毕业设计：基于python，RNN循环神经网络爬虫获取数据的价格预测系统.zip

package com.sheue.ml.rnn.lstm; import com.sheue.ml.layers.MatIniter; import com.sheue.ml.utils.Activer; import org.jblas.DoubleMatrix; import org.jblas.MatrixFunctions; import java.io.Serializable; import java.util.Map; public class LSTM implements Serializable { private static final long serialVersionUID = -7059290852389115565L; private int inSize; private int outSize; private int deSize; private DoubleMatrix Wxi; private DoubleMatrix Whi; private DoubleMatrix Wci; private DoubleMatrix bi; private DoubleMatrix Wxf; private DoubleMatrix Whf; private DoubleMatrix Wcf; private DoubleMatrix bf; private DoubleMatrix Wxc; private DoubleMatrix Whc; private DoubleMatrix bc; private DoubleMatrix Wxo; private DoubleMatrix Who; private DoubleMatrix Wco; private DoubleMatrix bo; private DoubleMatrix Why; private DoubleMatrix by; public LSTM(int inSize, int outSize, MatIniter initer) { this.inSize = inSize; this.outSize = outSize; if (initer.getType() == MatIniter.Type.Uniform) { this.Wxi = initer.uniform(inSize, outSize); this.Whi = initer.uniform(outSize, outSize); this.Wci = initer.uniform(outSize, outSize); this.bi = new DoubleMatrix(1, outSize); this.Wxf = initer.uniform(inSize, outSize); this.Whf = initer.uniform(outSize, outSize); this.Wcf = initer.uniform(outSize, outSize); this.bf = new DoubleMatrix(1, outSize); this.Wxc = initer.uniform(inSize, outSize); this.Whc = initer.uniform(outSize, outSize); this.bc = new DoubleMatrix(1, outSize); this.Wxo = initer.uniform(inSize, outSize); this.Who = initer.uniform(outSize, outSize); this.Wco = initer.uniform(outSize, outSize); this.bo = new DoubleMatrix(1, outSize); this.Why = initer.uniform(outSize, inSize); this.by = new DoubleMatrix(1, inSize); } else if (initer.getType() == MatIniter.Type.Gaussian) { this.Wxi = initer.gaussian(inSize, outSize); this.Whi = initer.gaussian(outSize, outSize); this.Wci = initer.gaussian(outSize, outSize); this.bi = new DoubleMatrix(1, outSize); this.Wxf = initer.gaussian(inSize, outSize); this.Whf = initer.gaussian(outSize, outSize); this.Wcf = initer.gaussian(outSize, outSize); this.bf = new DoubleMatrix(1, outSize); this.Wxc = initer.gaussian(inSize, outSize); this.Whc = initer.gaussian(outSize, outSize); this.bc = new DoubleMatrix(1, outSize); this.Wxo = initer.gaussian(inSize, outSize); this.Who = initer.gaussian(outSize, outSize); this.Wco = initer.gaussian(outSize, outSize); this.bo = new DoubleMatrix(1, outSize); this.Why = initer.gaussian(outSize, inSize); this.by = new DoubleMatrix(1, inSize); } } public LSTM(int inSize, int outSize, MatIniter initer, int deSize) { this(inSize, outSize, initer); this.deSize = deSize; this.Why = new DoubleMatrix(outSize, deSize); this.by = new DoubleMatrix(1, deSize); } public int getInSize() { return inSize; } private int getOutSize() { return outSize; } public int getDeSize() { return deSize; } public void active(int t, Map<String, DoubleMatrix> acts) { DoubleMatrix x = acts.get("x" + t); DoubleMatrix preH = null, preC = null; if (t == 0) { preH = new DoubleMatrix(1, getOutSize()); preC = preH.dup(); } else { preH = acts.get("h" + (t - 1)); preC = acts.get("c" + (t - 1)); } DoubleMatrix i = Activer.logistic(x.mmul(Wxi).add(preH.mmul(Whi)).add(preC.mmul(Wci)).add(bi)); DoubleMatrix f = Activer.logistic(x.mmul(Wxf).add(preH.mmul(Whf)).add(preC.mmul(Wcf)).add(bf)); DoubleMatrix gc = Activer.tanh(x.mmul(Wxc).add(preH.mmul(Whc)).add(bc)); DoubleMatrix c = f.mul(preC).add(i.mul(gc)); DoubleMatrix o = Activer.logistic(x.mmul(Wxo).add(preH.mmul(Who)).add(c.mmul(Wco)).add(bo)); DoubleMatrix gh = Activer.tanh(c); DoubleMatrix h = o.mul(gh); acts.put("i" + t, i); acts.put("f" + t, f); acts.put("gc" + t, gc); acts.put("c" + t, c); acts.put("o" + t, o); acts.put("gh" + t, gh); acts.put("h" + t, h); } public void bptt(Map<String, DoubleMatrix> acts, int lastT, double lr) { for (int t = lastT; t > -1; t--) { DoubleMatrix py = acts.get("py" + t); DoubleMatrix y = acts.get("y" + t); DoubleMatrix deltaY = py.sub(y); acts.put("dy" + t, deltaY); // cell output errors DoubleMatrix h = acts.get("h" + t); DoubleMatrix deltaH = null; if (t == lastT) { deltaH = Why.mmul(deltaY.transpose()).transpose(); } else { DoubleMatrix lateDgc = acts.get("dgc" + (t + 1)); DoubleMatrix lateDf = acts.get("df" + (t + 1)); DoubleMatrix lateDo = acts.get("do" + (t + 1)); DoubleMatrix lateDi = acts.get("di" + (t + 1)); deltaH = Why.mmul(deltaY.transpose()).transpose() .add(Whc.mmul(lateDgc.transpose()).transpose()) .add(Whi.mmul(lateDi.transpose()).transpose()) .add(Who.mmul(lateDo.transpose()).transpose()) .add(Whf.mmul(lateDf.transpose()).transpose()); } acts.put("dh" + t, deltaH); // output gates DoubleMatrix gh = acts.get("gh" + t); DoubleMatrix o = acts.get("o" + t); DoubleMatrix deltaO = deltaH.mul(gh).mul(deriveExp(o)); acts.put("do" + t, deltaO); // status DoubleMatrix deltaC = null; if (t == lastT) { deltaC = deltaH.mul(o).mul(deriveTanh(gh)) .add(Wco.mmul(deltaO.transpose()).transpose()); } else { DoubleMatrix lateDc = acts.get("dc" + (t + 1)); DoubleMatrix lateDf = acts.get("df" + (t + 1)); DoubleMatrix lateF = acts.get("f" + (t + 1)); DoubleMatrix lateDi = acts.get("di" + (t + 1)); deltaC = deltaH.mul(o).mul(deriveTanh(gh)) .add(Wco.mmul(deltaO.transpose()).transpose()) .add(lateF.mul(lateDc)) .add(Wcf.mmul(lateDf.transpose()).transpose()) .add(Wci.mmul(lateDi.transpose()).transpose()); } acts.put("dc" + t, deltaC); // cells DoubleMatrix gc = acts.get("gc" + t); DoubleMatrix i = acts.get("i" + t); DoubleMatrix deltaGc = deltaC.mul(i).mul(deriveTanh(gc)); acts.put("dgc" + t, deltaGc); DoubleMatrix preC = null; if (t > 0) { preC = acts.get("c" + (t - 1)); } else { preC = DoubleMatrix.zeros(1, h.length); } // forget gates DoubleMatrix f = acts.get("f" + t); DoubleMatrix deltaF = deltaC.mul(preC).mul(deriveExp(f)); acts.put("df" + t, deltaF); // input gates DoubleMatrix deltaI = deltaC.mul(gc).mul(deriveExp(i)); acts.put("di" + t, deltaI); } updateParameters(acts, lastT, lr); } private void updateParameters(Map<String, DoubleMatrix> acts, int lastT, double lr) { DoubleMatrix gWxi = new DoubleMatrix(Wxi.rows, Wxi.columns); DoubleMatrix gWhi = new DoubleMatrix(Whi.rows, Whi.columns);