神经网络python源码分享

神经网络的逻辑应该都是熟知的了，在这里想说明一下交叉验证

交叉验证方法：

看图大概就能理解了，大致就是先将数据集分成K份，对这K份中每一份都取不一样的比例数据进行训练和测试。得出K个误差，将这K个误差平均得到最终误差

这第一个部分是BP神经网络的建立

参数选取参照论文：基于数据挖掘技术的股价指数分析与预测研究_胡林林

import math
import random
import tushare as ts
import pandas as pd
random.seed(0)
def getData(id,start,end):
 df = ts.get_hist_data(id,start,end)
 DATA=pd.DataFrame(columns=['rate1', 'rate2','rate3','pos1','pos2','pos3','amt1','amt2','amt3','MA20','MA5','r'])
 P1 = pd.DataFrame(columns=['high','low','close','open','volume'])
 DATA2=pd.DataFrame(columns=['R'])
 DATA['MA20']=df['ma20']
 DATA['MA5']=df['ma5']
 P=df['close']
 P1['high']=df['high']
 P1['low']=df['low']
 P1['close']=df['close']
 P1['open']=df['open']
 P1['volume']=df['volume']

DATA['rate1']=(P1['close'].shift(1)-P1['open'].shift(1))/P1['open'].shift(1)
 DATA['rate2']=(P1['close'].shift(2)-P1['open'].shift(2))/P1['open'].shift(2)
 DATA['rate3']=(P1['close'].shift(3)-P1['open'].shift(3))/P1['open'].shift(3)
 DATA['pos1']=(P1['close'].shift(1)-P1['low'].shift(1))/(P1['high'].shift(1)-P1['low'].shift(1))
 DATA['pos2']=(P1['close'].shift(2)-P1['low'].shift(2))/(P1['high'].shift(2)-P1['low'].shift(2))
 DATA['pos3']=(P1['close'].shift(3)-P1['low'].shift(3))/(P1['high'].shift(3)-P1['low'].shift(3))
 DATA['amt1']=P1['volume'].shift(1)/((P1['volume'].shift(1)+P1['volume'].shift(2)+P1['volume'].shift(3))/3)
 DATA['amt2']=P1['volume'].shift(2)/((P1['volume'].shift(2)+P1['volume'].shift(3)+P1['volume'].shift(4))/3)
 DATA['amt3']=P1['volume'].shift(3)/((P1['volume'].shift(3)+P1['volume'].shift(4)+P1['volume'].shift(5))/3)
 templist=(P-P.shift(1))/P.shift(1)
 tempDATA = []
 for indextemp in templist:
   tempDATA.append(1/(1+math.exp(-indextemp*100)))
 DATA['r'] = tempDATA
 DATA=DATA.dropna(axis=0)
 DATA2['R']=DATA['r']
 del DATA['r']
 DATA=DATA.T
 DATA2=DATA2.T
 DATAlist=DATA.to_dict("list")
 result = []
 for key in DATAlist:
   result.append(DATAlist[key])
 DATAlist2=DATA2.to_dict("list")
 result2 = []
 for key in DATAlist2:
   result2.append(DATAlist2[key])
 return result
def getDataR(id,start,end):
 df = ts.get_hist_data(id,start,end)
 DATA=pd.DataFrame(columns=['rate1', 'rate2','rate3','pos1','pos2','pos3','amt1','amt2','amt3','MA20','MA5','r'])
 P1 = pd.DataFrame(columns=['high','low','close','open','volume'])
 DATA2=pd.DataFrame(columns=['R'])
 DATA['MA20']=df['ma20'].shift(1)
 DATA['MA5']=df['ma5'].shift(1)
 P=df['close']
 P1['high']=df['high']
 P1['low']=df['low']
 P1['close']=df['close']
 P1['open']=df['open']
 P1['volume']=df['volume']

