#!/usr/bin/env python # -*- coding: utf-8 -*- # author:mj time:2020/7/28 import pandas as pd import numpy as np import matplotlib.pyplot as plt import matplotlib matplotlib.rcParams['font.sans-serif'] = ['WenQuanYi Micro Hei'] matplotlib.rcParams['axes.unicode_minus'] = False import json import warnings warnings.filterwarnings('ignore') # 数据预处理函数 def data_pre(df): # 填充缺失值，并更新索引 # 现货填充 index1 = df['现货价格'].index[np.where(df['现货价格'].isnull())[0]] for item in index1: qian = df.loc[(item - 3):(item - 1), '现货价格'].values.tolist() hou = df.loc[(item + 1):(item + 3), '现货价格'].values.tolist() df.loc[index1[0], '现货价格'] = np.mean(qian + hou) # 期货填充 index2 = df['期货价格'].index[np.where(df['期货价格'].isnull())[0]] for item in index2: qian = df.loc[(item - 3):(item - 1), '期货价格'].values.tolist() hou = df.loc[(item + 1):(item + 3), '期货价格'].values.tolist() df.loc[item, '期货价格'] = np.nanmean(qian + hou) # 更新索引 df = df.reset_index(drop=True) # drop=True将之前的旧的索引删除 return df # 特征工程函数 def feature_processing(data, index, high_level, low_level, product_type): feature = [] i = index feature.append(data.iloc[i, 1]) # 记录现货绝对价格 feature.append(data.iloc[i, 2]) # 记录期货绝对价格 describe = data.iloc[(i - 60):i, 1].describe().values feature.append(np.round(describe[1], 3)) # 统计指标的平均值 feature.append(np.round(describe[2], 3)) # 统计指标的标准差 feature.append(describe[6]) # 统计指标的75%分位 describe1 = data.iloc[(i - 60):i, 2].describe().values feature.append(np.round(describe1[1])) # 统计指标的平均值 feature.append(np.round(describe1[2], 3)) # 统计指标的标准差 feature.append(describe1[6]) # 统计指标的75%分位 feature.append(data.iloc[i, 1] - data.iloc[i, 2]) # 记录现货期货基差 if product_type in ['螺纹钢', '热卷', '铁矿石']: feature.append(np.round(data.iloc[(i - 29):i, 1].corr(data.iloc[(i - 29):i, 2]), 3)) # 记录现货期货相关性系数 else: pass if (data.iloc[(i - 60), 1] - data.iloc[(i - 60), 2]) - (data.iloc[i, 1] - data.iloc[i, 2]) > 0: feature.append(1) else: feature.append(0) # 记录基差走势，变弱( >0 )则为1，变强( <=0 )则为0 if data.iloc[i, 1] - data.iloc[(i - 60), 1] > 0: feature.append(1) else: feature.append(0) # 记录现价走势，价格升( >0 )则为1，否则为0 if data.iloc[i, 2] - data.iloc[(i - 60), 2] > 0: feature.append(1) else: feature.append(0) # 记录期价走势，价格升( >0 )则为1，否则为0 # 下面是添加时间季度变量，后期需要进行特征编码的,为了将训练集和测试集都标准化为函数，这里统一人为编码 if data.iloc[i, 0].month in [1, 2, 3]: # 第一季度，1 0 0 0 feature.append(1) feature.append(0) feature.append(0) feature.append(0) elif data.iloc[i, 0].month in [4, 5, 6]: # 第二季度，0 1 0 0 feature.append(0) feature.append(1) feature.append(0) feature.append(0) elif data.iloc[i, 0].month in [7, 8, 9]: # 第三季度 0 0 1 0 feature.append(0) feature.append(0) feature.append(1) feature.append(0) elif data.iloc[i, 0].month in [10, 11, 12]: # 第四季度 0 0 0 1 feature.append(0) feature.append(0) feature.append(0) feature.append(1) return feature # 包含当天价格的历史价格data_excel表的表名,表中列明为：日期，现货价格，期货价格 def Model_BuyTaobao(input_path, out_path, days, product_type): df = pd.read_excel(input_path) if product_type == '螺纹钢': # 选择若干年，如 2017-1月以后和2010-2013，价格>3000的 # 如果是其它品种，可能会直接用所有时间段的，不需要进行时间区间的选择 df = df[(df.日期 > '2017-1-1') | (df.日期 > '2009-12-31') & (df.