ai.djl.modality.Classifications Maven / Gradle / Ivy
The newest version!
/*
* Copyright 2019 Amazon.com, Inc. or its affiliates. All Rights Reserved.
*
* Licensed under the Apache License, Version 2.0 (the "License"). You may not use this file except in compliance
* with the License. A copy of the License is located at
*
* http://aws.amazon.com/apache2.0/
*
* or in the "license" file accompanying this file. This file 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.
*/
package ai.djl.modality;
import ai.djl.ndarray.NDArray;
import ai.djl.ndarray.types.DataType;
import ai.djl.translate.Ensembleable;
import ai.djl.util.JsonSerializable;
import ai.djl.util.JsonUtils;
import com.google.gson.Gson;
import com.google.gson.JsonElement;
import com.google.gson.JsonSerializationContext;
import com.google.gson.JsonSerializer;
import java.lang.reflect.Type;
import java.nio.ByteBuffer;
import java.nio.charset.StandardCharsets;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.Collections;
import java.util.Comparator;
import java.util.Iterator;
import java.util.List;
import java.util.stream.Collectors;
/**
* {@code Classifications} is the container that stores the classification results for
* classification on a single input.
*/
public class Classifications implements JsonSerializable, Ensembleable {
private static final long serialVersionUID = 1L;
private static final Gson GSON =
JsonUtils.builder()
.registerTypeAdapter(Classifications.class, new ClassificationsSerializer())
.create();
@SuppressWarnings("serial")
protected List classNames;
@SuppressWarnings("serial")
protected List probabilities;
protected int topK;
/**
* Constructs a {@code Classifications} using a parallel list of classNames and probabilities.
*
* @param classNames the names of the classes
* @param probabilities the probabilities for each class for the input
*/
public Classifications(List classNames, List probabilities) {
this.classNames = classNames;
this.probabilities = probabilities;
this.topK = 5;
}
/**
* Constructs a {@code Classifications} using list of classNames parallel to an NDArray of
* probabilities.
*
* @param classNames the names of the classes
* @param probabilities the probabilities for each class for the input
*/
public Classifications(List classNames, NDArray probabilities) {
this(classNames, probabilities, 5);
}
/**
* Constructs a {@code Classifications} using list of classNames parallel to an NDArray of
* probabilities.
*
* @param classNames the names of the classes
* @param probabilities the probabilities for each class for the input
* @param topK the number of top classes to return
*/
public Classifications(List classNames, NDArray probabilities, int topK) {
this.classNames = classNames;
if (probabilities.getDataType() == DataType.FLOAT32) {
// Avoid converting float32 to float64 as this is not supported on MPS device
this.probabilities = new ArrayList<>();
for (float prob : probabilities.toFloatArray()) {
this.probabilities.add((double) prob);
}
} else {
NDArray array = probabilities.toType(DataType.FLOAT64, false);
this.probabilities =
Arrays.stream(array.toDoubleArray()).boxed().collect(Collectors.toList());
array.close();
}
this.topK = topK;
}
/**
* Returns the classes that were classified into.
*
* @return the classes that were classified into
*/
public List getClassNames() {
return classNames;
}
/**
* Returns the list of probabilities for each class (matching the order of the class names).
*
* @return the list of probabilities for each class (matching the order of the class names)
*/
public List getProbabilities() {
return probabilities;
}
/**
* Set the topK number of classes to be displayed.
*
* @param topK the number of top classes to return
*/
public final void setTopK(int topK) {
this.topK = topK;
}
/**
* Returns a classification item for each potential class for the input.
*
* @param the type of classification item for the task
* @return the list of classification items
*/
public List items() {
List list = new ArrayList<>(classNames.size());
for (int i = 0; i < classNames.size(); i++) {
list.add(item(i));
}
return list;
}
/**
* Returns the item at a given index based on the order used to construct the {@link
* Classifications}.
