+ * TODO: normalize input? + */ + private static void prepareDataForLearning(String fileName, boolean randomizeDataOrder, double trainingPercentage) throws IOException { + BufferedReader br = new BufferedReader(new FileReader(DATA_DIR_NAME + "/" + fileName)); + String currentLine = br.readLine(); // skip header + int expectedLineLength = currentLine.split(",").length; + BufferedWriter bwi = new BufferedWriter(new FileWriter(PREPARED_DATA_DIR_NAME + "/" + LEARNING_INPUT_FILE_NAME)); + BufferedWriter bwo = new BufferedWriter(new FileWriter(PREPARED_DATA_DIR_NAME + "/" + LEARNING_OUTPUT_FILE_NAME)); + BufferedWriter bwt = new BufferedWriter(new FileWriter(PREPARED_DATA_DIR_NAME + "/" + TESTING_FILE_NAME)); + ArrayList allLinesAsArrays = new ArrayList<>(); + String[] currentLineAsArray; + int currentLineLength; + + while ((currentLine = br.readLine()) != null) { + if (currentLine.contains("NA")) { + // we cannot use this for training + LOGGER.fine("Invalid line (contains NA) will be ignored: " + currentLine); + continue; + } + currentLineAsArray = currentLine.split(","); + currentLineLength = currentLineAsArray.length; + if (currentLineLength != expectedLineLength) { + // invalid formatting + LOGGER.fine("Invalid line (invalid format) will be ignored: " + currentLine); + continue; + } + + allLinesAsArrays.add(currentLineAsArray); + } + + if (randomizeDataOrder) { + Collections.shuffle(allLinesAsArrays); + } + + int dataSetLength = allLinesAsArrays.size(); + int maxIndexForTraining = (int) (dataSetLength * trainingPercentage); + for (String[] lineAsArray : allLinesAsArrays.subList(0, maxIndexForTraining)) { + bwi.write(String.join(",", Arrays.copyOfRange(lineAsArray, 1, expectedLineLength - 1)) + "\n"); // ignore field "id" + bwo.write(lineAsArray[expectedLineLength - 1] + "\n"); // last field has output + } + for (String[] lineAsArray : allLinesAsArrays.subList(maxIndexForTraining, dataSetLength - 1)) { + bwt.write(String.join(",", Arrays.copyOfRange(lineAsArray, 1, lineAsArray.length)) + "\n"); + } + + bwt.close(); + bwi.close(); + bwo.close(); + br.close(); + } + + private static void config() { + Locale.setDefault(Locale.ENGLISH); + System.setProperty("java.util.logging.SimpleFormatter.format", "%1$tF %1$tT %4$s\t%5$s%6$s%n"); + } +} \ No newline at end of file diff --git a/src/com/jlmd/simpleneuralnetwork/app/Mode.java b/src/com/jlmd/simpleneuralnetwork/app/Mode.java new file mode 100644 index 0000000..08cf890 --- /dev/null +++ b/src/com/jlmd/simpleneuralnetwork/app/Mode.java @@ -0,0 +1,28 @@ +package com.jlmd.simpleneuralnetwork.app; + +public enum Mode { + TRAIN("train"), TEST("test"), BATCH("batch"); + + private final String name; + + Mode(String name) { + this.name = name; + } + + public String getValue() { + return name; + } + + /** + * @param str Value of the enum item + * @return enum item or null if not found in enum + */ + public static Mode getModeFromString(String str) { + for (Mode m : values()) { + if (m.getValue().equals(str)) { + return m; + } + } + return null; + } +} diff --git a/src/com/jlmd/simpleneuralnetwork/app/TestResult.java b/src/com/jlmd/simpleneuralnetwork/app/TestResult.java new file mode 100644 index 0000000..0dc11f7 --- /dev/null +++ b/src/com/jlmd/simpleneuralnetwork/app/TestResult.java @@ -0,0 +1,65 @@ +package com.jlmd.simpleneuralnetwork.app; + +import java.util.ArrayList; + +/** + * Simple class to evaluate tests performed on a binary classifier + * TODO: implement real tools such as precision, recall, false positives, etc. + */ +public class TestResult { + + private int testedInputsForClass0 = 0; + private int testedInputsForClass1 = 0; + private int correctPredictionsForClass0 = 0; + private int correctPredictionsForClass1 = 0; + + public double getPrecisionForClass0() { + return ((double) this.correctPredictionsForClass0) / this.testedInputsForClass0; + } + + public double getPrecisionForClass1() { + return ((double) this.correctPredictionsForClass1) / this.testedInputsForClass1; + } + + public double getOverallPrecision() { + return ((double)(this.correctPredictionsForClass0 + this.correctPredictionsForClass1)) / (this.testedInputsForClass0 + this.testedInputsForClass1); + } + + public double getRatioClass0OverClass1() { + return ((double) this.testedInputsForClass0) / this.testedInputsForClass1; + } + + public int getTotalNumberOfTestsPerformed() { + return this.testedInputsForClass0 + this.testedInputsForClass1; + } + + public void addPrediction(int predictedClass, int correctClass) { + if (correctClass == 0) { + this.testedInputsForClass0++; + if (predictedClass == 0) { + this.correctPredictionsForClass0++; + } + } else { + this.testedInputsForClass1++; + if (predictedClass == 1) { + this.correctPredictionsForClass1++; + } + } + } + + public ArrayList exportMeasurements() { + ArrayList measurements = new