org.ojalgo.ann.CalculationLayer Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of ojalgo Show documentation
Show all versions of ojalgo Show documentation
oj! Algorithms - ojAlgo - is Open Source Java code that has to do with mathematics, linear algebra and optimisation.
/*
* Copyright 1997-2024 Optimatika
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
package org.ojalgo.ann;
import static org.ojalgo.function.constant.PrimitiveMath.*;
import java.util.function.DoubleUnaryOperator;
import org.ojalgo.function.PrimitiveFunction;
import org.ojalgo.matrix.store.MatrixStore;
import org.ojalgo.matrix.store.PhysicalStore;
import org.ojalgo.random.Uniform;
import org.ojalgo.structure.Structure2D;
final class CalculationLayer {
private ArtificialNeuralNetwork.Activator myActivator;
private final PhysicalStore myBias;
private final PhysicalStore myWeights;
CalculationLayer(final PhysicalStore.Factory factory, final int numberOfInputs, final int numberOfOutputs,
final ArtificialNeuralNetwork.Activator activator) {
super();
myWeights = factory.make(numberOfInputs, numberOfOutputs);
myBias = factory.make(1, numberOfOutputs);
myActivator = activator;
}
@Override
public boolean equals(final Object obj) {
if (this == obj) {
return true;
}
if (obj == null || !(obj instanceof CalculationLayer)) {
return false;
}
CalculationLayer other = (CalculationLayer) obj;
if (myActivator != other.myActivator) {
return false;
}
if (myBias == null) {
if (other.myBias != null) {
return false;
}
} else if (!myBias.equals(other.myBias)) {
return false;
}
if (myWeights == null) {
if (other.myWeights != null) {
return false;
}
} else if (!myWeights.equals(other.myWeights)) {
return false;
}
return true;
}
@Override
public int hashCode() {
final int prime = 31;
int result = 1;
result = prime * result + (myActivator == null ? 0 : myActivator.hashCode());
result = prime * result + (myBias == null ? 0 : myBias.hashCode());
result = prime * result + (myWeights == null ? 0 : myWeights.hashCode());
return result;
}
@Override
public String toString() {
StringBuilder tmpBuilder = new StringBuilder();
tmpBuilder.append("CalculationLayer [Weights=");
tmpBuilder.append(myWeights);
tmpBuilder.append(", Bias=");
tmpBuilder.append(myBias);
tmpBuilder.append(", Activator=");
tmpBuilder.append(myActivator);
tmpBuilder.append("]");
return tmpBuilder.toString();
}
void adjust(final PhysicalStore input, final PhysicalStore output, final PhysicalStore upstreamGradient,
final PhysicalStore downstreamGradient, final double learningRate, final double dropoutsFactor, final DoubleUnaryOperator regularisation) {
downstreamGradient.modifyMatching(MULTIPLY, output.onAll(myActivator.getDerivativeInTermsOfOutput()).transpose());
if (upstreamGradient != null) {
// No need to do this multiplication for the input layer
// input null to stop it...
myWeights.multiply(downstreamGradient, upstreamGradient);
}
if (regularisation != null) {
PrimitiveFunction.Unary modifier = arg -> arg + learningRate * regularisation.applyAsDouble(arg);
myWeights.modifyAll(modifier);
}
for (long j = 0L, nbOutput = myWeights.countColumns(); j < nbOutput; j++) {
for (long b = 0L, batchSize = input.countRows(); b < batchSize; b++) {
double gradient = downstreamGradient.doubleValue(j, b);
double ratedGradient = learningRate * gradient;
myBias.add(j, ratedGradient);
for (long i = 0L, nbInput = myWeights.countRows(); i < nbInput; i++) {
myWeights.add(i, j, ratedGradient * (input.doubleValue(b, i) / dropoutsFactor));
}
}
}
}
int countInputNodes() {
return Math.toIntExact(myWeights.countRows());
}
int countOutputNodes() {
return Math.toIntExact(myWeights.countColumns());
}
ArtificialNeuralNetwork.Activator getActivator() {
return myActivator;
}
double getBias(final int output) {
return myBias.doubleValue(output);
}
MatrixStore getLogicalWeights() {
return myWeights.below(myBias);
}
Structure2D getStructure() {
return myWeights;
}
double getWeight(final int input, final int output) {
return myWeights.doubleValue(input, output);
}
PhysicalStore invoke(final PhysicalStore input, final PhysicalStore output) {
myWeights.premultiply(input).onColumns(ADD, myBias).supplyTo(output);
myActivator.activate(output);
return output;
}
PhysicalStore invoke(final PhysicalStore input, final PhysicalStore output, final double probabilityToKeep) {
myWeights.premultiply(input).onColumns(ADD, myBias).supplyTo(output);
myActivator.activate(output, probabilityToKeep);
return output;
}
void randomise() {
double magnitude = ONE / Math.sqrt(this.countInputNodes());
Uniform randomiser = new Uniform(-magnitude, 2 * magnitude);
myWeights.fillAll(randomiser);
myBias.fillAll(randomiser);
}
void scale(final double factor) {
myWeights.modifyAll(MULTIPLY.second(factor));
}
void setActivator(final ArtificialNeuralNetwork.Activator activator) {
myActivator = activator;
}
void setBias(final int output, final double bias) {
myBias.set(output, bias);
}
void setWeight(final int input, final int output, final double weight) {
myWeights.set(input, output, weight);
}
}