org.nd4j.evaluation.classification.ROCMultiClass Maven / Gradle / Ivy
/*******************************************************************************
* Copyright (c) 2015-2018 Skymind, Inc.
*
* This program and the accompanying materials are made available under the
* terms of the Apache License, Version 2.0 which is available at
* https://www.apache.org/licenses/LICENSE-2.0.
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
* License for the specific language governing permissions and limitations
* under the License.
*
* SPDX-License-Identifier: Apache-2.0
******************************************************************************/
package org.nd4j.evaluation.classification;
import lombok.Data;
import lombok.EqualsAndHashCode;
import org.nd4j.base.Preconditions;
import org.nd4j.evaluation.BaseEvaluation;
import org.nd4j.evaluation.IEvaluation;
import org.nd4j.evaluation.IMetric;
import org.nd4j.evaluation.classification.ROC.Metric;
import org.nd4j.evaluation.curves.PrecisionRecallCurve;
import org.nd4j.evaluation.curves.RocCurve;
import org.nd4j.evaluation.serde.ROCArraySerializer;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.primitives.Triple;
import org.nd4j.shade.jackson.databind.annotation.JsonSerialize;
import java.io.Serializable;
import java.util.Arrays;
import java.util.List;
/**
* ROC (Receiver Operating Characteristic) for multi-class classifiers.
As per {@link ROC}, ROCBinary supports both exact (thersholdSteps == 0) and thresholded; see {@link ROC} for details.
*
* The ROC curves are produced by treating the predictions as a set of one-vs-all classifiers, and then calculating
* ROC curves for each. In practice, this means for N classes, we get N ROC curves.
*
* @author Alex Black
*/
@Data
@EqualsAndHashCode(callSuper = true)
public class ROCMultiClass extends BaseEvaluation {
public static final int DEFAULT_STATS_PRECISION = 4;
/**
* AUROC: Area under ROC curve
* AUPRC: Area under Precision-Recall Curve
*/
public enum Metric implements IMetric {AUROC, AUPRC;
@Override
public Class getEvaluationClass() {
return ROCMultiClass.class;
}
@Override
public boolean minimize() {
return false;
}
}
private int thresholdSteps;
private boolean rocRemoveRedundantPts;
@JsonSerialize(using = ROCArraySerializer.class)
private ROC[] underlying;
private List labels;
@EqualsAndHashCode.Exclude //Exclude axis: otherwise 2 Evaluation instances could contain identical stats and fail equality
protected int axis = 1;
protected ROCMultiClass(int axis, int thresholdSteps, boolean rocRemoveRedundantPts, List labels) {
this.thresholdSteps = thresholdSteps;
this.rocRemoveRedundantPts = rocRemoveRedundantPts;
this.axis = axis;
this.labels = labels;
}
public ROCMultiClass() {
//Default to exact
this(0);
}
/**
* @param thresholdSteps Number of threshold steps to use for the ROC calculation. Set to 0 for exact ROC calculation
*/
public ROCMultiClass(int thresholdSteps) {
this(thresholdSteps, true);
}
/**
* @param thresholdSteps Number of threshold steps to use for the ROC calculation. If set to 0: use exact calculation
* @param rocRemoveRedundantPts Usually set to true. If true, remove any redundant points from ROC and P-R curves
*/
public ROCMultiClass(int thresholdSteps, boolean rocRemoveRedundantPts) {
this.thresholdSteps = thresholdSteps;
this.rocRemoveRedundantPts = rocRemoveRedundantPts;
}
/**
* Set the axis for evaluation - this is the dimension along which the probability (and label classes) are present.
* For DL4J, this can be left as the default setting (axis = 1).
