-
Notifications
You must be signed in to change notification settings - Fork 1.8k
/
WeightedLossFunctionExample.java
82 lines (70 loc) · 3.53 KB
/
WeightedLossFunctionExample.java
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
/*******************************************************************************
*
*
*
* 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.
* See the NOTICE file distributed with this work for additional
* information regarding copyright ownership.
*
* 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.deeplearning4j.examples.quickstart.features.classimbalance;
import org.deeplearning4j.nn.conf.MultiLayerConfiguration;
import org.deeplearning4j.nn.conf.NeuralNetConfiguration;
import org.deeplearning4j.nn.conf.layers.DenseLayer;
import org.deeplearning4j.nn.conf.layers.OutputLayer;
import org.deeplearning4j.nn.multilayer.MultiLayerNetwork;
import org.deeplearning4j.nn.weights.WeightInit;
import org.nd4j.linalg.activations.Activation;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.factory.Nd4j;
import org.nd4j.linalg.learning.config.Sgd;
import org.nd4j.linalg.lossfunctions.impl.LossMCXENT;
/**
* @author Alex Black
*/
public class WeightedLossFunctionExample {
public static void main(String[] args) {
/*
Idea with a weighted loss function: it allows us to add a weight to the outputs.
For example, if we have 3 classes, and we consider predictions of the 3rd class to be more important, we might use
a weight array of [0.5,0.5,1.0]. This means that the first 2 outputs will contribute only half as much as they
normally would to the loss/score.
Note that the weights don't (and shouldn't necessarily) sum to 1.0 - and that a weight array of all 1s is equivalent
to having no weight array at all.
If the use case is dealing with class imbalance for classification, use smaller weights for frequently occurring
classes, and 1.0 or larger weights for infrequently occurring classes.
Training and the data pipelines when using weighted loss functions are identical to not using them, so this example
shows only how to configure the weighting.
*/
int numInputs = 4;
int numClasses = 3; //3 classes for classification
//Create the weights array. Note that we have 3 output classes, therefore we have 3 weights
INDArray weightsArray = Nd4j.create(new double[]{0.5, 0.5, 1.0});
MultiLayerConfiguration conf = new NeuralNetConfiguration.Builder()
.activation(Activation.RELU)
.weightInit(WeightInit.XAVIER)
.updater(new Sgd(0.1))
.list()
.layer(new DenseLayer.Builder().nIn(numInputs).nOut(5)
.build())
.layer(new DenseLayer.Builder().nIn(5).nOut(5)
.build())
.layer(new OutputLayer.Builder()
.lossFunction(new LossMCXENT(weightsArray)) // *** Weighted loss function configured here ***
.activation(Activation.SOFTMAX)
.nIn(5).nOut(numClasses).build())
.build();
//Initialize and use the model as before
MultiLayerNetwork model = new MultiLayerNetwork(conf);
model.init();
}
}