WIP perceptron train method

storm-ml/project.clj

@@ -6,4 +6,5 @@
   :javac-options {:debug "true" :fork "true"}
   :aot :all
   :dev-dependencies [[org.clojure/clojure "1.4.0"]
+                     [org.javatuples/javatuples "1.2"]
                      [storm "0.7.2"]])

storm-ml/src/jvm/storm/ml/Main.java

@@ -0,0 +1,40 @@
+package storm.ml;
+
+import java.lang.Boolean;
+import java.math.BigDecimal;
+import java.util.List;
+import java.util.ArrayList;
+
+import org.javatuples.Pair;
+
+import storm.ml.PerceptronTopologyBuilder;
+
+public class Main {
+  public static void main(String[] keywords) {
+      List<Pair<List<BigDecimal>, Boolean>> training_set = new ArrayList<Pair<List<BigDecimal>, Boolean>>(4);
+      List<BigDecimal> input_vector = new ArrayList<BigDecimal>(3);
+
+      input_vector.add(new BigDecimal(1));
+      input_vector.add(new BigDecimal(0));
+      input_vector.add(new BigDecimal(0));
+      training_set.add(new Pair<List<BigDecimal>, Boolean>(input_vector, true));
+
+      input_vector.add(new BigDecimal(1));
+      input_vector.add(new BigDecimal(0));
+      input_vector.add(new BigDecimal(1));
+      training_set.add(new Pair<List<BigDecimal>, Boolean>(input_vector, true));
+
+      input_vector.add(new BigDecimal(1));
+      input_vector.add(new BigDecimal(1));
+      input_vector.add(new BigDecimal(0));
+      training_set.add(new Pair<List<BigDecimal>, Boolean>(input_vector, true));
+
+      input_vector.add(new BigDecimal(1));
+      input_vector.add(new BigDecimal(1));
+      input_vector.add(new BigDecimal(1));
+      training_set.add(new Pair<List<BigDecimal>, Boolean>(input_vector, false));
+
+      PerceptronTopologyBuilder ptb = new PerceptronTopologyBuilder(3, 0.5, 0.1);
+      ptb.train(training_set);
+  }
+}

storm-ml/src/jvm/storm/ml/PerceptronTopologyBuilder.java

@@ -0,0 +1,59 @@
+package storm.ml;
+
+import java.lang.Integer;
+import java.lang.Double;
+import java.lang.Boolean;
+import java.math.BigDecimal;
+import java.util.List;
+import java.util.ArrayList;
+
+import org.javatuples.Pair;
+
+public class PerceptronTopologyBuilder {
+    public final Integer size;
+    public final Double threshold;
+    public final Double learning_rate;
+
+    private List<BigDecimal> weights;
+
+    public PerceptronTopologyBuilder(Integer size, Double threshold, Double learning_rate) {
+        this.size = size;                   // size of the weight array and input
+        this.threshold = threshold;         // margin to determine positive results
+        this.learning_rate = learning_rate; // adaptation factor for the weights
+
+        this.weights = new ArrayList<BigDecimal>(size);
+        int i; for (i=0; i<size; i++)
+            this.weights.add(new BigDecimal(0));
+    }
+
+    private BigDecimal dot_product(List<BigDecimal> vector1, List<BigDecimal> vector2) {
+        BigDecimal result = new BigDecimal(0);
+        int i; for (i=0; i<this.size; i++)
+            result.add(vector1.get(i).multiply(vector2.get(i)));
+
+        return result;
+    }
+
+    public void train(List<Pair<List<BigDecimal>, Boolean>> training_set) {
+        while (true) {
+            int error_count = 0;
+            for (Pair<List<BigDecimal>, Boolean> training_pair : training_set) {
+                List<BigDecimal> input_vector = training_pair.getValue0();
+                Integer desired_output        = training_pair.getValue1() ? 1 : 0;
+                System.out.println(String.format("%s", this.weights));
+
+                int result = dot_product(input_vector, this.weights).compareTo(new BigDecimal(threshold)) > 0 ? 1 : 0;
+
+                int error = desired_output - result;
+                if (error != 0) {
+                    error_count += 1;
+                    int i; for (i=0; i<this.size; i++)
+                        this.weights.set(i, this.weights.get(i).add(input_vector.get(i).multiply(new BigDecimal(this.learning_rate * error))));
+                }
+            }
+            if (error_count == 0)
+                break;
+            System.out.println();
+        }
+    }
+}