Author: tommaso
Date: Fri Jun 21 19:58:00 2013
New Revision: 1495568
URL: http://svn.apache.org/r1495568
Log:
fixed indent
Modified:
labs/yay/trunk/core/src/main/java/org/apache/yay/core/BackPropagationLearningStrategy.java
labs/yay/trunk/core/src/main/java/org/apache/yay/core/LogisticRegressionCostFunction.java
labs/yay/trunk/core/src/main/java/org/apache/yay/core/TanhFunction.java
labs/yay/trunk/core/src/test/java/org/apache/yay/core/BackPropagationLearningStrategyTest.java
labs/yay/trunk/core/src/test/java/org/apache/yay/core/LogisticRegressionCostFunctionTest.java
Modified:
labs/yay/trunk/core/src/main/java/org/apache/yay/core/BackPropagationLearningStrategy.java
URL:
http://svn.apache.org/viewvc/labs/yay/trunk/core/src/main/java/org/apache/yay/core/BackPropagationLearningStrategy.java?rev=1495568&r1=1495567&r2=1495568&view=diff
==============================================================================
---
labs/yay/trunk/core/src/main/java/org/apache/yay/core/BackPropagationLearningStrategy.java
(original)
+++
labs/yay/trunk/core/src/main/java/org/apache/yay/core/BackPropagationLearningStrategy.java
Fri Jun 21 19:58:00 2013
@@ -78,7 +78,7 @@ public class BackPropagationLearningStra
double newCost =
costFunction.calculateAggregatedCost(trainingExamples, hypothesis);
if (newCost > cost) {
- throw new RuntimeException("failed to converge at iteration " +
iterations + " with alpha "+ alpha +" : cost going from " + cost + " to " +
newCost);
+ throw new RuntimeException("failed to converge at iteration " +
iterations + " with alpha " + alpha + " : cost going from " + cost + " to " +
newCost);
} else if (cost == newCost || newCost < threshold || iterations >
MAX_ITERATIONS) {
System.out.println("successfully converged with alpha " + alpha + "
after " + iterations + " iterations with cost " + newCost + " and parameters "
+ Arrays.toString(hypothesis.getParameters()));
break;
@@ -97,91 +97,90 @@ public class BackPropagationLearningStra
hypothesis.setParameters(updatedWeights);
iterations++;
- }
- }
- catch (Exception e) {
- throw new WeightLearningException("error during backprop learning", e);
}
+ } catch (Exception e) {
+ throw new WeightLearningException("error during backprop learning", e);
+ }
return updatedWeights;
}
- private RealMatrix[] calculateDerivatives(RealMatrix[] weightsMatrixSet,
TrainingSet<Double, Double> trainingExamples) throws WeightLearningException {
- // set up the accumulator matrix(es)
- RealMatrix[] triangle = new RealMatrix[weightsMatrixSet.length];
- RealVector[] deltaVectors = new RealVector[weightsMatrixSet.length];
-
- int noOfMatrixes = weightsMatrixSet.length - 1;
- double count = 0;
- for (TrainingExample<Double, Double> trainingExample :
trainingExamples) {
- try {
- // get activations from feed forward propagation
- RealVector[] activations =
predictionStrategy.debugOutput(ConversionUtils.toValuesCollection(trainingExample.getFeatures()),
weightsMatrixSet);
-
- // calculate output error (corresponding to the last delta^l)
- RealVector nextLayerDelta = calculateOutputError(trainingExample,
activations);
-
- deltaVectors[noOfMatrixes] = nextLayerDelta;
-
- // back prop the error and update the deltas accordingly
- for (int l = noOfMatrixes; l > 0; l--) {
- RealVector currentActivationsVector = activations[l - 1];
- nextLayerDelta = calculateDeltaVector(weightsMatrixSet[l],
currentActivationsVector, nextLayerDelta);
-
- // collect delta vectors for this example
- deltaVectors[l - 1] = nextLayerDelta;
- }
-
- RealVector[] newActivations = new RealVector[activations.length];
- newActivations[0] =
ConversionUtils.toRealVector(ConversionUtils.toValuesCollection(trainingExample.getFeatures()));
- for (int k = 0; k < activations.length - 1; k++) {
- newActivations[k+1] = activations[k];
- }
-
- // update triangle (big delta matrix)
- updateTriangle(triangle, newActivations, deltaVectors,
weightsMatrixSet);
-
- } catch (Exception e) {
