具体程序：
Ind.py
# -*- coding: utf-8 -*-"""Created on Wed Nov 27 13:46:50 @author: nuoyanli"""import pandas as pdimport numpy as np# 移动平均线指标的计算def MA(data, N1, N2, N3):MAN1 = pd.Series(data['Clsprc'].values).rolling(N1).mean()MAN2 = pd.Series(data['Clsprc'].values).rolling(N2).mean()MAN3 = pd.Series(data['Clsprc'].values).rolling(N3).mean()return (MAN1, MAN2, MAN3)# 指数平滑异同平均线的计算def MACD(data):EMA12 = pd.Series(data['Clsprc'].values).ewm(halflife=12).mean()EMA26 = pd.Series(data['Clsprc'].values).ewm(halflife=26).mean()DIF = EMA12 - EMA26DEA = np.zeros((len(DIF)))MACD = np.zeros((len(DIF)))for t in range(len(DIF)):if t == 0:DEA[t] = DIF[t]if t > 0:DEA[t] = (2 * DIF[t] + 8 * DEA[t - 1]) / 10MACD[t] = 2 * (DIF[t] - DEA[t])return MACD# 随机指标KDJ的计算def KDJ(data, N):# Lmin=pd.rolling_min(data['Loprc'].values,N)Lmin = pd.Series(data['Loprc'].values).rolling(N).min()Lmax = pd.Series(data['Hiprc'].values).rolling(N).max()RSV = (data['Clsprc'].values - Lmin) / (Lmax - Lmin)K = np.zeros((len(RSV)))D = np.zeros((len(RSV)))J = np.zeros((len(RSV)))for t in range(N, len(data)):if t == 0:K[t] = RSV[t]D[t] = RSV[t]if t > 0:K[t] = 2 / 3 * K[t - 1] + 1 / 3 * RSV[t]D[t] = 2 / 3 * D[t - 1] + 1 / 3 * K[t]J[t] = 3 * D[t] - 2 * K[t]return (K, D, J)# 相对强弱指标的计算def RSI(data, N):z = np.zeros(len(data) - 1)z[data.iloc[1:, 2].values - data.iloc[0:-1, 2].values >= 0] = 1z[data.iloc[1:, 2].values - data.iloc[0:-1, 2].values < 0] = -1z1 = pd.Series(z == 1).rolling(N).sum()z2 = pd.Series(z == -1).rolling(N).sum()rsi = np.zeros((len(data)))for t in range(N - 1, len(data) - 1):rsi[t] = z1[t] / (z1[t] + z2[t])return rsi# 乖离率指标def BIAS(data, N):bias = np.zeros((len(data)))man = pd.Series(data.iloc[:, 2].values).rolling(N).mean()for t in range(N - 1, len(data)):bias[t] = (data.iloc[t, 2] - man[t]) / man[t]return bias# 能量潮指标的计算def OBV(data):obv = np.zeros((len(data)))for t in range(len(data)):if t == 0:obv[t] = data['Dnshrtrd'].values[t]if t > 0:if data['Clsprc'].values[t] >= data['Clsprc'].values[t - 1]:obv[t] = obv[t - 1] + data['Dnshrtrd'].values[t]if data['Clsprc'].values[t] < data['Clsprc'].values[t - 1]:obv[t] = obv[t - 1] - data['Dnshrtrd'].values[t]return obv# 涨跌趋势（分类指标/预测指标）的计算def cla(data):y = np.zeros(len(data))z = np.zeros(len(y) - 1)for i in range(len(z)):z[data.iloc[1:, 2].values - data.iloc[0:1, 2].values > 0] = 1z[data.iloc[1:, 2].values - data.iloc[0:-1, 2].values == 0] = 0z[data.iloc[1:, 2].values - data.iloc[0:-1, 2].values < 0] = -1y[i] = z[i]return y
理解：
Series：是一种类似于一维数组的对象，它由一组数据（各种NumPy数据类型）以及一组与之相关的数据标签（即索引）组成。 例如： 从结果可以看出Series的字符串表现形式为：索引在左边，值在右边。 在上面的例子中我并没有为数据指定具体索引，它会自动创建一个0到N-1（N为数据的长度）的整数型索引。series rolling：移动窗口，为了提升数据的准确性，将某个点的取值扩大到包含这个点的一段区间，用区间来进行判断，对于pd.Series(s).rolling(window=3).mean()就是每次往后移动一位计算3位的平均值代替当前位置。
benefit.py
