t_learn 0.1.1.0 → 0.1.1.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: b6597d2b7b9978e0c84e5e87920fb4f3fb765887
-  data.tar.gz: 8c54a272723b8640a8d89e0e8937f84a1ab1a279
+  metadata.gz: bcf9731528ca98ed43cfe5f2a7686707def2e6a9
+  data.tar.gz: a90853e05203740a90c3c5ce6cf9fb13f0ab36bf
 SHA512:
-  metadata.gz: 12134be65f46bbe5d9dec8aaaee3b90c8e3c02f9602f30419a75e7a74cb1c1683ca73725cbcbc081e16918114f357b292e5ceadd5fdd4117de9dcf5ae105641a
-  data.tar.gz: 1198ca3aaa4551929242a30e0afa117676702d26ebccc6d7eda6bcadebbfda9a1fb3a9aa7e2150d257a0f9f9a06ae3c3294cd37b2573a2459206d2b68ebc68b4
+  metadata.gz: 7b25db6487317f85009cb795bec1060e71cdca91804ce74f013e8933d3dfc6f1128c65d02829cac5956b31747cc159907f92fecdb7a910ef18219c95759aa8a5
+  data.tar.gz: 86dfa12d25b29010c0e6ccebbb120e2f28062e70bdece7e43d830e79d503038fd808553218f0a8d621f11f800be22cd2e30b8e940c5c752b7ba28e008e60a403

data/.rake_tasks~

@@ -0,0 +1,7 @@
+build
+clean
+clobber
+install
+install:local
+release[remote]
+spec

data/README.md

@@ -24,25 +24,23 @@ Or install it yourself as:
 respect for keras.
 
 ``` ruby
-def test_fun(x)
-    return ( x*0.9 ).round(2)
-end
-
 
 model = TLearn::FeedForwardNeuralNetwork.new
+
 model.add_layer(2)
-model.add_layer(5)
+model.add_layer(3)
 model.add_layer(1)
 
-x_train = [[0.1, 1.0],[0.2, 1.0], [0.4, 1.0], [0.6, 1.0]]
+x_train = [[0.0, 0.0],[0.0, 1.0], [1.0, 0.0], [1.0, 1.0]]
+y_train = [[ 0.0 ], [ 1.0 ],[ 1.0 ],[ 0.0 ]]
+model.fit(x_train, y_train, 50000)
 
-y_train = [[ test_fun(x_train[0][0]) ], [ test_fun(x_train[1][0]) ],[ test_fun(x_train[2][0]) ],[ test_fun(x_train[3][0]) ]]
-model.fit(x_train, y_train, 500000)
+x_test = x_train
+y_test = y_train
 
-x_test = [[0.1, 1.0],[0.2, 1.0], [0.4, 1.0], [0.6, 1.0]]
-y_test = [[ test_fun(x_train[0][0]) ], [ test_fun(x_train[1][0]) ],[ test_fun(x_train[2][0]) ],[ test_fun(x_train[3][0]) ]]
+err_rate = model.evaluate(x_test, y_test)
 
-model.evaluate(x_test, y_test)
+puts "err rate: #{err_rate}%"
 
 ```
 
@@ -51,10 +49,11 @@ model.evaluate(x_test, y_test)
 
 ```
 
-x [0.1,  1.0],  y 0.09 ,  output 0.22505163646378912
-x [0.2,  1.0],  y 0.18 ,  output 0.2817288022885251
-x [0.4,  1.0],  y 0.36 ,  output 0.3699200581887254
-x [0.6,  1.0],  y 0.54 ,  output 0.42524180537036876
+x [0.0,  0.0],  y [0.0] ,  output [0.03286460161620565]
+x [0.0,  1.0],  y [1.0] ,  output [0.9733866321804969]
+x [1.0,  0.0],  y [1.0] ,  output [0.9731963536942299]
+x [1.0,  1.0],  y [0.0] ,  output [0.014481150692655216]
+err rate: 2.5190691608533524%
 