- throw new WeightLearningException("error during derivatives
calculation", e);
- }
- count++;
+ private RealMatrix[] calculateDerivatives(RealMatrix[] weightsMatrixSet,
TrainingSet<Double, Double> trainingExamples) throws WeightLearningException {
+ // set up the accumulator matrix(es)
+ RealMatrix[] triangle = new RealMatrix[weightsMatrixSet.length];
+ RealVector[] deltaVectors = new RealVector[weightsMatrixSet.length];
+
+ int noOfMatrixes = weightsMatrixSet.length - 1;
+ double count = 0;
+ for (TrainingExample<Double, Double> trainingExample : trainingExamples) {
+ try {
+ // get activations from feed forward propagation
+ RealVector[] activations =
predictionStrategy.debugOutput(ConversionUtils.toValuesCollection(trainingExample.getFeatures()),
weightsMatrixSet);
+
+ // calculate output error (corresponding to the last delta^l)
+ RealVector nextLayerDelta = calculateOutputError(trainingExample,
activations);
+
+ deltaVectors[noOfMatrixes] = nextLayerDelta;
+
+ // back prop the error and update the deltas accordingly
+ for (int l = noOfMatrixes; l > 0; l--) {
+ RealVector currentActivationsVector = activations[l - 1];
+ nextLayerDelta = calculateDeltaVector(weightsMatrixSet[l],
currentActivationsVector, nextLayerDelta);
+
+ // collect delta vectors for this example
+ deltaVectors[l - 1] = nextLayerDelta;
}
- return createDerivatives(triangle, count);
+ RealVector[] newActivations = new RealVector[activations.length];
+ newActivations[0] =
ConversionUtils.toRealVector(ConversionUtils.toValuesCollection(trainingExample.getFeatures()));
+ for (int k = 0; k < activations.length - 1; k++) {
+ newActivations[k + 1] = activations[k];
+ }
+
+ // update triangle (big delta matrix)
+ updateTriangle(triangle, newActivations, deltaVectors,
weightsMatrixSet);
+
+ } catch (Exception e) {
+ throw new WeightLearningException("error during derivatives
calculation", e);
+ }
+ count++;
}
- private RealMatrix[] updateWeights(RealMatrix[] weightsMatrixSet,
RealMatrix[] derivatives, double alpha) {
- RealMatrix[] updatedParameters = new
RealMatrix[weightsMatrixSet.length];
- for (int l = 0; l < weightsMatrixSet.length; l++) {
- double[][] updatedWeights = weightsMatrixSet[l].getData();
- for (int i = 0; i < updatedWeights.length; i++) {
- for (int j = 0; j < updatedWeights[i].length; j++) {
- updatedWeights[i][j] = updatedWeights[i][j] - alpha *
derivatives[l].getData()[i][j];
- }
- }
- updatedParameters[l] = new Array2DRowRealMatrix(updatedWeights);
+ return createDerivatives(triangle, count);
+ }
+
+ private RealMatrix[] updateWeights(RealMatrix[] weightsMatrixSet,
RealMatrix[] derivatives, double alpha) {
+ RealMatrix[] updatedParameters = new RealMatrix[weightsMatrixSet.length];
+ for (int l = 0; l < weightsMatrixSet.length; l++) {
+ double[][] updatedWeights = weightsMatrixSet[l].getData();
+ for (int i = 0; i < updatedWeights.length; i++) {
+ for (int j = 0; j < updatedWeights[i].length; j++) {
+ updatedWeights[i][j] = updatedWeights[i][j] - alpha *
derivatives[l].getData()[i][j];
}
- return updatedParameters;
+ }
+ updatedParameters[l] = new Array2DRowRealMatrix(updatedWeights);
}
+ return updatedParameters;
+ }
- private RealMatrix[] createDerivatives(RealMatrix[] triangle, double
count) {
- RealMatrix[] derivatives = new RealMatrix[triangle.length];
- for (int i = 0; i < triangle.length; i++) {
- // TODO : introduce regularization diversification on bias term
(currently not regularized)
- derivatives[i] = triangle[i].scalarMultiply(1d / count);
- }
- return derivatives;
+ private RealMatrix[] createDerivatives(RealMatrix[] triangle, double count) {
+ RealMatrix[] derivatives = new RealMatrix[triangle.length];
+ for (int i = 0; i < triangle.length; i++) {
+ // TODO : introduce regularization diversification on bias term
(currently not regularized)
+ derivatives[i] = triangle[i].scalarMultiply(1d / count);
}
+ return derivatives;
+ }
- private void updateTriangle(RealMatrix[] triangle, RealVector[]
