# coding: UTF-8 import configparser import os import psutil import scipy.io as scio import numpy as np import pandas as pd import sys from backtest_framework.print_type import PrintType as p import copy import warnings from sklearn.linear_model import LinearRegression import scipy.stats as st from wby_utils.path.Path import get_all_dirs warnings.filterwarnings("ignore") class DataApi: def getConfig(self): """ 赋值路径用函数 :param: :return: """ try: self.config = configparser.ConfigParser() self.config.read(self.configPath, encoding="utf-8") except Exception as e: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "配置文件读取失败!") else: self.local_db_path = self.config['data_path']['local_db'] self.dataPathList.append(self.config['data_path']['base_data_path']) self.factorNpzPath = self.config['data_path']['derivative_factor_npz_path'] # 因子路径 self.dataPathList.append(self.factorNpzPath) self.result_path = self.config['data_path']['result_path'] # 结果路径 self.dataPathList.append(self.result_path) def getLpp(self): """ 获取回测下标用函数 :return: int，回测下标 """ return self.lpp def setLpp(self, lpp): """ 设置回测下标用函数 :return: """ self.lpp = lpp def version(process_name): """ 展示版本用函数 :param process_name: 运行程序类名，char类型 :return: """ crossbar = "===============================================" version = "==" + process_name + " for Python3, Version: 0.0.1==" p.version_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, crossbar) p.version_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, version) p.version_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, crossbar) def loadDerivativeFactorToGlobal(self, factorName, shapeType, typeLen): """ 从指定目录读取因子数据用函数 :param factorName:因子名 :param shapeType:char类型，‘tensor’表示三维矩阵型因子，’matrix‘表示矩阵型因子，’vector‘表示列表型因子 :param typeLen:typeLen在当type为’vector‘的时候的列数，‘tensor’二维和三维的长度 :return: """ if factorName in self.globalSave.keys(): p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "请求的因子已被导入！") return tradeDate = self.getAllData("trade_date_int") filePath_npz = self.factorNpzPath + factorName + ".npz" filePath_mat = self.factorNpzPath + factorName + ".mat" try: try: data = np.load(filePath_npz) single_key = data.files[0] factorData = data[single_key] except: factorData = scio.loadmat(filePath_mat)[factorName] site = np.max([1, np.shape(factorData)[0] - self.factorGeneratorTradeDateCount - 1]) self.startDate = str(tradeDate[site][0]) self.endDate = str(tradeDate[-1][0]) if shapeType == "matrix": factorData = self.myFactorReshape(factorData) elif shapeType == 'vector': factor = np.nan * np.zeros((len(tradeDate), factorData.shape[1])) factor[:factorData.shape[0], :factorData.shape[1]] = factorData factorData = factor elif shapeType == 'tensor': factor = np.nan * np.zeros((len(tradeDate), factorData.shape[1], factorData.shape[2])) factor[:factorData.shape[0], :factorData.shape[1], :factorData.shape[2]] = factorData factorData = factor except: self.startDate = str(tradeDate[0][0]) self.endDate = str(tradeDate[-1][0]) if shapeType == 'matrix': factorData = self.myFactorReshape() elif shapeType == 'vector': factorData = np.nan * np.zeros((len(tradeDate), typeLen)) elif shapeType == 'tensor': if len(typeLen) == 1: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "typeLen参数不能为标量！") factorData = np.nan * np.zeros((len(tradeDate), typeLen[0], typeLen[1])) self.globalSave[factorName] = factorData p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "导入" + factorName + "文件！") def setGlobal(self, globalName, globalValue): """ 存储全局变量用函数 :param globalName: char类型，全局变量名 :param globalValue: 任意类型，全局变量值 :return: """ self.globalSave[globalName] = globalValue def getGlobal(self, globalName): """ 获取全局变量用函数 :param globalName: char类型，全局变量名 :return: 任意类型， 全局变量值 """ try: return self.globalSave[globalName] except: p.warning_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, f"请求的全局变量不存在！