Renamed evaluate to loss

Author: xEcho1337

src/main/java/net/echo/brain4j/model/Model.java

@@ -117,7 +117,7 @@ public void connect(WeightInit weightInit, boolean update) {
      * @param set dataset for testing
      * @return the error of the model
      */
-    public abstract double evaluate(DataSet<R> set);
+    public abstract double loss(DataSet<R> set);
 
     /**
      * Trains the model for one epoch.
@@ -126,6 +126,18 @@ public void connect(WeightInit weightInit, boolean update) {
      */
     public abstract void fit(DataSet<R> dataSet);
 
+    /**
+     * Trains the model for the given number of epoches.
+     *
+     * @param dataSet dataset for training
+     * @param epoches number of epoches
+     */
+    public void fit(DataSet<R> dataSet, int epoches) {
+        for (int i = 0; i < epoches; i++) {
+            fit(dataSet);
+        }
+    }
+
     /**
      * Predicts the output for the given input.
      *

src/main/java/net/echo/brain4j/model/impl/Sequential.java

@@ -12,7 +12,6 @@
 import net.echo.brain4j.loss.LossFunctions;
 import net.echo.brain4j.model.Model;
 import net.echo.brain4j.model.initialization.WeightInit;
-import net.echo.brain4j.structure.Neuron;
 import net.echo.brain4j.structure.Synapse;
 import net.echo.brain4j.structure.cache.StatesCache;
 import net.echo.brain4j.training.BackPropagation;
@@ -98,7 +97,7 @@ public Sequential compile(WeightInit weightInit, LossFunctions function, Optimiz
     }
 
     @Override
-    public double evaluate(DataSet<DataRow> set) {
+    public double loss(DataSet<DataRow> set) {
         reloadMatrices();
 
         AtomicReference<Double> totalError = new AtomicReference<>(0.0);

src/main/java/net/echo/brain4j/model/impl/Transformer.java

@@ -42,7 +42,7 @@ public void connect(WeightInit weightInit, boolean update) {
     }
 
     @Override
-    public double evaluate(DataSet<Object> set) {
+    public double loss(DataSet<Object> set) {
         return 0;
     }
 

src/main/java/net/echo/brain4j/training/techniques/SmartTrainer.java

@@ -48,7 +48,7 @@ public <R> void start(Model<R, ?, ?> model, DataSet<R> dataSet, double lossThres
 
             if (epoches % evaluateEvery == 0) {
                 long start = System.nanoTime();
-                this.loss = model.evaluate(dataSet);
+                this.loss = model.loss(dataSet);
                 long took = System.nanoTime() - start;
 
                 this.listeners.forEach(listener -> listener.onEvaluated(dataSet, epoches, loss, took));
@@ -91,7 +91,7 @@ public <R> void startFor(Model<R, ?, ?> model, DataSet<R> dataSet, int epochesAm
 
             if (epoches % evaluateEvery == 0) {
                 long start = System.nanoTime();
-                this.loss = model.evaluate(dataSet);
+                this.loss = model.loss(dataSet);
                 long took = System.nanoTime() - start;
 
                 this.listeners.forEach(listener -> listener.onEvaluated(dataSet, epoches, loss, took));

src/main/java/net/echo/brain4j/utils/MLUtils.java

@@ -17,7 +17,7 @@ public class MLUtils {
      * @param <T>     the type of the enum
      * @return the best matching enum constant
      */
-    public static <T extends Enum<T>> T findBestMatch(Vector outputs, Class<T> clazz) {
+    public static <T extends Enum<T>> T parse(Vector outputs, Class<T> clazz) {
         return clazz.getEnumConstants()[indexOfMaxValue(outputs)];
     }
 

src/main/java/net/echo/brain4j/utils/Vector.java

@@ -2,6 +2,7 @@
 
 import java.util.Arrays;
 import java.util.Iterator;
+import java.util.List;
 import java.util.Random;
 import java.util.function.Supplier;
 
@@ -68,6 +69,16 @@ public static Vector zeros(int size) {
         return new Vector(size).fill(0.0);
     }
 
+    public static Vector parse(List<String> pixels) {
+        Vector vector = new Vector(pixels.size());
+
+        for (int i = 0; i < pixels.size(); i++) {
+            vector.set(i, Double.parseDouble(pixels.get(i)));
+        }
+
+        return vector;
+    }
+
     public void set(int index, double value) {
         data[index] = (float) value;
     }

src/test/java/ApproxExample.java

@@ -134,7 +134,7 @@ public void trainAndVisualize() {
                     updateGraph(x, trueOutput, predictedOutput);
                 }
 