* Axis should be set as follows:
* For 2D (OutputLayer), shape [minibatch, numClasses] - axis = 1
* For 3D, RNNs/CNN1D (DL4J RnnOutputLayer), NCW format, shape [minibatch, numClasses, sequenceLength] - axis = 1
* For 3D, RNNs/CNN1D (DL4J RnnOutputLayer), NWC format, shape [minibatch, sequenceLength, numClasses] - axis = 2
* For 4D, CNN2D (DL4J CnnLossLayer), NCHW format, shape [minibatch, channels, height, width] - axis = 1
* For 4D, CNN2D, NHWC format, shape [minibatch, height, width, channels] - axis = 3
*
* @param axis Axis to use for evaluation
*/
public void setAxis(int axis){
this.axis = axis;
}
/**
* Get the axis - see {@link #setAxis(int)} for details
*/
public int getAxis(){
return axis;
}
@Override
public void reset() {
underlying = null;
}
@Override
public String stats() {
return stats(DEFAULT_STATS_PRECISION);
}
public String stats(int printPrecision) {
StringBuilder sb = new StringBuilder();
int maxLabelsLength = 15;
if (labels != null) {
for (String s : labels) {
maxLabelsLength = Math.max(s.length(), maxLabelsLength);
}
}
String patternHeader = "%-" + (maxLabelsLength + 5) + "s%-12s%-10s%-10s";
String header = String.format(patternHeader, "Label", "AUC", "# Pos", "# Neg");
String pattern = "%-" + (maxLabelsLength + 5) + "s" //Label
+ "%-12." + printPrecision + "f" //AUC
+ "%-10d%-10d"; //Count pos, count neg
sb.append(header);
if (underlying != null) {
for (int i = 0; i < underlying.length; i++) {
double auc = calculateAUC(i);
String label = (labels == null ? String.valueOf(i) : labels.get(i));
sb.append("\n").append(String.format(pattern, label, auc, getCountActualPositive(i),
getCountActualNegative(i)));
}
sb.append("Average AUC: ").append(String.format("%-12." + printPrecision + "f", calculateAverageAUC()));
if(thresholdSteps > 0){
sb.append("\n");
sb.append("[Note: Thresholded AUC/AUPRC calculation used with ").append(thresholdSteps)
.append(" steps); accuracy may reduced compared to exact mode]");
}
} else {
//Empty evaluation
sb.append("\n-- No Data --\n");
}
return sb.toString();
}
/**
* Evaluate the network, with optional metadata
*
* @param labels Data labels
* @param predictions Network predictions
* @param recordMetaData Optional; may be null. If not null, should have size equal to the number of outcomes/guesses
*
*/
@Override
public void eval(INDArray labels, INDArray predictions, INDArray mask, final List recordMetaData) {
Triple p = BaseEvaluation.reshapeAndExtractNotMasked(labels, predictions, mask, axis);
if(p == null){
//All values masked out; no-op
return;
}
INDArray labels2d = p.getFirst();
INDArray predictions2d = p.getSecond();
INDArray maskArray = p.getThird();
Preconditions.checkState(maskArray == null, "Per-output masking for ROCMultiClass is not supported");
if(labels2d.dataType() != predictions2d.dataType())
labels2d = labels2d.castTo(predictions2d.dataType());
// FIXME: int cast
int n = (int) labels2d.size(1);
if (underlying == null) {
underlying = new ROC[n];
for (int i = 0; i < n; i++) {
underlying[i] = new ROC(thresholdSteps, rocRemoveRedundantPts);
}
}
if (underlying.length != labels2d.size(1)) {
throw new IllegalArgumentException(
"Cannot evaluate data: number of label classes does not match previous call. " + "Got "
+ labels2d.size(1) + " labels (from array shape "
+ Arrays.toString(labels2d.shape()) + ")"
+ " vs. expected number of label classes = " + underlying.length);
}
for (int i = 0; i < n; i++) {
INDArray prob = predictions2d.getColumn(i, true); //Probability of class i
INDArray label = labels2d.getColumn(i, true);
//Workaround for: https://github.com/deeplearning4j/deeplearning4j/issues/7305
if(prob.rank() == 0)
prob = prob.reshape(1,1);
if(label.rank() == 0)
label = label.reshape(1,1);
underlying[i].eval(label, prob);
}
}
/**
* Get the (one vs. all) ROC curve for the specified class
* @param classIdx Class index to get the ROC curve for
* @return ROC curve for the given class
*/
public RocCurve getRocCurve(int classIdx) {
