diff --git a/src/main/java/org/dromara/easyai/rnnJumpNerveCenter/CustomManager.java b/src/main/java/org/dromara/easyai/rnnJumpNerveCenter/CustomManager.java new file mode 100644 index 0000000..bcb8675 --- /dev/null +++ b/src/main/java/org/dromara/easyai/rnnJumpNerveCenter/CustomManager.java @@ -0,0 +1,207 @@ +package org.dromara.easyai.rnnJumpNerveCenter; + + +import org.dromara.easyai.matrixTools.Matrix; +import org.dromara.easyai.config.SentenceConfig; +import org.dromara.easyai.entity.CreatorModel; +import org.dromara.easyai.entity.SemanticsBack; +import org.dromara.easyai.entity.TalkBody; +import org.dromara.easyai.function.Tanh; +import org.dromara.easyai.naturalLanguage.word.WordEmbedding; +import org.dromara.easyai.rnnJumpNerveEntity.MyWordFeature; +import org.dromara.easyai.rnnJumpNerveEntity.NerveCenter; +import org.dromara.easyai.rnnJumpNerveEntity.SensoryNerve; + +import java.util.*; + +@Deprecated() +public class CustomManager {// + private final WordEmbedding wordEmbedding; + private NerveJumpManager semanticsManager;//网络 + private final int vectorDimension;//特征纵向维度 + private final int maxFeatureLength;//特征最长长度 + private final float studyPoint;//词向量学习学习率 + private final int minLength;//最小长度 + private final int answerMaxLength;//回答最长长度 + private int times; + private final float param; + private final float powerTh; + private final int rzModel; + + public CustomManager(WordEmbedding wordEmbedding, SentenceConfig config) { + this.minLength = config.getMinLength(); + this.wordEmbedding = wordEmbedding; + this.vectorDimension = config.getQaWordVectorDimension(); + this.maxFeatureLength = config.getMaxWordLength(); + this.studyPoint = config.getWeStudyPoint(); + this.answerMaxLength = config.getMaxAnswerLength(); + this.powerTh = config.getSentenceTrustPowerTh(); + this.times = config.getTimes(); + this.param = config.getParam(); + this.rzModel = config.getRzModel(); + if (this.times < 1) { + this.times = 1; + } + } + + public void init() throws Exception { + semanticsManager = new NerveJumpManager(vectorDimension, vectorDimension, wordEmbedding.getWordList().size(), maxFeatureLength + answerMaxLength - 1, new Tanh(), false, + studyPoint, rzModel, param); + semanticsManager.setPowerTh(powerTh); + semanticsManager.initRnn(true, true, true, true, maxFeatureLength); + List nerveCenterList = semanticsManager.getNerveCenterList(); + for (NerveCenter nerveCenter : nerveCenterList) { + nerveCenter.setWordEmbedding(wordEmbedding); + } + } + + public void insertModel(CreatorModel creatorModel) throws Exception { + semanticsManager.insertModelParameter(creatorModel.getSemanticsModel()); + } + + public String getAnswer(String question, long eventID) throws Exception { + SemanticsBack semanticsBack = new SemanticsBack(); + if (question.length() > maxFeatureLength) { + question = question.substring(0, maxFeatureLength); + } + MyWordFeature myWordFeature = wordEmbedding.getEmbedding(question, eventID, false); + List featureList = myWordFeature.getFirstFeatureList(); + Matrix featureMatrix = myWordFeature.getFeatureMatrix(); + featureMatrix = insertZero(featureMatrix, featureMatrix.getX()); + int[] storeys = getStoreys2(question.length(), maxFeatureLength); + studySemanticsNerve(eventID, featureList, false, null, semanticsBack, featureMatrix, storeys, question.length()); + return semanticsBack.getWord(); + } + + public CreatorModel study(List talkBodies) throws Exception { + Random random = new Random(); + CreatorModel creatorModel = new CreatorModel(); + int maxTime = maxFeatureLength * answerMaxLength * times; + int wordSize = talkBodies.size() * maxTime; + int index = 0; + for (int i = 0; i < maxTime; i++) {//第一阶段学习 + System.out.println("生成模型学习完成次数：" + (i + 1) + "总共次数:" + maxTime); + for (TalkBody talkBody : talkBodies) { + index++; + String question = talkBody.getQuestion(); + String answer = talkBody.getAnswer(); + if (question.length() > maxFeatureLength) { + question = question.substring(0, maxFeatureLength); + } + if (answer.length() > answerMaxLength) { + answer = answer.substring(0, answerMaxLength); + } + String word = question + answer; + MyWordFeature myWordFeature = wordEmbedding.getEmbedding(word, 1, false); + semanticsStudy(myWordFeature, question, answer, 1, random); + float point = (float) index / (float) (wordSize) * 100; + String result = String.format("%.6f", point); + System.out.println("训练进度：" + result + "%"); + } + } + creatorModel.setSemanticsModel(semanticsManager.getModelParameter()); + return creatorModel; + } + + private int[] getStoreys2(int len, int finishIndex) { + int[] storeys = new int[len + 1]; + for (int i = 0; i < storeys.length; i++) { + if (i < storeys.length - 1) { + storeys[i] = i; + } else { + storeys[i] = finishIndex; + } + } + return storeys; + } + + private int[] getStoreys(int len, Random random, int startIndex) { + int[] storeys; + if (len < minLength) { + storeys = new int[len]; + for (int i = 0; i < len; i++) { + storeys[i] = i + startIndex; + } + } else { + List list = new ArrayList<>(); + for (int i = 1; i < len; i++) { + list.add(i); + } + int myLen = (int) (minLength + (float) Math.random() * (len - minLength + 1)); + storeys = new int[myLen]; + if (startIndex > 0) { + storeys[0] = startIndex; + } + for (int i = 1; i < myLen; i++) { + int index = random.nextInt(list.size()); + storeys[i] = list.get(index) + startIndex; + list.remove(index); + } + Arrays.sort(storeys); + } + return storeys; + } + + private int[] pushArray(int[] arr, int value) { + int[] myArr = new int[arr.length + 1]; + for (int i = 0; i < myArr.length; i++) { + if (i < myArr.length - 1) { + myArr[i] = arr[i]; + } else { + myArr[i] = value; + } + } + return myArr; + } + + + private Matrix insertZero(Matrix feature, int index) throws Exception { + Matrix matrix = new Matrix(feature.getX() + 1, feature.getY()); + int x = matrix.getX(); + int y = matrix.getY(); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + if (i < index) { + matrix.setNub(i, j, feature.getNumber(i, j)); + } else if (i > index) { + matrix.setNub(i, j, feature.getNumber(i - 1, j)); + } + } + } + return matrix; + } + + private void semanticsStudy(MyWordFeature myWordFeature, String question, String answer, long eventID, Random random) throws Exception { + Matrix featureMatrix = insertZero(myWordFeature.getFeatureMatrix(), question.length()); + List firstFeatureList = myWordFeature.getFirstFeatureList(); + System.out.println("训练question:" + question + ",answer:" + answer); + if (question.length() > 1 && !answer.isEmpty()) { + int[] questionStoreys = getStoreys(question.length(), random, 0); + int[] answerStoreys = getStoreys(answer.length() + 1, random, maxFeatureLength); + Map E = new HashMap<>(); + for (int i = 0; i < answerStoreys.length - 1; i++) { + //System.out.println("====================================="); + E.clear(); + int index = answerStoreys[i + 1] - maxFeatureLength; + questionStoreys = pushArray(questionStoreys, answerStoreys[i]); + String myAnswer = answer.substring(index - 1, index); + int wordID = wordEmbedding.getID(myAnswer) + 1; + E.put(wordID, 1f); + studySemanticsNerve(eventID, firstFeatureList, true, E, null, featureMatrix, questionStoreys, question.length()); + } + } + } + + private void studySemanticsNerve(long eventId, List featureList, boolean isStudy, Map E, SemanticsBack semanticsBack, + Matrix rnnMatrix, int[] storeys, int questionLength) throws Exception { + List sensoryNerves = semanticsManager.getSensoryNerves(); + if (sensoryNerves.size() == featureList.size()) { + for (int i = 0; i < sensoryNerves.size(); i++) { + sensoryNerves.get(i).postMessage(eventId, featureList.get(i), isStudy, E, semanticsBack, rnnMatrix, storeys, questionLength); + } + } else { + throw new Exception("1size not equals,feature size:" + featureList.size() + "," + + "sensorySize:" + sensoryNerves.size()); + } + } +}