DATA['rate1']=(P1['close'].shift(1)-P1['open'].shift(1))/P1['open'].shift(1)
 DATA['rate2']=(P1['close'].shift(2)-P1['open'].shift(2))/P1['open'].shift(2)
 DATA['rate3']=(P1['close'].shift(3)-P1['open'].shift(3))/P1['open'].shift(3)
 DATA['pos1']=(P1['close'].shift(1)-P1['low'].shift(1))/(P1['high'].shift(1)-P1['low'].shift(1))
 DATA['pos2']=(P1['close'].shift(2)-P1['low'].shift(2))/(P1['high'].shift(2)-P1['low'].shift(2))
 DATA['pos3']=(P1['close'].shift(3)-P1['low'].shift(3))/(P1['high'].shift(3)-P1['low'].shift(3))
 DATA['amt1']=P1['volume'].shift(1)/((P1['volume'].shift(1)+P1['volume'].shift(2)+P1['volume'].shift(3))/3)
 DATA['amt2']=P1['volume'].shift(2)/((P1['volume'].shift(2)+P1['volume'].shift(3)+P1['volume'].shift(4))/3)
 DATA['amt3']=P1['volume'].shift(3)/((P1['volume'].shift(3)+P1['volume'].shift(4)+P1['volume'].shift(5))/3)
 templist=(P-P.shift(1))/P.shift(1)
 tempDATA = []
 for indextemp in templist:
   tempDATA.append(1/(1+math.exp(-indextemp*100)))
 DATA['r'] = tempDATA
 DATA=DATA.dropna(axis=0)
 DATA2['R']=DATA['r']
 del DATA['r']
 DATA=DATA.T
 DATA2=DATA2.T
 DATAlist=DATA.to_dict("list")
 result = []
 for key in DATAlist:
   result.append(DATAlist[key])
 DATAlist2=DATA2.to_dict("list")
 result2 = []
 for key in DATAlist2:
   result2.append(DATAlist2[key])
 return result2
def rand(a, b):
 return (b - a) * random.random() + a
def make_matrix(m, n, fill=0.0):
 mat = []
 for i in range(m):
   mat.append([fill] * n)
 return mat
def sigmoid(x):
 return 1.0 / (1.0 + math.exp(-x))
def sigmod_derivate(x):
 return x * (1 - x)
class BPNeuralNetwork:
 def __init__(self):
   self.input_n = 0
   self.hidden_n = 0
   self.output_n = 0
   self.input_cells = []
   self.hidden_cells = []
   self.output_cells = []
   self.input_weights = []
   self.output_weights = []
   self.input_correction = []
   self.output_correction = []

def setup(self, ni, nh, no):
   self.input_n = ni + 1
   self.hidden_n = nh
   self.output_n = no
   # init cells
   self.input_cells = [1.0] * self.input_n
   self.hidden_cells = [1.0] * self.hidden_n
   self.output_cells = [1.0] * self.output_n
   # init weights
   self.input_weights = make_matrix(self.input_n, self.hidden_n)
   self.output_weights = make_matrix(self.hidden_n, self.output_n)
   # random activate
   for i in range(self.input_n):
     for h in range(self.hidden_n):
       self.input_weights[i][h] = rand(-0.2, 0.2)
   for h in range(self.hidden_n):
     for o in range(self.output_n):
       self.output_weights[h][o] = rand(-2.0, 2.0)
   # init correction matrix
   self.input_correction = make_matrix(self.input_n, self.hidden_n)
   self.output_correction = make_matrix(self.hidden_n, self.output_n)