# -*- coding: utf-8 -*-"""Created on Wed Nov 27 13:43:56 @author: nuoyanli"""from predict import result1import pandas as pddata = pd.read_excel('./数据/data.xlsx')r_list = [] # 存放收益率r_trd = [] # 存放交易日期for t in range(len(result1) - 1):# 如果预测结果为1，执行投资策略if result1['预测结果'].values[t] == 1:p1 = data.loc[data['Trddt'].values == result1['交易日期'].values[t], 'Clsprc'].valuesdt = data.loc[data['Trddt'].values > result1['交易日期'].values[t], ['Trddt', 'Clsprc']]dt = dt.sort_values('Trddt')p2 = dt['Clsprc'].values[0]r = (p2 - p1) / p1r_list.append(r)r_trd.append(result1['交易日期'].values[t])r_total = sum(r_list)print(r_total)trd_r = {'交易日期': r_trd, '收益率': r_list}trd_r = pd.DataFrame(trd_r)print(trd_r)
predict.py
# -*- coding: utf-8 -*-"""Created on Wed Nov 27 14:23:52 @author: nuoyanli"""from main import Dataimport pandas as pdfrom sklearn.preprocessing import StandardScaler# 提取训练和测试数据x1 = Data['交易日期'] >= '-01-01'x2 = Data['交易日期'] <= '-11-30'xx = x1 & x2index = xx.values == Trueindex1 = xx.values == Falsex_train = Data.iloc[index, 1:15]y_train = Data.iloc[index, [15]]x_test = Data.iloc[index1, 1:15]y_test = Data.iloc[index1, [15]]# 数据标准化scaler = StandardScaler()scaler.fit(x_train)x_train = scaler.transform(x_train)x_test = scaler.transform(x_test)##支持向量机模型from sklearn import svmclf = svm.SVC()clf.fit(x_train, y_train)result = clf.predict(x_test)sc = clf.score(x_train, y_train)# print(sc)print(sc)result = pd.DataFrame(result)# 提取预测样本的交易日期ff = Data.iloc[index1, 0]# 将预测结果与实践结果整合在一起，进行比较pm1 = {'交易日期': ff.values, '预测结果': result.iloc[:, 0].values, '实际结果': y_test.iloc[:, 0].values}result1 = pd.DataFrame(pm1)z = result1['预测结果'].values - result1['实际结果'].valuesR = len(z[z == 0]) / len(z)# 预测的准确率print(R)print(result1)list1 = result1['交易日期']list2 = result1['预测结果']list3 = result1['实际结果']D = {'time': list1, 'predict': list2, 'true': list3}D = pd.DataFrame(D)D.to_excel('DD.xlsx')
main.py
# -*- coding: utf-8 -*-"""Created on Wed Nov 27 14:03:16 @author: nuoyanli"""import Indimport pandas as pddata = pd.read_excel('./数据/data.xlsx')MA = Ind.MA(data, 5, 10, 20)macd = Ind.MACD(data)kdj = Ind.KDJ(data, 9)rsi6 = Ind.RSI(data, 6)rsi12 = Ind.RSI(data, 12)rsi24 = Ind.RSI(data, 24)bias5 = Ind.BIAS(data, 5)bias10 = Ind.BIAS(data, 10)bias20 = Ind.BIAS(data, 20)obv = Ind.OBV(data)y = Ind.cla(data)# 将计算出的技术指标与交易日期以及股价的涨跌趋势利用字典整合在一起pm = {'交易日期': data['Trddt'].values}PM = pd.DataFrame(pm)DF = {'MA5': MA[0], 'MA10': MA[1], 'MA20': MA[2], 'MACD': macd,'K': kdj[0], 'D': kdj[1], 'J': kdj[2], 'RSI6': rsi6, 'RSI12': rsi12,'RSI24': rsi24, 'BIAS5': bias5, 'BIAS10': bias10, 'BIAS20': bias20, 'OBV': obv}DF = pd.DataFrame(DF)s1 = PM.join(DF)y1 = {'涨跌趋势': y}ZZ = pd.DataFrame(y1)s2 = s1.join(ZZ)# 去掉空值ss = s2.dropna()# 将ss中第6列不为0的值提取出来，存放到Data中Data = ss[ss.iloc[:, 6].values != 0]print(Data)Data.to_excel('res.xlsx')