{globalName}") return None def getAllData(self, key): """ 获取数据文件用函数 :param key: char类型，数据名称 :return: 数据文件，np.array格式 """ if key not in self.allData: npzFile = key + '.npz' matFile = key + '.mat' for dataPath in self.dataPathList: if npzFile in os.listdir(dataPath): data = np.load(dataPath + "\\\\" + npzFile, allow_pickle=True) single_key = data.files[0] value = data[single_key] self.allData[key] = value p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "因子文件" + key + "成功导入！") return self.allData[key] elif matFile in os.listdir(dataPath): value = scio.loadmat(dataPath + matFile)[key] self.allData[key] = value p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "因子文件" + key + "成功导入！") return self.allData[key] p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "因子文件导入失败！找不到因子文件" + key + "，请更新因子！" + dataPath) return self.allData[key] def getDataPackage(self, key): """ 获取数据文件用函数 :param key: char类型，数据名称 :return: 数据文件，np.array格式 """ if key not in self.allData: npzFile = key + '.npz' matFile = key + '.mat' for dataPath in self.dataPathList: if npzFile in os.listdir(dataPath): data = np.load(dataPath + "\\\\" + npzFile, allow_pickle=True) self.allData[key] = data p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "数据包" + key + "成功导入！") return self.allData[key] elif matFile in os.listdir(dataPath): value = scio.loadmat(dataPath + matFile)[key] self.allData[key] = value p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "数据包" + key + "成功导入！") return self.allData[key] p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "数据包导入失败！找不到数据包" + key + "，请更新数据包！") return self.allData[key] def getFieldTickData(self, code, tradeDate): """ 获取tick交易数据用函数 :param code: char类型，股票代码 :param tradeDate:int类型，交易日期 :return: np.array类型，tick交易数据 """ trade_date = self.getAllData('trade_date_int') tradeDateSite = np.where(trade_date == int(tradeDate))[0][0] if not tradeDateSite: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "请求的日期为非交易日！") elif tradeDateSite > self.getLpp(): p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "禁止提取未来数据！") else: quotePath = self.local_db_path + 'quote_data\\stock\\xt_tick_data\\' + str(tradeDate) + '\\' dataFile = f"{quotePath}{code}.npz" ret = np.load(dataFile, allow_pickle=True) return ret def getFieldMinuteData(self, fieldName, tradeDate): """ 获取日内交易数据用函数 :param fieldName: char类型，日内交易数据名称 :param tradeDate: int类型，交易日期 :return: np.array类型，日内交易数据 """ trade_date = self.getAllData('trade_date_int') tradeDateSite = np.where(trade_date == int(tradeDate))[0][0] if not tradeDateSite: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "请求的日期为非交易日！") elif tradeDateSite > self.getLpp(): p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "禁止提取未来数据！") else: if fieldName[-2:] == '1m': quotePath = self.local_db_path + 'quote_data\\stock\\minute_data\\1m\\' elif fieldName[-2:] == '5m' and not fieldName[-3:] == '15m': quotePath = self.local_db_path + 'quote_data\\stock\\minute_data\\5m\\' elif fieldName[-3:] == '10m': quotePath = self.local_db_path + 'quote_data\\stock\\minute_data\\10m\\' elif fieldName[-3:] == '15m': quotePath = self.local_db_path + 'quote_data\\stock\\minute_data\\15m\\' elif fieldName[-3:] == '30m': quotePath = self.local_db_path + 'quote_data\\stock\\minute_data\\30m\\' elif fieldName[-3:] == '60m': quotePath = self.local_db_path + 'quote_data\\stock\\minute_data\\60m\\' else: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "请求的因子文件为非日内行情数据！") dataFile = f"{quotePath}{fieldName}_{tradeDate}.npz" data = np.load(dataFile, allow_pickle=True) single_key = data.files[0] value = data[single_key] ret = self.quoteCorrection(value) return ret def getStartTradeDate(self, startTradeDate, change): """ 设置开始交易日期标签用函数 :param startTradeDate: int类型，开始交易日期 :return: startDateFlag数值，int类型 """ tradeDate = self.getAllData('trade_date_int') if startTradeDate and startTradeDate != np.inf: date = int(startTradeDate) ret = 0 minDistant = np.inf for i in list(range(0, len(tradeDate))): if tradeDate[i] - date < minDistant and tradeDate[i] - date >= 0: minDistant = tradeDate[i] - date ret = i else: ret = 0 if change == True: self.startDateFlag = ret return ret def getEndTradeDate(self, endTradeDate=""): """ 设置结束交易日期标签用函数 :param endTradeDate: int类型，结束交易日期 :return: endDateFlag数值，int类型 """ tradeDate = self.getAllData('trade_date_int') if endTradeDate and endTradeDate != np.inf: date = int(endTradeDate) ret = 0 minDistant = np.inf for i in list(range(0, len(tradeDate))): if