def predict(self, inputs):
   # activate input layer
   for i in range(self.input_n - 1):
     self.input_cells[i] = inputs[i]
   # activate hidden layer
   for j in range(self.hidden_n):
     total = 0.0
     for i in range(self.input_n):
       total += self.input_cells[i] * self.input_weights[i][j]
     self.hidden_cells[j] = sigmoid(total)
   # activate output layer
   for k in range(self.output_n):
     total = 0.0
     for j in range(self.hidden_n):
       total += self.hidden_cells[j] * self.output_weights[j][k]
     self.output_cells[k] = sigmoid(total)
   return self.output_cells[:]
 def back_propagate(self, case, label, learn, correct):
   # feed forward
   self.predict(case)
   # get output layer error
   output_deltas = [0.0] * self.output_n
   for o in range(self.output_n):
     error = label[o] - self.output_cells[o]
     output_deltas[o] = sigmod_derivate(self.output_cells[o]) * error
   # get hidden layer error
   hidden_deltas = [0.0] * self.hidden_n
   for h in range(self.hidden_n):
     error = 0.0
     for o in range(self.output_n):
       error += output_deltas[o] * self.output_weights[h][o]
     hidden_deltas[h] = sigmod_derivate(self.hidden_cells[h]) * error
   # update output weights
   for h in range(self.hidden_n):
     for o in range(self.output_n):
       change = output_deltas[o] * self.hidden_cells[h]
       self.output_weights[h][o] += learn * change + correct * self.output_correction[h][o]
       self.output_correction[h][o] = change
   # update input weights
   for i in range(self.input_n):
     for h in range(self.hidden_n):
       change = hidden_deltas[h] * self.input_cells[i]
       self.input_weights[i][h] += learn * change + correct * self.input_correction[i][h]
       self.input_correction[i][h] = change
   # get global error
   error = 0.0
   for o in range(len(label)):
     error += 0.5 * (label[o] - self.output_cells[o]) ** 2
   return error
 def train(self, cases, labels, limit=10000, learn=0.05, correct=0.1):
   for i in range(limit):
     error = 0.0
     for i in range(len(cases)):
       label = labels[i]
       case = cases[i]
       error += self.back_propagate(case, label, learn, correct)
 def test(self,id):
   result=getData("000001", "2015-01-05", "2015-01-09")
   result2=getDataR("000001", "2015-01-05", "2015-01-09")
   self.setup(11, 5, 1)
   self.train(result, result2, 10000, 0.05, 0.1)
   for t in resulttest:
     print(self.predict(t))

下面是选取14-15年数据进行训练，16年数据作为测试集，调仓周期为20个交易日，大约1个月，对上证50中的股票进行预测，选取预测的涨幅前10的股票买入，对每只股票分配一样的资金,初步运行没有问题，但就是太慢了，等哪天有空了再运行

import BPnet
import tushare as ts
import pandas as pd
import math
import xlrd
import datetime as dt
import time

#
#nn =BPnet.BPNeuralNetwork()
#nn.test('000001')
#for i in ts.get_sz50s()['code']:
holdList=pd.DataFrame(columns=['time','id','value'])
share=ts.get_sz50s()['code']
time2=ts.get_k_data('000001')['date']
newtime = time2[400:640]
newcount=0
for itime in newtime:
 print(itime)
 if newcount ％ 20 == 0:
       sharelist = pd.DataFrame(columns=['time','id','value'])
   for ishare in share:
     backwardtime = time.strftime('％Y-％m-％d',time.localtime(time.mktime(time.strptime(itime,'％Y-％m-％d'))-432000*4))
     trainData = BPnet.getData(ishare, '2014-05-22',itime)
     trainDataR = BPnet.getDataR(ishare, '2014-05-22',itime)
     testData = BPnet.getData(ishare, backwardtime,itime)
     try:
       print(testData)
       testData = testData[-1]
       print(testData)
       nn = BPnet.BPNeuralNetwork()
       nn.setup(11, 5, 1)
       nn.train(trainData, trainDataR, 10000, 0.05, 0.1)
       value = nn.predict(testData)
       newlist= pd.DataFrame({'time':itime,"id":ishare,"value":value},index=["0"])
       sharelist = sharelist.append(newlist,ignore_index=True)
     except:
       pass
   sharelist=sharelist.sort(columns ='value',ascending=False)
   sharelist = sharelist[:10]
   holdList=holdList.append(sharelist,ignore_index=True)
 newcount+=1
 print(holdList)

总结

神经网络理论基础及Python实现详解

Python语言实现百度语音识别API的使用实例

Python通过matplotlib绘制动画简单实例

如有不足之处，欢迎留言指出。感谢朋友们对本站的支持！

来源：https://www.cnblogs.com/yunerlalala/p/6240722.html