-                double error = model.evaluate(dataSet);
+                double error = model.loss(dataSet);
                 errorLabel.setText("Error: " + String.format("%.4f", error) + " Took: " + String.format("%.2f", took) + " ms Epoch: " + i);
                 System.out.print("\rEpoch #" + j + " Error: " + error + " Took: " + took + " ms");
             }

src/test/java/RNNExample.java

@@ -22,7 +22,7 @@ public void start() {
             model.fit(dataSet);
 
             if (i % 1000 == 0) {
-                double loss = model.evaluate(dataSet);
+                double loss = model.loss(dataSet);
 
                 System.out.println(i + ". Loss: " + loss);
             }

src/test/java/XorExample.java

@@ -31,7 +31,7 @@ private void start() {
 
         double took = (System.nanoTime() - start) / 1e6;
 
-        System.out.println("Loss: " + model.evaluate(dataSet));
+        System.out.println("Loss: " + model.loss(dataSet));
         System.out.println("Took: " + took + " ms");
 
         for (DataRow row : dataSet) {

src/test/java/conv/ConvExample.java

@@ -7,20 +7,16 @@
 import net.echo.brain4j.layer.impl.convolution.InputLayer;
 import net.echo.brain4j.loss.LossFunctions;
 import net.echo.brain4j.model.impl.Sequential;
-import net.echo.brain4j.model.initialization.WeightInit;
 import net.echo.brain4j.training.data.DataRow;
 import net.echo.brain4j.training.optimizers.impl.Adam;
-import net.echo.brain4j.training.techniques.SmartTrainer;
-import net.echo.brain4j.training.techniques.TrainListener;
-import net.echo.brain4j.training.updater.impl.StochasticUpdater;
 import net.echo.brain4j.utils.DataSet;
 import net.echo.brain4j.utils.MLUtils;
 import net.echo.brain4j.utils.Vector;
-import org.apache.commons.io.FileUtils;
+import org.apache.commons.csv.CSVFormat;
+import org.apache.commons.csv.CSVParser;
 
-import java.io.File;
+import java.io.FileReader;
 import java.io.IOException;
-import java.util.Arrays;
 import java.util.List;
 
 public class ConvExample {
@@ -36,15 +32,7 @@ private void start() throws IOException {
 
         System.out.println(model.getStats());
 
-        SmartTrainer trainer = new SmartTrainer(1, 1);
-        trainer.addListener(new TrainListener<DataRow>() {
-            @Override
-            public void onEvaluated(DataSet<DataRow> dataSet, int epoch, double loss, long took) {
-                System.out.println("Epoch #" + epoch + " Loss: " + loss);
-            }
-        });
-        trainer.startFor(model, dataSet, 100, 0.01);
-
+        model.fit(dataSet, 10);
         model.save("mnist-conv.json");
 
         int incorrect = 0;
@@ -91,20 +79,22 @@ private Sequential getModel() {
 
     private DataSet<DataRow> getDataSet() throws IOException {
         DataSet<DataRow> dataSet = new DataSet<>();
-        List<String> lines = FileUtils.readLines(new File("dataset.csv"), "UTF-8");
 
-        for (int j = 0; j < 150 * 2; j++) {
-            String line = lines.get(j);
-            String[] parts = line.split(",");
-            double[] inputs = Arrays.stream(parts, 1, parts.length).mapToDouble(x -> Double.parseDouble(x) / 255).toArray();
+        FileReader reader = new FileReader("dataset.csv");
+        CSVParser parser = new CSVParser(reader, CSVFormat.EXCEL);
 
-            Vector output = new Vector(10);
+        parser.forEach(record -> {
+            List<String> columns = record.toList();
 
-            int value = Integer.parseInt(parts[0]);
-            output.set(value, 1);
+            String label = columns.getFirst();
+            List<String> pixels = columns.subList(1, columns.size());
 
-            dataSet.getData().add(new DataRow(Vector.of(inputs), output));
-        }
+            Vector output = new Vector(10);
+            output.set(Integer.parseInt(label), 1);
+
+            Vector input = Vector.parse(pixels).divide(255);
+            dataSet.add(new DataRow(input, output));
+        });
 
         return dataSet;
     }