activations, RealVector[] deltaVectors, RealMatrix[] weightsMatrixSet) {
- for (int l = weightsMatrixSet.length - 1; l >= 0; l--) {
- RealMatrix realMatrix = deltaVectors[l].outerProduct(activations[l]);
- if (triangle[l] == null) {
- triangle[l] = realMatrix;
- } else {
- triangle[l] = triangle[l].add(realMatrix);
- }
+ private void updateTriangle(RealMatrix[] triangle, RealVector[] activations,
RealVector[] deltaVectors, RealMatrix[] weightsMatrixSet) {
+ for (int l = weightsMatrixSet.length - 1; l >= 0; l--) {
+ RealMatrix realMatrix = deltaVectors[l].outerProduct(activations[l]);
+ if (triangle[l] == null) {
+ triangle[l] = realMatrix;
+ } else {
+ triangle[l] = triangle[l].add(realMatrix);
}
+ }
}
private RealVector calculateDeltaVector(RealMatrix thetaL, RealVector
activationsVector, RealVector nextLayerDelta) {
@@ -198,11 +197,9 @@ public class BackPropagationLearningStra
int sampleOutputIntValue = trainingExample.getOutput().intValue();
if (sampleOutputIntValue < sampleOutput.length) {
sampleOutput[sampleOutputIntValue] = 1d;
- }
- else if (sampleOutput.length == 1){
+ } else if (sampleOutput.length == 1) {
sampleOutput[0] = trainingExample.getOutput();
- }
- else {
+ } else {
throw new RuntimeException("problem with multiclass output mapping");
}
RealVector learnedOutputRealVector = new ArrayRealVector(sampleOutput); //
turn example output to a vector
Modified:
labs/yay/trunk/core/src/main/java/org/apache/yay/core/LogisticRegressionCostFunction.java
URL:
http://svn.apache.org/viewvc/labs/yay/trunk/core/src/main/java/org/apache/yay/core/LogisticRegressionCostFunction.java?rev=1495568&r1=1495567&r2=1495568&view=diff
==============================================================================
---
labs/yay/trunk/core/src/main/java/org/apache/yay/core/LogisticRegressionCostFunction.java
(original)
+++
labs/yay/trunk/core/src/main/java/org/apache/yay/core/LogisticRegressionCostFunction.java
Fri Jun 21 19:58:00 2013
@@ -44,7 +44,7 @@ public class LogisticRegressionCostFunct
@Override
public Double calculateAggregatedCost(TrainingSet<Double, Double>
trainingExamples,
- Hypothesis<RealMatrix, Double, Double>
hypothesis) throws Exception {
+ Hypothesis<RealMatrix, Double, Double>
hypothesis) throws Exception {
Double errorTerm = calculateErrorTerm(hypothesis, trainingExamples);
Double regularizationTerm = calculateRegularizationTerm(hypothesis,
trainingExamples);
Modified:
labs/yay/trunk/core/src/main/java/org/apache/yay/core/TanhFunction.java
URL:
http://svn.apache.org/viewvc/labs/yay/trunk/core/src/main/java/org/apache/yay/core/TanhFunction.java?rev=1495568&r1=1495567&r2=1495568&view=diff
==============================================================================
--- labs/yay/trunk/core/src/main/java/org/apache/yay/core/TanhFunction.java
(original)
+++ labs/yay/trunk/core/src/main/java/org/apache/yay/core/TanhFunction.java Fri
Jun 21 19:58:00 2013
@@ -24,8 +24,8 @@ import org.apache.yay.ActivationFunction
* Tanh activation function
*/
public class TanhFunction implements ActivationFunction<Double> {
- @Override
- public Double apply(Double signal) {
- return Math.tanh(signal);
- }
+ @Override
+ public Double apply(Double signal) {
+ return Math.tanh(signal);
+ }
}
Modified:
labs/yay/trunk/core/src/test/java/org/apache/yay/core/BackPropagationLearningStrategyTest.java
URL:
http://svn.apache.org/viewvc/labs/yay/trunk/core/src/test/java/org/apache/yay/core/BackPropagationLearningStrategyTest.java?rev=1495568&r1=1495567&r2=1495568&view=diff
==============================================================================
---
labs/yay/trunk/core/src/test/java/org/apache/yay/core/BackPropagationLearningStrategyTest.java
(original)
+++
labs/yay/trunk/core/src/test/java/org/apache/yay/core/BackPropagationLearningStrategyTest.java
Fri Jun 21 19:58:00 2013
@@ -18,9 +18,6 @@
*/
package org.apache.yay.core;
-import java.util.ArrayList;
-import java.util.Collection;
-
import org.apache.commons.math3.linear.Array2DRowRealMatrix;
