include Numo
+include DNN::Layers
+include DNN::Activations
+include DNN::Optimizers
+Model = DNN::Model
+MNIST = DNN::MNIST
+
+x_train, y_train = MNIST.load_train
+x_test, y_test = MNIST.load_test
+
+x_train = Numo::SFloat.cast(x_train).reshape(x_train.shape[0], 28, 28)
+x_test = Numo::SFloat.cast(x_test).reshape(x_test.shape[0], 28, 28)
+
+x_train /= 255
+x_test /= 255
+
+y_train = DNN::Util.to_categorical(y_train, 10)
+y_test = DNN::Util.to_categorical(y_test, 10)
+
+model = Model.new
+
+model << InputLayer.new([28, 28])
+
+model << LSTM.new(200)
+model << LSTM.new(200, return_sequences: false)
+
+model << Dense.new(10)
+model << SoftmaxWithLoss.new
+
+model.compile(Adam.new)
+
+model.train(x_train, y_train, 10, batch_size: 100, test: [x_test, y_test])

data/lib/dnn.rb

@@ -1,6 +1,9 @@
-require "numo/narray"
-
-Xumo = Numo
+if defined? Cumo
+  Xumo = Cumo
+else
+  require "numo/narray"
+  Xumo = Numo
+end
 
 Xumo::SFloat.srand(rand(2**64))
 

data/lib/dnn/core/rnn_layers.rb

@@ -17,15 +17,47 @@ module DNN
                      weight_initializer: nil,
                      bias_initializer: nil,
                      weight_decay: 0)
-      super()
-      @num_nodes = num_nodes
-      @stateful = stateful
-      @return_sequences = return_sequences
-      @weight_initializer = (weight_initializer || RandomNormal.new)
-      @bias_initializer = (bias_initializer || Zeros.new)
-      @weight_decay = weight_decay
-      @layers = []
-      @h = nil
+        super()
+        @num_nodes = num_nodes
+        @stateful = stateful
+        @return_sequences = return_sequences
+        @weight_initializer = (weight_initializer || RandomNormal.new)
+        @bias_initializer = (bias_initializer || Zeros.new)
+        @weight_decay = weight_decay
+        @layers = []
+        @h = nil
+      end
+
+      def forward(xs)
+        @xs_shape = xs.shape
+        hs = Xumo::SFloat.zeros(xs.shape[0], @time_length, @num_nodes)
+        h = (@stateful && @h) ? @h : Xumo::SFloat.zeros(xs.shape[0], @num_nodes)
+        xs.shape[1].times do |t|
+          x = xs[true, t, false]
+          h = @layers[t].forward(x, h)
+          hs[true, t, false] = h
+        end
+        @h = h
+        @return_sequences ? hs : h
+      end
+
+      def backward(dh2s)
+        @grads[:weight] = Xumo::SFloat.zeros(*@params[:weight].shape)
+        @grads[:weight2] = Xumo::SFloat.zeros(*@params[:weight2].shape)
+        @grads[:bias] = Xumo::SFloat.zeros(*@params[:bias].shape)
+        unless @return_sequences
+          dh = dh2s
+          dh2s = Xumo::SFloat.zeros(dh.shape[0], @time_length, dh.shape[1])
+          dh2s[true, -1, false] = dh
+        end
+        dxs = Xumo::SFloat.zeros(@xs_shape)
+        dh = 0
+        (0...dh2s.shape[1]).to_a.reverse.each do |t|
+          dh2 = dh2s[true, t, false]
+          dx, dh = @layers[t].backward(dh2 + dh)
+          dxs[true, t, false] = dx
+        end
+        dxs
       end
 
       def to_hash(merge_hash = nil)
@@ -34,7 +66,6 @@ module DNN
           num_nodes: @num_nodes,
           stateful: @stateful,
           return_sequences: @return_sequences,
-          activation: @activation.to_hash,
           weight_initializer: @weight_initializer.to_hash,
           bias_initializer: @bias_initializer.to_hash,
           weight_decay: @weight_decay,
@@ -114,38 +145,6 @@ module DNN
         @activation = (activation || Tanh.new)
       end
 