日期 <= '2013-12-31')] data = data_pre(df) elif product_type in ['热卷', '铁矿石']: data = data_pre(df) elif product_type in ['焦煤', '焦炭']: data = df.dropna() data = data.reset_index(drop=True) # 搜索 guancha = {} # 观察 for i in range(len(data) - days): l1_jicha = [] l1_jicha.append(data.iloc[i, 1] - data.iloc[i, 2]) l1_jicha.append(data.iloc[(i + int(days / 3) - 1), 1] - data.iloc[(i + int(days / 3) - 1), 2]) l1_jicha.append(data.iloc[(i + int(days * 2 / 3) - 1), 1] - data.iloc[(i + int(days * 2 / 3) - 1), 2]) l1_jicha.append(data.iloc[(i + days - 1), 1] - data.iloc[(i + days - 1), 2]) guancha[i] = (l1_jicha[0] - l1_jicha[1]) + (l1_jicha[0] - l1_jicha[2]) + (l1_jicha[0] - l1_jicha[3]) biaoxing_list = [item for item in guancha.values() if item > 0] low_level = np.percentile(biaoxing_list, 30) # 146 high_level = np.percentile(biaoxing_list, 80) # 535 dict_label = {} for i in range(len(data) - days): l1_jicha = [] l1_jicha.append(data.iloc[i, 1] - data.iloc[i, 2]) l1_jicha.append(data.iloc[(i + int(days / 3) - 1), 1] - data.iloc[(i + int(days / 3) - 1), 2]) l1_jicha.append(data.iloc[(i + int(days * 2 / 3) - 1), 1] - data.iloc[(i + int(days * 2 / 3) - 1), 2]) l1_jicha.append(data.iloc[(i + days - 1), 1] - data.iloc[(i + days - 1), 2]) panduan = (l1_jicha[0] - l1_jicha[1]) + (l1_jicha[0] - l1_jicha[2]) + (l1_jicha[0] - l1_jicha[3]) if panduan > high_level: label = 3 elif panduan >= low_level and panduan <= high_level: label = 2 elif panduan > 0 and panduan < low_level: label = 1 else: label = 0 dict_label[i] = label # print(sum(dict_label.values())) # 推演t时间之前60天的数据进行特征工程,考虑季节 train = {} time_index_z = [] for i in range(len(data)): if i >= 60: feature = feature_processing(data, i, high_level, low_level, product_type) train[i] = feature time_index_z.append(data.iloc[i, 0]) import string XX = [values for key, values in train.items()] # 用于最终放到模型中预测后绘图的 DF_Z = pd.DataFrame(XX, columns=list(string.ascii_letters[0:len(XX[0])]), index=time_index_z) # 特征数据X，结果标签数据y, 用于模型训练和测试的， 长度为1658 print(len(train)) X = [] y = [] time_index = [] for key, values in train.items(): if key in [k for k, v in dict_label.items()]: X.append(values) y.append(dict_label[key]) time_index.append(data.iloc[key, 0]) # 这里不需要再进行编码了，加上时间索引time_index df_X = pd.DataFrame(X, columns=list(string.ascii_letters[0:len(X[0])]), index=time_index) X = df_X.values # print(X[0:2])print(y[0:2]) # 随机切分训练集测试集，进行随机森林分类模型建立（随机森林比决策树好很多） from sklearn.tree import DecisionTreeClassifier from sklearn.ensemble import RandomForestClassifier from sklearn.model_selection import train_test_split train_X, test_X, train_y, test_y = train_test_split(X, y, test_size=0.1, random_state=1) # tree_model = DecisionTreeClassifier(max_depth=10, criterion='entropy') # 如果想让每次决策树的模型是一样的，需要设置参数random_state=1, tree_model = RandomForestClassifier(max_depth=10, criterion='entropy',random_state=1) tree_model.fit(train_X, train_y) tree_score = tree_model.score(train_X, train_y) tree_score1 = tree_model.score(test_X, test_y) print('决策树模型训练得分:', tree_score) print('决策树模型测试得分:', tree_score1) flag = '模型训练成功' result_Z = tree_model.predict(XX) DF_Z['所有最终结果'] = list(result_Z) # 准备绘图数据hui_data hui_data = data[data.