import org.apache.commons.math3.linear.RealMatrix;
import org.apache.yay.PredictionStrategy;
@@ -29,6 +26,9 @@ import org.apache.yay.TrainingSet;
import org.apache.yay.core.utils.ExamplesFactory;
import org.junit.Test;
+import java.util.ArrayList;
+import java.util.Collection;
+
import static junit.framework.Assert.assertFalse;
import static junit.framework.Assert.assertNotNull;
@@ -90,7 +90,7 @@ public class BackPropagationLearningStra
RealMatrix[] initialWeights = new RealMatrix[3];
initialWeights[0] = new Array2DRowRealMatrix(new double[][]{{0d, 0d, 0d},
{1d, Math.random(), Math.random()}, {1d, Math.random(), Math.random()}, {1d,
Math.random(), Math.random()}});
initialWeights[1] = new Array2DRowRealMatrix(new double[][]{{0d, 0d, 0d,
0d}, {1d, Math.random(), Math.random(), Math.random()}, {1d, Math.random(),
Math.random(), Math.random()}, {1d, Math.random(), Math.random(),
Math.random()}});
- initialWeights[2] = new Array2DRowRealMatrix(new
double[][]{{1d,Math.random(), Math.random(), Math.random()}});
+ initialWeights[2] = new Array2DRowRealMatrix(new double[][]{{1d,
Math.random(), Math.random(), Math.random()}});
Collection<TrainingExample<Double, Double>> samples =
createSamples(500000, 2);
TrainingSet<Double, Double> trainingSet = new TrainingSet<Double,
Double>(samples);
@@ -106,10 +106,10 @@ public class BackPropagationLearningStra
Collection<TrainingExample<Double, Double>> trainingExamples = new
ArrayList<TrainingExample<Double, Double>>(size);
Double[] featureValues = new Double[noOfFeatures];
for (int i = 0; i < size; i++) {
- for (int j = 0; j < noOfFeatures; j++) {
- featureValues[j] = Math.random();
- }
- trainingExamples.add(ExamplesFactory.createDoubleTrainingExample(1d,
featureValues));
+ for (int j = 0; j < noOfFeatures; j++) {
+ featureValues[j] = Math.random();
+ }
+ trainingExamples.add(ExamplesFactory.createDoubleTrainingExample(1d,
featureValues));
}
return trainingExamples;
}
Modified:
labs/yay/trunk/core/src/test/java/org/apache/yay/core/LogisticRegressionCostFunctionTest.java
URL:
http://svn.apache.org/viewvc/labs/yay/trunk/core/src/test/java/org/apache/yay/core/LogisticRegressionCostFunctionTest.java?rev=1495568&r1=1495567&r2=1495568&view=diff
==============================================================================
---
labs/yay/trunk/core/src/test/java/org/apache/yay/core/LogisticRegressionCostFunctionTest.java
(original)
+++
labs/yay/trunk/core/src/test/java/org/apache/yay/core/LogisticRegressionCostFunctionTest.java
Fri Jun 21 19:58:00 2013
@@ -24,12 +24,6 @@ import org.apache.yay.CostFunction;
import org.apache.yay.NeuralNetwork;
import org.apache.yay.TrainingExample;
import org.apache.yay.TrainingSet;
-import org.apache.yay.core.FeedForwardStrategy;
-import org.apache.yay.core.LogisticRegressionCostFunction;
-import org.apache.yay.core.MaxSelectionFunction;
-import org.apache.yay.core.NeuralNetworkFactory;
-import org.apache.yay.core.SigmoidFunction;
-import org.apache.yay.core.VoidLearningStrategy;
import org.apache.yay.core.utils.ExamplesFactory;
import org.junit.Before;
import org.junit.Test;
@@ -44,7 +38,7 @@ import static org.junit.Assert.assertTru
*/
public class LogisticRegressionCostFunctionTest {
- private CostFunction<RealMatrix,Double,Double> costFunction;
+ private CostFunction<RealMatrix, Double, Double> costFunction;
private TrainingSet<Double, Double> trainingSet;
@Before
@@ -62,7 +56,6 @@ public class LogisticRegressionCostFunct
trainingExamples.add(example4);
trainingSet = new TrainingSet<Double, Double>(trainingExamples);
-
}
@Test
@@ -75,7 +68,6 @@ public class LogisticRegressionCostFunct
new VoidLearningStrategy<Double, Double>(), new
FeedForwardStrategy(
new SigmoidFunction()), new MaxSelectionFunction<Double>());
-
Double cost = costFunction.calculateAggregatedCost(trainingSet,
neuralNetwork);
assertTrue("cost should not be negative", cost > 0d);
}
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
