【转】基于keras 的神经网络股价预测模型

  1 from matplotlib.dates import DateFormatter, WeekdayLocator, DayLocator, MONDAY,YEARLY
  2 from matplotlib.finance import quotes_historical_yahoo_ohlc, candlestick_ohlc
  3 #import matplotlib
  4 import tushare as ts
  5 import pandas as pd
  6 import matplotlib.pyplot as plt
  7 from matplotlib.pylab import date2num
  8 import datetime
  9 import numpy as np
 10 from pandas import DataFrame
 11 from numpy import row_stack,column_stack
 12
 13 df=ts.get_hist_data(‘601857‘,start=‘2016-06-15‘,end=‘2017-11-06‘)
 14 dd=df[[‘open‘,‘high‘,‘low‘,‘close‘]]
 15
 16 #print(dd.values.shape[0])
 17
 18 dd1=dd .sort_index()
 19
 20 dd2=dd1.values.flatten()
 21
 22 g1=dd2[::-1]
 23
 24 g2=g1[0:120]
 25
 26 g3=g2[::-1]
 27
 28 gg=DataFrame(g3)
 29
 30 gg.T.to_excel(‘gg.xls‘)
 31
 32
 33
 34 #dd3=pd.DataFrame(dd2)
 35 #dd3.T.to_excel(‘d8.xls‘)
 36
 37 g=dd2[0:140]
 38 for i in range(dd.values.shape[0]-34):
 39
 40     s=dd2[i*4:i*4+140]
 41     g=row_stack((g,s))
 42
 43 fg=DataFrame(g)
 44
 45 print(fg)
 46 fg.to_excel(‘fg.xls‘)
 47
 48
 49 #-*- coding: utf-8 -*-
 50 #建立、训练多层神经网络，并完成模型的检验
 51 #from __future__ import print_function
 52 import pandas as pd
 53
 54
 55 inputfile1=‘fg.xls‘ #训练数据
 56 testoutputfile = ‘test_output_data.xls‘ #测试数据模型输出文件
 57 data_train = pd.read_excel(inputfile1) #读入训练数据(由日志标记事件是否为洗浴)
 58 data_mean = data_train.mean()
 59 data_std = data_train.std()
 60 data_train1 = (data_train-data_mean)/5  #数据标准化
 61
 62 y_train = data_train1.iloc[:,120:140].as_matrix() #训练样本标签列
 63 x_train = data_train1.iloc[:,0:120].as_matrix() #训练样本特征
 64 #y_test = data_test.iloc[:,4].as_matrix() #测试样本标签列
 65
 66 from keras.models import Sequential
 67 from keras.layers.core import Dense, Dropout, Activation
 68
 69 model = Sequential() #建立模型
 70 model.add(Dense(input_dim = 120, output_dim = 240)) #添加输入层、隐藏层的连接
 71 model.add(Activation(‘relu‘)) #以Relu函数为激活函数
 72 model.add(Dense(input_dim = 240, output_dim = 120)) #添加隐藏层、隐藏层的连接
 73 model.add(Activation(‘relu‘)) #以Relu函数为激活函数
 74 model.add(Dense(input_dim = 120, output_dim = 120)) #添加隐藏层、隐藏层的连接
 75 model.add(Activation(‘relu‘)) #以Relu函数为激活函数
 76 model.add(Dense(input_dim = 120, output_dim = 20)) #添加隐藏层、输出层的连接
 77 model.add(Activation(‘sigmoid‘)) #以sigmoid函数为激活函数
 78 #编译模型，损失函数为binary_crossentropy，用adam法求解
 79 model.compile(loss=‘mean_squared_error‘, optimizer=‘adam‘)
 80
 81 model.fit(x_train, y_train, nb_epoch = 100, batch_size = 8) #训练模型
 82 model.save_weights(‘net.model‘) #保存模型参数
 83
 84 inputfile2=‘gg.xls‘ #预测数据
 85 pre = pd.read_excel(inputfile2)
 86
 87 pre_mean = data_mean[0:120]
 88 pre_std = pre.std()
 89 pre1 = (pre-pre_mean)/5  #数据标准化
 90
 91 pre2 = pre1.iloc[:,0:120].as_matrix() #预测样本特征
 92 r = pd.DataFrame(model.predict(pre2))
 93 rt=r*5+data_mean[120:140].as_matrix()
 94 print(rt.round(2))
 95
 96
 97
 98 rt.to_excel(‘rt.xls‘)
 99
100 #print([email protected]_train.iloc[:,116:120].std().values+data_mean[116:120].as_matrix())
101
102
103
104 a=list(df.index[0:-1])
105
106 b=a[0]
107
108 c= datetime.datetime.strptime(b,‘%Y-%m-%d‘)
109
110 d = date2num(c)
111
112
113 c1=[d+i+1 for i in range(5)]
114 c2=np.array([c1])
115
116 r1=rt.values.flatten()
117 r2=r1[0:4]
118 for i in range(4):
119
120     r3=r1[i*4+4:i*4+8]
121     r2=row_stack((r2,r3))
122
123 c3=column_stack((c2.T,r2))
124 r5=DataFrame(c3)
125
126 if len(c3) == 0:
127     raise SystemExit
128
129 fig, ax = plt.subplots()
130 fig.subplots_adjust(bottom=0.2)
131
132 #ax.xaxis.set_major_locator(mondays)
133 #ax.xaxis.set_minor_locator(alldays)
134 #ax.xaxis.set_major_formatter(mondayFormatter)
135 #ax.xaxis.set_minor_formatter(dayFormatter)
136
137 #plot_day_summary(ax, quotes, ticksize=3)
138 candlestick_ohlc(ax, c3, width=0.6, colorup=‘r‘, colordown=‘g‘)
139
140 ax.xaxis_date()
141 ax.autoscale_view()
142 plt.setp(plt.gca().get_xticklabels(), rotation=45, horizontalalignment=‘right‘)
143
144 ax.grid(True)
145 #plt.title(‘000002‘)
146 plt.show()