assertIndex(classIdx);
return underlying[classIdx].getRocCurve();
}
/**
* Get the (one vs. all) Precision-Recall curve for the specified class
* @param classIdx Class to get the P-R curve for
* @return Precision recall curve for the given class
*/
public PrecisionRecallCurve getPrecisionRecallCurve(int classIdx) {
assertIndex(classIdx);
return underlying[classIdx].getPrecisionRecallCurve();
}
/**
* Calculate the AUC - Area Under ROC Curve
* Utilizes trapezoidal integration internally
*
* @return AUC
*/
public double calculateAUC(int classIdx) {
assertIndex(classIdx);
return underlying[classIdx].calculateAUC();
}
/**
* Calculate the AUPRC - Area Under Curve Precision Recall
* Utilizes trapezoidal integration internally
*
* @return AUC
*/
public double calculateAUCPR(int classIdx) {
assertIndex(classIdx);
return underlying[classIdx].calculateAUCPR();
}
/**
* Calculate the macro-average (one-vs-all) AUC for all classes
*/
public double calculateAverageAUC() {
assertIndex(0);
double sum = 0.0;
for (int i = 0; i < underlying.length; i++) {
sum += calculateAUC(i);
}
return sum / underlying.length;
}
/**
* Calculate the macro-average (one-vs-all) AUCPR (area under precision recall curve) for all classes
*/
public double calculateAverageAUCPR() {
double sum = 0.0;
for (int i = 0; i < underlying.length; i++) {
sum += calculateAUCPR(i);
}
return sum / underlying.length;
}
/**
* Get the actual positive count (accounting for any masking) for the specified class
*
* @param outputNum Index of the class
*/
public long getCountActualPositive(int outputNum) {
assertIndex(outputNum);
return underlying[outputNum].getCountActualPositive();
}
/**
* Get the actual negative count (accounting for any masking) for the specified output/column
*
* @param outputNum Index of the class
*/
public long getCountActualNegative(int outputNum) {
assertIndex(outputNum);
return underlying[outputNum].getCountActualNegative();
}
/**
* Merge this ROCMultiClass instance with another.
* This ROCMultiClass instance is modified, by adding the stats from the other instance.
*
* @param other ROCMultiClass instance to combine with this one
*/
@Override
public void merge(ROCMultiClass other) {
if (this.underlying == null) {
this.underlying = other.underlying;
return;
} else if (other.underlying == null) {
return;
}
//Both have data
if (underlying.length != other.underlying.length) {
throw new UnsupportedOperationException("Cannot merge ROCBinary: this expects " + underlying.length
+ "outputs, other expects " + other.underlying.length + " outputs");
}
for (int i = 0; i < underlying.length; i++) {
this.underlying[i].merge(other.underlying[i]);
}
}
public int getNumClasses() {
if (underlying == null) {
return -1;
}
return underlying.length;
}
private void assertIndex(int classIdx) {
if (underlying == null) {
throw new IllegalStateException("Cannot get results: no data has been collected");
}
if (classIdx < 0 || classIdx >= underlying.length) {
throw new IllegalArgumentException("Invalid class index (" + classIdx
+ "): must be in range 0 to numClasses = " + underlying.length);
}
}
public static ROCMultiClass fromJson(String json){
return fromJson(json, ROCMultiClass.class);
}
public double scoreForMetric(Metric metric, int idx){
assertIndex(idx);
switch (metric){
case AUROC:
return calculateAUC(idx);
case AUPRC:
return calculateAUCPR(idx);
default:
throw new IllegalStateException("Unknown metric: " + metric);
}
}
@Override
public double getValue(IMetric metric){
if(metric instanceof Metric){
if(metric == Metric.AUPRC)
return calculateAverageAUCPR();
else if(metric == Metric.AUROC)
return calculateAverageAUC();
else
throw new IllegalStateException("Can't get value for non-ROC Metric " + metric);
} else
throw new IllegalStateException("Can't get value for non-ROC Metric " + metric);
}
@Override
public ROCMultiClass newInstance() {
return new ROCMultiClass(axis, thresholdSteps, rocRemoveRedundantPts, labels);
}
}