 ```
 

data/lib/t_learn/feedforward_neural_network.rb

@@ -7,19 +7,21 @@ module TLearn
   # ==
   #
   class FeedForwardNeuralNetwork
-    attr_accessor :layer_list, :layer_size, :link_list, :node_id, :learning_rate
-    def initialize(learning_rate=0.3)
+    attr_accessor :layer_list, :layer_size, :link_list, :node_id, :learning_rate, :err_list, :threshold
+    def initialize(learning_rate=0.1, threshold=0.0)
       @layer_size = 0    #layer iterator
       @layer_list = Array.new
       @link_list = Hash.new
       @node_id = 0
       @learning_rate = learning_rate
+      @err_list = Array.new
+      @threshold = threshold
     end
 
     def add_layer(node_num)
       node_list = Array.new()
       node_num.times do |num|
-        node = Node.new(0.2)
+        node = Node.new(0.0,"sig", @threshold)
         node.set_id(@node_id)
         node_list.push(node)
         @node_id += 1
@@ -40,7 +42,7 @@ module TLearn
     def connect_nodes
       @layer_list[@layer_size - 1].each do |from_node|
         @layer_list[@layer_size].each do |to_node|
-          @link_list["#{from_node.id}_#{to_node.id}"] = 0.0
+          @link_list["#{from_node.id}_#{to_node.id}"] = rand(-1.0...1.0)
         end
       end
     end
@@ -54,11 +56,15 @@ module TLearn
     def fit(x_train, y_train, epoch)
       # input teacher_datas
       epoch.times do 
+        epoch_err = 0.0 
         x_train.zip(y_train).each do |x, y|
+          x, y = x_train.zip(y_train).sample
+          # puts "x #{x}, y #{y}"
           propagation(x)
-          # back_propagation 
+          epoch_err += calc_ave_err(y)
           back_propagation(y)
         end
+        @err_list.push(epoch_err)
       end
     end
 
@@ -75,12 +81,22 @@ module TLearn
             @layer_list[layer_num].each do |from_node|
               sum_all_from_node += @link_list["#{from_node.id}_#{to_node.id}"] * from_node.w
             end
-            to_node.update_w(sum_all_from_node)
+            to_node.update_w(sum_all_from_node + 1.0)
           end
         end
       end
     end
 
+    def calc_ave_err(y)
+      sum_err = 0.0
+      @layer_list[@layer_size - 1].each_with_index do |node, i|
+        sum_err += calc_err(node.w,y[i]).abs
+      end
+      ave_err = (sum_err)/y.size
+      return ave_err 
+    end
+
+    
     #
     # === 
     # 
@@ -88,56 +104,62 @@ module TLearn
     #
     def back_propagation(y)
       delta = {}
-      ( @layer_size - 1).downto(1) do |layer_num|
+      ( @layer_size - 1).downto(0) do |layer_num|
         if ( @layer_size - 1) == layer_num   # if output layer
-          @layer_list[layer_num].each_with_index do |to_node, i|
-            @layer_list[layer_num - 1].each do |from_node|
-              delta["#{from_node.id}_#{to_node.id}"] = - calc_err(to_node.w,y[i]) * to_node.w * (1.0 - to_node.w)
-              # puts "delta[#{from_node}_#{to_node}]  #{delta['#{from_node}_#{to_node}']}"
-              delta_weight = -1.0 * @learning_rate * delta["#{from_node.id}_#{to_node.id}"] * to_node.w
-              @link_list["#{from_node.id}_#{to_node.id}"] = @link_list["#{from_node.id}_#{to_node.id}"] + delta_weight ;
-            end
+          @layer_list[layer_num].each_with_index do |output_node, i|
+              delta["#{output_node.id}"] = -1.0 * calc_err(y[i], output_node.w) * output_node.w * (1.0 -output_node.w)
           end
         else 
-          @layer_list[layer_num].each do |to_node|
-            @layer_list[layer_num - 1].each do |from_node|
-              delta["#{from_node.id}_#{to_node.id}"] = calc_delta(delta,layer_num, to_node) * to_node.w * (1.0 - to_node.w)
-              delta_weight = -1.0 * @learning_rate * delta["#{from_node.id}_#{to_node.id}"] * to_node.w
-              @link_list["#{from_node.id}_#{to_node.id}"] = @link_list["#{from_node.id}_#{to_node.id}"] + delta_weight 
+          @layer_list[layer_num].each do |from_node|
+            # リンクの更新
+            @layer_list[layer_num + 1].each do |to_node|
+              update_weight = -1.0 * @learning_rate * delta["#{to_node.id}"] * from_node.w
+              @link_list["#{from_node.id}_#{to_node.id}"] = @link_list["#{from_node.id}_#{to_node.id}"] + update_weight 
             end
+            # その層のdeltaの更新
+            delta["#{from_node.id}"] = calc_delta(delta,layer_num, from_node) * from_node.w * (1.0 - from_node.w)
           end
         end
       end
     end
 
