Add File

src/main/java/org/dromara/easyai/transFormer/nerve/SoftMax.java

@@ -0,0 +1,147 @@
+package org.dromara.easyai.transFormer.nerve;
+
+
+import org.dromara.easyai.i.OutBack;
+import org.dromara.easyai.matrixTools.Matrix;
+import org.dromara.easyai.matrixTools.MatrixOperation;
+
+import java.util.ArrayList;
+import java.util.List;
+
+public class SoftMax extends Nerve {
+    private final List<OutNerve> outNerves;
+    private final boolean isShowLog;
+    private final MatrixOperation matrixOperation = new MatrixOperation();
+    private final float timePunValue;//时间惩罚系数
+
+    public SoftMax(List<OutNerve> outNerves, boolean isShowLog
+            , int sensoryNerveNub, int hiddenNerveNub, int outNerveNub, float timePunValue) throws Exception {
+        super(0, "softMax", 0, null, sensoryNerveNub, hiddenNerveNub, outNerveNub,
+                null, 0, 0, 1);
+        this.timePunValue = timePunValue;
+        this.outNerves = outNerves;
+        this.isShowLog = isShowLog;
+    }
+
+    private Matrix getSigmaByColum(Matrix matrix) throws Exception {
+        int x = matrix.getX();
+        int y = matrix.getY();
+        Matrix myMatrix = new Matrix(1, y);
+        float pun = 1;
+        for (int i = x - 1; i >= 0; i--) {
+            for (int j = 0; j < y; j++) {
+                float value = myMatrix.getNumber(0, j) + matrix.getNumber(i, j) * pun;
+                myMatrix.setNub(0, j, value);
+            }
+            pun = pun * timePunValue;
+        }
+        return myMatrix;
+    }
+
+    @Override
+    protected void toOut(long eventId, Matrix parameter, boolean isStudy, OutBack outBack, List<Integer> E, boolean outAllPro) throws Exception {
+        boolean allReady = insertMatrixParameter(eventId, parameter);
+        if (allReady) {
+            Matrix feature = reMatrixFeatures.get(eventId).getMatrix();//特征
+            reMatrixFeatures.remove(eventId);
+            int x = feature.getX();
+            if (isStudy) {
+                // System.out.println("分类数量:" + feature.getY());
+                if (E.size() != x) {
+                    throw new Exception("期望的序列长度与实际序列不相等！请检查期望E，补充漏掉的序列");
+                }
+                Matrix allError = null;
+                for (int i = 0; i < x; i++) {
+                    //Matrix row = feature.getRow(i);
+                    Matrix r = feature.getSonOfMatrix(0, 0, i + 1, feature.getY());
+                    Matrix row = getSigmaByColum(r);
+                    Mes mes = softMax(true, row, false);//输出值
+                    int key = E.get(i);
+                    if (isShowLog) {
+                        System.out.println("softMax==" + key + ",out==" + mes.poi + ",nerveId==" + mes.typeID);
+                    }
+                    Matrix errors = error(mes, key, i, x);
+                    if (i == 0) {
+                        allError = errors;
+                    } else {
+                        allError = matrixOperation.add(allError, errors);
+                        //allError = matrixOperation.pushVector(allError, errors, true);
+                    }
+                }
+                int size = outNerves.size();
+                for (int i = 0; i < size; i++) {
+                    Matrix errorMatrix = allError.getColumn(i);
+                    outNerves.get(i).getGBySoftMax(errorMatrix, eventId);
+                }
+            } else {
+                if (outBack != null) {
+                    Matrix row = getSigmaByColum(feature);
+                    Mes mes = softMax(false, row, outAllPro);//输出值
+                    outBack.getBack(mes.poi, mes.typeID, eventId);
+                    outBack.getBackMatrix(row, 1, eventId);
+                    if (outAllPro) {
+                        outBack.getSoftMaxBack(eventId, mes.softMax);
+                    }
+                } else {
+                    throw new Exception("not find outBack");
+                }
+            }
+        }
+    }
+
+    private Matrix error(Mes mes, int key, int featureIndex, int allSize) throws Exception {
+        int t = key - 1;
+        List<Float> softMax = mes.softMax;
+        Matrix matrix = new Matrix(allSize, softMax.size());
+        for (int i = 0; i < softMax.size(); i++) {
+            float self = softMax.get(i);
+            float myError;
+            if (i != t) {
+                myError = -self;
+            } else {
+                myError = 1 - self;
+            }
+            float pun = 1;
+            for (int j = featureIndex; j >= 0; j--) {
+                float error = myError * pun;
+                matrix.setNub(j, i, error);
+                pun = pun * timePunValue;
+            }
+        }
+        return matrix;
+    }
+
+    private Mes softMax(boolean isStudy, Matrix matrix, boolean outAllPro) throws Exception {//计算当前输出结果
+        float sigma = 0;
+        int id = 0;
+        float poi = 0;
+        Mes mes = new Mes();
+        int size = matrix.getY();
+        for (int j = 0; j < size; j++) {
+            float value = matrix.getNumber(0, j);
+            sigma = (float) Math.exp(value) + sigma;
+        }
+        List<Float> softMax = new ArrayList<>();
+        for (int i = 0; i < size; i++) {
+            float eSelf = (float) Math.exp(matrix.getNumber(0, i));
+            float value = eSelf / sigma;
+            if (isStudy || outAllPro) {
+                softMax.add(value);
+            }
+            if (value > poi) {
+                poi = value;
+                id = i + 1;
+            }
+        }
+        mes.softMax = softMax;
+        mes.typeID = id;
+        mes.poi = poi;
+        return mes;
+    }
+
+    static class Mes {
+        int typeID;
+        float poi;
+        List<Float> softMax;
+    }
+}