-      def forward(xs)
-        @xs_shape = xs.shape
-        hs = Xumo::SFloat.zeros(xs.shape[0], @time_length, @num_nodes)
-        h = (@stateful && @h) ? @h : Xumo::SFloat.zeros(xs.shape[0], @num_nodes)
-        xs.shape[1].times do |t|
-          x = xs[true, t, false]
-          h = @layers[t].forward(x, h)
-          hs[true, t, false] = h
-        end
-        @h = h
-        @return_sequences ? hs : h
-      end
-
-      def backward(dh2s)
-        @grads[:weight] = Xumo::SFloat.zeros(*@params[:weight].shape)
-        @grads[:weight2] = Xumo::SFloat.zeros(*@params[:weight2].shape)
-        @grads[:bias] = Xumo::SFloat.zeros(*@params[:bias].shape)
-        unless @return_sequences
-          dh = dh2s
-          dh2s = Xumo::SFloat.zeros(dh.shape[0], @time_length, dh.shape[1])
-          dh2s[true, -1, false] = dh
-        end
-        dxs = Xumo::SFloat.zeros(@xs_shape)
-        dh = 0
-        (0...dh2s.shape[1]).to_a.reverse.each do |t|
-          dh2 = dh2s[true, t, false]
-          dx, dh = @layers[t].backward(dh2 + dh)
-          dxs[true, t, false] = dx
-        end
-        dxs
-      end
-
       def to_hash
         super({activation: @activation.to_hash})
       end
@@ -374,38 +373,6 @@ module DNN
         super
       end
 
-      def forward(xs)
-        @xs_shape = xs.shape
-        hs = Xumo::SFloat.zeros(xs.shape[0], @time_length, @num_nodes)
-        h = (@stateful && @h) ? @h : Xumo::SFloat.zeros(xs.shape[0], @num_nodes)
-        xs.shape[1].times do |t|
-          x = xs[true, t, false]
-          h = @layers[t].forward(x, h)
-          hs[true, t, false] = h
-        end
-        @h = h
-        @return_sequences ? hs : h
-      end
-
-      def backward(dh2s)
-        @grads[:weight] = Xumo::SFloat.zeros(*@params[:weight].shape)
-        @grads[:weight2] = Xumo::SFloat.zeros(*@params[:weight2].shape)
-        @grads[:bias] = Xumo::SFloat.zeros(*@params[:bias].shape)
-        unless @return_sequences
-          dh = dh2s
-          dh2s = Xumo::SFloat.zeros(dh.shape[0], @time_length, dh.shape[1])
-          dh2s[true, -1, false] = dh
-        end
-        dxs = Xumo::SFloat.zeros(@xs_shape)
-        dh = 0
-        (0...dh2s.shape[1]).to_a.reverse.each do |t|
-          dh2 = dh2s[true, t, false]
-          dx, dh = @layers[t].backward(dh2 + dh)
-          dxs[true, t, false] = dx
-        end
-        dxs
-      end
-
       private
     
       def init_params

data/lib/dnn/version.rb

@@ -1,3 +1,3 @@
 module DNN
-  VERSION = "0.6.2"
+  VERSION = "0.6.3"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: ruby-dnn
 version: !ruby/object:Gem::Version
-  version: 0.6.2
+  version: 0.6.3
 platform: ruby
 authors:
 - unagiootoro
@@ -101,8 +101,9 @@ files:
 - bin/console
 - bin/setup
 - examples/cifar10_example.rb
+- examples/mnist_conv2d_example.rb
 - examples/mnist_example.rb
-- examples/mnist_example2.rb
+- examples/mnist_lstm_example.rb
 - examples/xor_example.rb
 - lib/dnn.rb
 - lib/dnn/core/activations.rb