-    def calc_err(w, teacher_data)
-      return (teacher_data - w )
+    def calc_err(teacher_data, w)
+      return (teacher_data -w)
     end
 
     def calc_delta(delta,layer_i, from_node)
       sum = 0.0
       @layer_list[layer_i+1].each do |to_node|
-        sum += delta["#{from_node.id}_#{to_node.id}"] * from_node.w 
+        sum += delta["#{to_node.id}"] * @link_list["#{from_node.id}_#{to_node.id}"]
       end
       return sum
     end
 
     def evaluate(x_test, y_test)
       # compare teacher_datas and output of nn
-      sum = 0.0
+      sum_err = 0.0
       x_test.zip(y_test).each do |x, y|
         propagation(x)
-        @layer_list[@layer_size -1].zip(y).each do |output, y_|
-          puts "x #{x}, y #{y_} , output #{output.w}"
-          sum += 1 if output.w == y_
+        output = []
+        err = 0.0
+        @layer_list[@layer_size -1].zip(y).each do |o, y_f|
+          output.push(o.w)
+          err += (y_f - o.w).abs
         end
+        sum_err += (err/y_test[0].size)
+        puts "x #{x}, y #{y} , output #{output}"
       end 
-      return (sum/y_test.size) * 100.0
+      return (sum_err/y_test.size) * 100.0
+      # return 0.0
+    end
+
+    def get_output_layer
+      return @layer_list[@layer_size-1]
     end
 
     class Node
       attr_accessor :w,:active_function, :threshold, :id
-      def initialize(w = 0.0, active_function = "sig", threshold = 0.5)
+      def initialize(w = 0.0, active_function = "sig", threshold = 0.0)
         @w = w 
         @threshold = threshold 
         @active_function = active_function
@@ -158,8 +180,7 @@ module TLearn
       end
 
       def sigmoid_fun(x, a=1)
-        result= (1.0/(1.0+Math.exp(-1.0 * a * x))) ;
-        return result
+        return (1.0/(1.0+Math.exp(-1.0 * a * x)))
       end
     end
   end

data/lib/t_learn/version.rb

@@ -1,3 +1,3 @@
 module TLearn
-  VERSION = "0.1.1.0"
+  VERSION = "0.1.1.2"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: t_learn
 version: !ruby/object:Gem::Version
-  version: 0.1.1.0
+  version: 0.1.1.2
 platform: ruby
 authors:
 - Tcom242242
 autorequire: 
 bindir: exe
 cert_chain: []
-date: 2016-12-27 00:00:00.000000000 Z
+date: 2017-01-01 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: bundler
@@ -60,6 +60,7 @@ extensions: []
 extra_rdoc_files: []
 files:
 - ".gitignore"
+- ".rake_tasks~"
 - ".rspec"
 - ".travis.yml"
 - CODE_OF_CONDUCT.md