ArrayList<>(); + measurements.add("NbTests"); + measurements.add(this.getTotalNumberOfTestsPerformed()); + measurements.add("RatioClass0OverClass1"); + measurements.add(this.getRatioClass0OverClass1()); + measurements.add("OverallPrecision"); + measurements.add(this.getOverallPrecision()); + measurements.add("PrecisionForClass0"); + measurements.add(this.getPrecisionForClass0()); + measurements.add("PrecisionForClass1"); + measurements.add(this.getPrecisionForClass1()); + + return measurements; + } +} \ No newline at end of file diff --git a/src/com/jlmd/simpleneuralnetwork/example/Main.java b/src/com/jlmd/simpleneuralnetwork/example/Main.java deleted file mode 100644 index d1a522f..0000000 --- a/src/com/jlmd/simpleneuralnetwork/example/Main.java +++ /dev/null @@ -1,35 +0,0 @@ -package com.jlmd.simpleneuralnetwork.example; - -import com.jlmd.simpleneuralnetwork.neuralnetwork.entity.Result; -import com.jlmd.simpleneuralnetwork.neuralnetwork.utils.DataUtils; -import com.jlmd.simpleneuralnetwork.neuralnetwork.NeuralNetwork; -import com.jlmd.simpleneuralnetwork.neuralnetwork.entity.Error; -import com.jlmd.simpleneuralnetwork.neuralnetwork.callback.INeuralNetworkCallback; - -/** - * @author jlmd - */ -public class Main { - public static void main(String [] args){ - System.out.println("Starting neural network sample... "); - - float[][] x = DataUtils.readInputsFromFile("data/x.txt"); - int[] t = DataUtils.readOutputsFromFile("data/t.txt"); - - NeuralNetwork neuralNetwork = new NeuralNetwork(x, t, new INeuralNetworkCallback() { - @Override - public void success(Result result) { - float[] valueToPredict = new float[] {-0.205f, 0.780f}; - System.out.println("Success percentage: " + result.getSuccessPercentage()); - System.out.println("Predicted result: " + result.predictValue(valueToPredict)); - } - - @Override - public void failure(Error error) { - System.out.println("Error: " + error.getDescription()); - } - }); - - neuralNetwork.startLearning(); - } -} diff --git a/src/com/jlmd/simpleneuralnetwork/neuralnetwork/Analyzer.java b/src/com/jlmd/simpleneuralnetwork/neuralnetwork/Analyzer.java index 0b91026..4f9fdcb 100644 --- a/src/com/jlmd/simpleneuralnetwork/neuralnetwork/Analyzer.java +++ b/src/com/jlmd/simpleneuralnetwork/neuralnetwork/Analyzer.java @@ -2,11 +2,13 @@ import com.jlmd.simpleneuralnetwork.neuralnetwork.transfer.ITransferFunction; +import java.io.Serializable; + /** * Generate out function using weights and input values * @author jlmd */ -public class Analyzer { +public class Analyzer implements Serializable { private float[] fOutArray; private float[][] wWeights; private float[] vWeights; diff --git a/src/com/jlmd/simpleneuralnetwork/neuralnetwork/entity/Result.java b/src/com/jlmd/simpleneuralnetwork/neuralnetwork/entity/Result.java index 97cce79..2d92135 100644 --- a/src/com/jlmd/simpleneuralnetwork/neuralnetwork/entity/Result.java +++ b/src/com/jlmd/simpleneuralnetwork/neuralnetwork/entity/Result.java @@ -3,11 +3,13 @@ import com.jlmd.simpleneuralnetwork.neuralnetwork.Analyzer; import com.jlmd.simpleneuralnetwork.neuralnetwork.parser.result.IResultParser; +import java.io.Serializable; + /** * Entity to save obtained result * @author jlmd */ -public class Result { +public class Result implements Serializable { private float successPercentage; private float quadraticError; private Analyzer analyzer; diff --git a/src/com/jlmd/simpleneuralnetwork/neuralnetwork/parser/result/BinaryResultParser.java b/src/com/jlmd/simpleneuralnetwork/neuralnetwork/parser/result/BinaryResultParser.java index 42aef8d..f73c247 100644 --- a/src/com/jlmd/simpleneuralnetwork/neuralnetwork/parser/result/BinaryResultParser.java +++ b/src/com/jlmd/simpleneuralnetwork/neuralnetwork/parser/result/BinaryResultParser.java @@ -1,10 +1,12 @@ package com.jlmd.simpleneuralnetwork.neuralnetwork.parser.result; +import java.io.Serializable; + /** * Binary implementation of result parser * @author jlmd */ -public class BinaryResultParser implements IResultParser { +public class BinaryResultParser implements IResultParser, Serializable { @Override public int countSuccesses(int success, float fOut, float t){ if ((fOut < 0.5 && t == 0) || (fOut >= 0.5 && t == 1)) diff --git a/src/com/jlmd/simpleneuralnetwork/neuralnetwork/transfer/SigmoidFunction.java b/src/com/jlmd/simpleneuralnetwork/neuralnetwork/transfer/SigmoidFunction.java index e10b47b..e48c657 100644 --- a/src/com/jlmd/simpleneuralnetwork/neuralnetwork/transfer/SigmoidFunction.java +++ b/src/com/jlmd/simpleneuralnetwork/neuralnetwork/transfer/SigmoidFunction.java @@ -1,10 +1,12 @@ package com.jlmd.simpleneuralnetwork.neuralnetwork.transfer; +import java.io.Serializable; + /** * Sigmoid implementation of transfer function that limit the value between 0 and 1 * @author jlmd */ -public class SigmoidFunction implements ITransferFunction { +public class SigmoidFunction implements ITransferFunction, Serializable { @Override public float transfer(float value) { return (float)(1/(1+Math.exp(-value)));