*
* @param index the index of the item to return
* @param the type of classification item for the task
* @return the item at the given index, equivalent to {@code classifications.items().get(index)}
*/
@SuppressWarnings("unchecked")
public T item(int index) {
return (T) new Classification(classNames.get(index), probabilities.get(index));
}
/**
* Returns a list of the top classes.
*
* @param the type of the classification item for the task
* @return the list of classification items for the best classes in order of best to worst
*/
public List topK() {
return topK(topK);
}
/**
* Returns a list of the top {@code k} best classes.
*
* @param k the number of classes to return
* @param the type of the classification item for the task
* @return the list of classification items for the best classes in order of best to worst
*/
public List topK(int k) {
List items = items();
items.sort(Comparator.comparingDouble(Classification::getProbability).reversed());
int count = Math.min(items.size(), k);
return items.subList(0, count);
}
/**
* Returns the most likely class for the classification.
*
* @param the type of the classification item for the task
* @return the classification item
*/
public T best() {
return item(probabilities.indexOf(Collections.max(probabilities)));
}
/**
* Returns the result for a particular class name.
*
* @param className the class name to get results for
* @param the type of the classification item for the task
* @return the (first if multiple) classification item
*/
public T get(String className) {
int size = classNames.size();
for (int i = 0; i < size; i++) {
if (classNames.get(i).equals(className)) {
return item(i);
}
}
return null;
}
/** {@inheritDoc} */
@Override
public String toJson() {
return GSON.toJson(this) + '\n';
}
/** {@inheritDoc} */
@Override
public String getAsString() {
return toJson();
}
/** {@inheritDoc} */
@Override
public ByteBuffer toByteBuffer() {
return ByteBuffer.wrap(toJson().getBytes(StandardCharsets.UTF_8));
}
/** {@inheritDoc} */
@Override
public String toString() {
StringBuilder sb = new StringBuilder();
sb.append('[').append(System.lineSeparator());
for (Classification item : topK(topK)) {
sb.append('\t').append(item).append(System.lineSeparator());
}
sb.append(']');
return sb.toString();
}
/** {@inheritDoc} */
@Override
public Classifications ensembleWith(Iterator it) {
int size = probabilities.size();
List newProbabilities = new ArrayList<>(size);
newProbabilities.addAll(probabilities);
int count = 1;
while (it.hasNext()) {
++count;
Classifications c = it.next();
for (int i = 0; i < size; ++i) {
newProbabilities.set(i, newProbabilities.get(i) + c.probabilities.get(i));
}
if (!c.classNames.equals(classNames)) {
throw new IllegalArgumentException(
"Found a classNames mismatch during ensembling. All input Classifications"
+ " should have the same classNames, but some were different");
}
}
final int total = count;
newProbabilities.replaceAll(p -> p / total);
return new Classifications(classNames, newProbabilities);
}
/**
* A {@code Classification} stores the classification result for a single class on a single
* input.
*/
public static class Classification {
private String className;
private double probability;
/**
* Constructs a single class result for a classification.
*
* @param className the class name of the result
* @param probability the probability of the result
*/
public Classification(String className, double probability) {
this.className = className;
this.probability = probability;
}
/**
* Returns the class name.
*
* @return the class name
*/
public String getClassName() {
return className;
}
/**
* Returns the probability.
*
* Probability explains how accurately the classifier identified the target class.
*
* @return the probability
*/
public double getProbability() {
return probability;
}
/** {@inheritDoc} */
@Override
public String toString() {
StringBuilder sb = new StringBuilder(100);
sb.append("{\"class\": \"").append(className).append("\", \"probability\": ");
if (probability < 0.00001) {
sb.append(String.format("%.1e", probability));
} else {
probability = (int) (probability * 100000) / 100000f;
sb.append(String.format("%.5f", probability));
}
sb.append('}');
return sb.toString();
}
}
/** A customized Gson serializer to serialize the {@code Classifications} object. */
public static final class ClassificationsSerializer implements JsonSerializer {
/** {@inheritDoc} */
@Override
public JsonElement serialize(Classifications src, Type type, JsonSerializationContext ctx) {
List list = src.topK();
return ctx.serialize(list);
}
}
}