brainy 1.0.1 → 1.0.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: fd89c615bcbb935df0b74079d0a9d966e3ee7a1b
-  data.tar.gz: cfb4ba364fbb4e22988ea52ad1a009a7a409921b
+  metadata.gz: 9fd74abee11cdbbf0e0b79f302134fe7722ed67f
+  data.tar.gz: f867df45585f08ae396bee5799b74f611bcbb225
 SHA512:
-  metadata.gz: 44ddc7ca29fa248784df73e8b843427a7b31876f6372013e634cfa6383bfadf0aace3008b77340d9d50854cae61e602444ad5ffbed2ad8b87a0c5d128226015e
-  data.tar.gz: fb22aab27dc09690df51a40d121c6eb6b4285fa3cb3f1f4cc32f634e504786828c4bf57eaeaa913f2275b360c837162e97f7a26350539d4016645d5fb2c4cfb9
+  metadata.gz: 3cc3538aa573f9097ce3cebfe5a6d439ed580077b58ea93f5d1771d409f1d02fc9efc148110a58082665faa4f246a5ba781df6cca153d2b1f303d4b5752038f6
+  data.tar.gz: e44212b55f265290610ad81d03f79c12ee0cc0e47a34e7745491dfa49bea272d525fa600a0d8cfd7fb76076f1fdf1fbaecb30061421ef1bb258e9118ace20207

data/README.md

@@ -25,5 +25,5 @@ mse = 1000.times.map do
   (o - net.evaluate([i]).first) ** 2
 end.reduce(:+) / 1000
 
-puts "your MSE: #{ mse.round(3) }" # smaller is better
+puts "your MSE: #{ mse.round(6) }" # smaller is better
 ```

data/lib/brainy/network.rb

@@ -4,26 +4,32 @@ module Brainy
 
     def initialize(input_count, hidden_count, output_count, learning_rate)
       @layers = [
-          Array.new(hidden_count, Vector.elements(Array.new(input_count + 1, rand(-1.0..1.0)))),
-          Array.new(output_count, Vector.elements(Array.new(hidden_count + 1, rand(-1.0..1.0))))
+          Array.new(hidden_count) { Vector.elements(Array.new(input_count + 1) { rand(-1.0..1.0) }) },
+          Array.new(output_count) { Vector.elements(Array.new(hidden_count + 1) { rand(-1.0..1.0) }) }
       ]
       @learning_rate = learning_rate
-      @activate = lambda { |x| 1 / (1 + Math.exp(-1 * x)) }
-      @activate_prime = lambda { |x| x * (1 - x) }
+    end
+
+    def activate(x)
+      1 / (1 + Math.exp(-1 * x))
+    end
+
+    def activate_prime(x)
+      x * (1 - x)
     end
 
     def evaluate(inputs)
       @layers.reduce(Vector.elements(inputs)) do |input, layer|
-        input = Vector.elements(input.to_a + [1])
-        output = layer.map { |node| @activate.call(node.inner_product(input)) }
+        input = Vector.elements(input.to_a + [1.0])
+        output = layer.map { |node| activate(node.inner_product(input)) }
         Vector.elements(output)
       end
     end
 
     def train!(inputs, expected)
-      inputs = Vector.elements(inputs + [1])
-      hidden_outs = Vector.elements(@layers.first.map { |node| @activate.call(node.inner_product(inputs)) } + [1])
-      output_outs = Vector.elements(@layers.last.map { |node| @activate.call(node.inner_product(hidden_outs)) })
+      inputs = Vector.elements(inputs + [1.0])
+      hidden_outs = Vector.elements(@layers.first.map { |node| activate(node.inner_product(inputs)) } + [1.0])
+      output_outs = Vector.elements(@layers.last.map { |node| activate(node.inner_product(hidden_outs)) })
       output_deltas = get_output_deltas(expected, output_outs)
       hidden_deltas = get_hidden_deltas(hidden_outs, @layers.last, output_deltas)
       @layers[1] = get_updated_weights(@layers.last, hidden_outs, output_deltas)
@@ -32,14 +38,14 @@ module Brainy
 
     def get_output_deltas(expected, output)
       expected.zip(output.to_a).map do |expect, out|
-        (out - expect) * @activate_prime.call(out)
+        (out - expect) * activate_prime(out)
       end
     end
 
     def get_hidden_deltas(hidden_outs, output_nodes, output_deltas)
       hidden_outs.each_with_index.map do |out, index|
         error = output_nodes.zip(output_deltas).map { |weights, delta| weights[index] * delta }.reduce(:+)
-        error * @activate_prime.call(out)
+        error * activate_prime(out)
       end
     end
 

data/lib/brainy/version.rb

@@ -1,3 +1,3 @@
 module Brainy
-  VERSION = '1.0.1'
+  VERSION = '1.0.2'
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: brainy
 version: !ruby/object:Gem::Version
-  version: 1.0.1
+  version: 1.0.2
 platform: ruby
 authors:
 - Robert Scott Reis