1 /*
2 * Licensed to the Apache Software Foundation (ASF) under one or more
3 * contributor license agreements. See the NOTICE file distributed with
4 * this work for additional information regarding copyright ownership.
5 * The ASF licenses this file to You under the Apache License, Version 2.0
6 * (the "License"); you may not use this file except in compliance with
7 * the License. You may obtain a copy of the License at
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 */
17
18 package org.apache.commons.math4.neuralnet.sofm.util;
19
20 import java.util.function.DoubleUnaryOperator;
21 import java.util.function.LongToDoubleFunction;
22
23 import org.apache.commons.math4.neuralnet.internal.NeuralNetException;
24
25 /**
26 * Decay function whose shape is similar to a sigmoid.
27 * <br>
28 * Class is immutable.
29 *
30 * @since 3.3
31 */
32 public class QuasiSigmoidDecayFunction implements LongToDoubleFunction {
33 /** Sigmoid. */
34 private final DoubleUnaryOperator sigmoid;
35 /** See {@link #value(long)}. */
36 private final double scale;
37
38 /**
39 * Creates an instance.
40 * The function {@code f} will have the following properties:
41 * <ul>
42 * <li>{@code f(0) = initValue}</li>
43 * <li>{@code numCall} is the inflexion point</li>
44 * <li>{@code slope = f'(numCall)}</li>
45 * </ul>
46 *
47 * @param initValue Initial value, i.e. {@link #applyAsDouble(long) applyAsDouble(0)}.
48 * @param slope Value of the function derivative at {@code numCall}.
49 * @param numCall Inflexion point.
50 * @throws IllegalArgumentException if {@code initValue <= 0},
51 * {@code slope >= 0} or {@code numCall <= 0}.
52 */
53 public QuasiSigmoidDecayFunction(double initValue,
54 double slope,
55 long numCall) {
56 if (initValue <= 0) {
57 throw new NeuralNetException(NeuralNetException.NOT_STRICTLY_POSITIVE, initValue);
58 }
59 if (slope >= 0) {
60 throw new NeuralNetException(NeuralNetException.TOO_LARGE, slope, 0);
61 }
62 if (numCall <= 1) {
63 throw new NeuralNetException(NeuralNetException.TOO_SMALL, numCall, 1);
64 }
65
66 final double k = initValue;
67 final double m = numCall;
68 final double b = 4 * slope / initValue;
69 sigmoid = x -> k / (1 + Math.exp(b * (m - x)));
70
71 final double y0 = sigmoid.applyAsDouble(0d);
72 scale = k / y0;
73 }
74
75 /**
76 * Computes the value of the learning factor.
77 *
78 * @param numCall Current step of the training task.
79 * @return the value of the function at {@code numCall}.
80 */
81 @Override
82 public double applyAsDouble(long numCall) {
83 return scale * sigmoid.applyAsDouble((double) numCall);
84 }
85 }