merged changes with master

spec/network_spec.cr

@@ -17,14 +17,17 @@ describe SHAInet::Network do
 
     xor = SHAInet::Network.new
     xor.add_layer(:input, 2, :memory)
-    1.times { |x| xor.add_layer(:hidden, 2, :memory) }
+    1.times { |x| xor.add_layer(:hidden, 3, :memory) }
     xor.add_layer(:output, 1, :memory)
     xor.fully_connect
 
-    # data, cost_function, activation_function, epochs, error_threshold, learning_rate, momentum)
-    xor.train(training_data, :mse, :sigmoid, 10000, 0.000001)
+    # data, training_type, cost_function, activation_function, epochs, error_threshold (sum of errors), learning_rate, momentum)
+    xor.train(training_data, :sgdm, :mse, :sigmoid, 10000, 0.001)
 
     (xor.run([0, 0]).first < 0.1).should eq(true)
+    (xor.run([1, 0]).first > 0.9).should eq(true)
+    (xor.run([0, 1]).first > 0.9).should eq(true)
+    (xor.run([1, 1]).first < 0.1).should eq(true)
   end
 
   it "works on iris dataset" do
@@ -36,7 +39,6 @@ describe SHAInet::Network do
     iris = SHAInet::Network.new
     iris.add_layer(:input, 4, :memory)
     iris.add_layer(:hidden, 5, :memory)
-    iris.add_layer(:hidden, 5, :memory)
     iris.add_layer(:output, 3, :memory)
     iris.fully_connect
 
@@ -53,7 +55,36 @@ describe SHAInet::Network do
     normalized = SHAInet::TrainingData.new(inputs, outputs)
     normalized.normalize_min_max
     puts normalized
-    iris.train(normalized.data, :mse, :sigmoid, 10000, 0.000001)
+    iris.train(normalized.data, :sgdm, :mse, :sigmoid, 20000, 0.1)
+    iris.run(normalized.normalized_inputs.first)
+    puts "Expected output is: [0,0,1]"
+  end
+
+  it "works on iris dataset with batch train with Rprop" do
+    label = {
+      "setosa"     => [0.to_f64, 0.to_f64, 1.to_f64],
+      "versicolor" => [0.to_f64, 1.to_f64, 0.to_f64],
+      "virginica"  => [1.to_f64, 0.to_f64, 0.to_f64],
+    }
+    iris = SHAInet::Network.new
+    iris.add_layer(:input, 4, :memory)
+    iris.add_layer(:hidden, 5, :memory)
+    iris.add_layer(:output, 3, :memory)
+    iris.fully_connect
+
+    outputs = Array(Array(Float64)).new
+    inputs = Array(Array(Float64)).new
+    CSV.each_row(File.read(__DIR__ + "/test_data/iris.csv")) do |row|
+      row_arr = Array(Float64).new
+      row[0..-2].each do |num|
+        row_arr << num.to_f64
+      end
+      inputs << row_arr
+      outputs << label[row[-1]]
+    end
+    normalized = SHAInet::TrainingData.new(inputs, outputs)
+    normalized.normalize_min_max
+    iris.train_batch(normalized.data, :rprop, :mse, :sigmoid, 20000, 0.1)
     result = iris.run(normalized.normalized_inputs.first)
     ((result.first < 0.1) && (result[1] < 0.1) && (result.last > 0.9)).should eq(true)
   end

src/shainet/functions.cr

@@ -142,4 +142,17 @@ module SHAInet
 
     return input_size, vocabulary_v, payloads_v
   end
+
+  # # Other # #
+
+  # Used in Rprop
+  def self.sign(input : GenNum)
+    if input > 0
+      return +1
+    elsif input < 0
+      return -1
+    else
+      return 0
+    end
+  end
 end