日期 > '2018-12-31'] hui_data.set_index(["日期"], inplace=True) tianjialie = [hui_data.iloc[i, 1] if item.strftime("%Y-%m-%d") == data.iloc[-1, 0].strftime("%Y-%m-%d") else np.nan for i, item in enumerate(hui_data.index.tolist())] hui_data['结果'] = tianjialie tianjialie1 = [hui_data.iloc[i, 1] if DF_Z.loc[item]['所有最终结果'] > 0 else np.nan for i, item in enumerate(hui_data.index.tolist())] tianjialie1[-1] = np.nan hui_data['历史结果'] = tianjialie1 result = [result_Z[-1], data.iloc[-1, 0].strftime("%Y-%m-%d")] # 当天的结果和时间 # 绘制结果图形 plt.figure(figsize=(8, 5)) plt.plot(hui_data.index, hui_data['现货价格'], label='现货价格') plt.plot(hui_data.index, hui_data['期货价格'], label='期货价格') plt.xticks(rotation=30) def jidu(date): if date.month in [1, 2, 3]: jd = 1 elif date.month in [4, 5, 6]: jd = 2 elif date.month in [7, 8, 9]: jd = 3 elif date.month in [10, 11, 12]: jd = 4 return jd plt.title('期现买入套保模型策略（' + product_type + ')', fontsize=20) info = '参考要素:基差' + str(np.round(feature[8], 2)) + ',现期绝对价格' + str(np.round(feature[0], 2)) + '/' + str( np.round(feature[1], 3)) + ',当天为第' + str(jidu(data.iloc[-1, 0])) + '季度' # lim = plt.axis() # plt.text(lim[0] + (lim[1] - lim[0]) / 2, lim[3] + (lim[3] - lim[2]) / 14, info, ha='center', va='top', fontsize=10, c='k') plt.xlabel(info) for a, b in zip(hui_data.index, hui_data['结果']): if a.strftime("%Y-%m-%d") == result[1]: result_json = [] if result[0] == 0: annatation = '不适合对' + product_type + '当前活跃合约进行买入操作' result.append(annatation) plt.plot(hui_data.index, hui_data['历史结果'], 'o', ms=2, c='r', label='适合买入套保点') plt.text(a, b + 15, '不买入', ha='center', va='bottom', fontsize=20, c='r') elif result[0] == 1: annatation = '适合对' + product_type + '当前活跃合约进行买入操作,买入等级1颗星,持仓时间为' + str(days) + '天' result.append(annatation) plt.plot(hui_data.index, hui_data['历史结果'], 'o', ms=2, c='r', label='适合买入套保点') plt.plot(hui_data.index, hui_data['结果'], 'o', ms=10, c='r') plt.text(a, b + 15, '买入\n*', ha='center', va='bottom', fontsize=20, c='r') elif result[0] == 2: annatation = '适合对' + product_type + '当前活跃合约进行买入操作,买入等级2颗星,持仓时间为' + str(days) + '天' result.append(annatation) plt.plot(hui_data.index, hui_data['历史结果'], 'o', ms=2, c='r', label='适合买入套保点') plt.plot(hui_data.index, hui_data['结果'], 'o', ms=10, c='r') plt.text(a, b + 15, '买入\n**', ha='center', va='bottom', fontsize=20, c='r') elif result[0] == 3: annatation = '适合对' + product_type + '当前活跃合约进行买入操作,买入等级3颗星,持仓时间为' + str(days) + '天' result.append(annatation) plt.plot(hui_data.index, hui_data['历史结果'], 'o', ms=2, c='r', label='适合买入套保点') plt.plot(hui_data.index, hui_data['结果'], 'o', ms=10, c='r') plt.text(a, b + 15, '买入\n***', ha='center', va='bottom', fontsize=20, c='r') plt.gcf().autofmt_xdate() plt.ylabel('期/现货价格', fontsize=10) plt.legend(loc='upper left') plt.savefig(out_path + '/tksmr.png', dpi=100) #plt.show() result_json.append(result[1] + result[2]) # 类似2020-04-23不适合对XX当前活跃合约进行买入操作 result_json.append(info) with open(out_path + '/' + product_type + '买入套保推荐结果.json', 'w', encoding='utf-8') as f: json.dump(result_json, f, ensure_ascii=False) result_output = result[1] + result[2] return result_output if __name__ == "__main__": # 买入套保模型 days = 30 # 持仓时间 product_type = '铁矿石' # 螺纹钢 热卷 铁矿石 焦煤 焦炭 input_path = "/www/wwwroot/file.jsfsdata.com/tks_mr.xls" # 输入数据相对路径，指向当前目录的file目录 out_path = "/www/wwwroot/file.jsfsdata.com/rs" # 输出json的保存相对路径，指向当前目录的file目录 result_output = Model_BuyTaobao(input_path, out_path, days, product_type) print(result_output)