date - tradeDate[i] < minDistant and date - tradeDate[ i] >= 0: minDistant = date - tradeDate[i] ret = i else: ret = len(tradeDate) self.endDateFlag = ret return ret def myFactorReshape(self, inputData): """ 规范数据形状用函数 :param inputData: np.array类型，输入的数据 :return: np.array类型，重塑形状之后的数据 """ tradeDate = self.getAllData("trade_date_int") stock_list = self.getAllData("stock_code") if not inputData is None: if inputData.shape[1] < len(stock_list): ret = np.hstack((inputData, np.nan * np.zeros( (inputData.shape[0], len(stock_list) - inputData.shape[1])))) else: ret = inputData if inputData.shape[0] < len(tradeDate): ret = np.vstack((ret, np.nan * np.zeros( (len(tradeDate) - ret.shape[0], ret.shape[1])))) else: ret = np.nan * np.zeros([len(tradeDate), len(stock_list)]) return ret def myWinsorizeCalc(self, inputValue, sigma, calcType): """ 去极值用函数 :param inputValue: np.array类型，输入的数据 :param sigma: int类型，控制宽度参数 :param calcType: char类型，“std”、“median”、“Briner”，代表去极值方法 :return: np.array类型，去极值后的数据 """ inputValue_ = copy.deepcopy(inputValue) if calcType == 'std' or calcType == 'median': if calcType == 'std': meanValue = np.nanmean(inputValue_) stdValue = np.nanstd(inputValue_) elif calcType == 'median': meanValue = np.nanmedian(inputValue_) stdValue = np.nanmedian(np.abs(inputValue_ - meanValue)) upperThreshold = meanValue + stdValue * sigma lowerThreshold = meanValue - stdValue * sigma while len(np.where(inputValue_ > upperThreshold)[1]) > 0 or len( np.where(inputValue_ < lowerThreshold)[1]) > 0: inputValue_[0, np.where( inputValue_ > upperThreshold)[1]] = upperThreshold inputValue_[0, np.where( inputValue_ < lowerThreshold)[1]] = lowerThreshold if calcType == 'std': meanValue = np.nanmean(inputValue_) stdValue = np.nanstd(inputValue_) elif calcType == 'median': meanValue = np.nanmedian(inputValue_) stdValue = np.nanmedian(np.abs(inputValue_ - meanValue)) upperThreshold = meanValue + stdValue * sigma lowerThreshold = meanValue - stdValue * sigma elif calcType == 'Briner': inputValue_ = self.myStandardizeCalc(inputValue_) s_plus = max(0, min(1, 0.5 / np.nanmax(inputValue_ - sigma))) s_minus = max(0, min(1, -0.5 / np.nanmin(inputValue_ + sigma))) inputValue_[0, np.where( inputValue_ > sigma)[1]] = sigma * (1 - s_plus) + inputValue_[ 0, np.where(inputValue_ > sigma)[1]] * s_plus inputValue_[0, np.where( inputValue_ < -sigma)[1]] = -sigma * (1 - s_minus) + inputValue_[ 0, np.where(inputValue_ < -sigma)[1]] * s_minus else: meanValue = np.nan stdValue = np.nan p.warning_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "标准化类型的入参有误！") return inputValue_ def myStandardizeCalc(self, inputValue, calcType, filter): """ 标准化用函数 :param inputValue: np.array类型，输入的数据 :param calcType: char类型，“zscore”或“0-1”，代表标准化方法 :return: np.array类型，标准化后的数据 """ if calcType == 'zscore': return (inputValue - np.nanmean(inputValue)) / np.nanstd(inputValue) elif calcType == '0-1': return (inputValue - np.nanmin(inputValue)) / ( np.nanmax(inputValue) - np.nanmin(inputValue)) elif calcType == 'ppf': ppf = st.norm.ppf(1 - np.sum(~np.isnan(inputValue)) / np.sum(~np.isnan(filter))) return ppf + (np.nanmax(inputValue) - ppf) / (np.nanmax(inputValue) - np.nanmin(inputValue)) * ( inputValue - np.nanmin(inputValue)) def myNanFillAsData(self, inputValue, data): """ 缺失值填充用函数 :param inputValue: np.array类型，输入的数据 :param data: float类型，填充缺失值用参数 :return: np.array类型，填充缺失值后的数据 """ stockTradeDayCount = self.getFieldData('stockTradeDayCount', 1) inTheMarket = np.nan * np.zeros([1, np.shape(stockTradeDayCount)[1]]) inTheMarket[0, np.where(stockTradeDayCount > 0)[1]] = 1 for i in range(np.shape(inTheMarket)[1]): if np.isnan(inTheMarket[0, i]) == False and np.isnan( inputValue[0, i]): inputValue[0, i] = data return inputValue * inTheMarket def myIndustryFlag(zx): """ 生成各股行业one-hot矩阵用函数 :param zx: np.array类型，中信行业数据 :return: np.array类型，各股行业one-hot矩阵 """ ret = np.zeros([int(np.nanmax(zx)), int(np.shape(zx)[1])]) for i in list(range(int(np.nanmax(zx)))): ret[i, np.where(zx == i + 1)[1]] = 1 return ret def removeAllData(self, key, usedMemoProcess, usedMemoPercent): """ 从内存移除数据用函数 :param key: list类型，包含数据名称的列表 :param usedMemoProcess: float类型，python进程占用内存GB上限 :param usedMemoPercent: float类型，计算机内存占用百分比上限 :return: """ if not usedMemoProcess and not usedMemoPercent: for i in range(len(key)): self.allData.pop(key[i]) p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "请求的因子/数据包" + key[i] + "删除完毕！") else: mem_Process = round( psutil.Process(os.getpid()).memory_info().rss / 1024 / 1024 / 1024, 2) memPercent = round(psutil.virtual_memory().percent / 100, 2) if (usedMemoProcess and mem_Process > usedMemoProcess) or ( usedMemoPercent and memPercent > usedMemoPercent): for i in range(len(key)): self.allData.pop(key[i]) p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "请求的因子/数据包" + key[i] + "删除完毕！") def getStockName(self): """ 获取股票名称用函数 :param :return: np.array类型，股票名称列表 """ return self.getAllData("stockName") def getStockCode(self): """ 获取股票数字代码用函数 :param :return: np.array类型，股票数字代码列表 """ stock_list = self.getStockList() return np.array(list(map(lambda x: int(x[:6]), stock_list))) def getStockList(self): """ 获取股票代码用函数 :param :return: np.array类型，股票代码列表 """ return self.getAllData("stock_code") def getTradeDateList(self, varargin): """ 返回交易日历，不传入参数返回历史所有的交易日历，传入参数返回固定交易日数的交易日历 :param days: 返回固定交易日期数 :return: List """ ret = self.getAllData("trade_date_int") if varargin: return ret[self.getLpp() - varargin + 1:self.getLpp() + 1, :] else: return ret[:self.getLpp() + 1, :] def isLastDate(self): """ 判断是否为最后一个回测截面，如果是，返回True，否则返回False :return: bool """ return self.lpp == self.endDateFlag def getStockIndexQuotes(self, indexCode, varargin): # TODO """ 获取股票指数行情数据 :param indexCode:指数代码 :param varargin:可选参数，截取的日期行数 :return: """ lpp = self.lpp nowDate = self.getNowDate() indexQuotes = self.getAllData("stockIndexQuotes")[()][indexCode] site = np.where(indexQuotes[:, 0] == nowDate)[0][0] if varargin: return indexQuotes[lpp - varargin + 1:lpp + 1] else: return indexQuotes[:lpp + 1] def getStockIndexDailyQuotes(self, indexCode, tradeDate): """ 获取股票指数日内行情数据 :param indexCode:指数代码 :param tradeDate:行情日期 :return: """ return self.getAllData("stockIndexDailyQuotes")[()][indexCode][tradeDate] def getFieldData(self, fieldName, varargin): """ 获取因子矩阵用函数 :param fieldName:因子名 :param varargin: 返回的因子行数 :return: """ lpp = self.lpp if varargin: return self.getAllData(fieldName)[lpp - varargin + 1:lpp + 1] else: return self.getAllData(fieldName)[:lpp + 1] def getFieldDataByIndex(self, fieldName, index): """ 获取因子矩阵用函数 :param fieldName:因子名 :param index: 回测下标 :return: """ return self.getAllData(fieldName)[index].reshape(1, -1) def saveGlobal(self, globalName): """ 存储因子到指定文件夹目录 :param globalName:请求存储的因子名 :return: """ factorData = self.getGlobal(globalName) if sum(factorData[-1] == 0) + sum(np.isnan( factorData[-1])) == factorData.shape[1]: p.warning_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "衍生因子 " + globalName + " 异常，结果未保存！") else: np.savez_compressed(self.factorNpzPath + globalName + '.npz', factorData) if self.saveFactorResultAsMat == True: scio.savemat(self.factorNpzPath + globalName + '.mat', {globalName: factorData}) p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "衍生因子 " + globalName + " 保存完成！") def saveGlobalUnconditionally(self, globalName): """ 无条件存储因子到指定文件夹目录 :param globalName:请求存储的因子名 :return: """ factorData = self.getGlobal(globalName) np.savez_compressed(self.factorNpzPath + globalName + '.npz', factorData) if self.saveFactorResultAsMat == True: scio.savemat(self.factorNpzPath + globalName + '.mat', {globalName: factorData}) p.disp_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "衍生因子 " + globalName + " 保存完成！") def getBarCount(self): """ 返回bar计数 :return: """ return self.barCount def getNowDate(self, type): """ 返回回测当日下标的日期 :return: """ if type == 'int': return self.getAllData("trade_date_int")[self.lpp][0] elif type == 'datetime': return self.getAllData("trade_date_datetime")[self.lpp][0] else: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "type类型有误！") def getYesterdayDate(self, type): """ 返回回测前一日下标的日期 :return: """ if type == 'int': return self.getAllData("trade_date_int")[self.lpp - 1][0] elif type == 'datetime': return self.getAllData("trade_date_datetime")[self.lpp - 1][0] else: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "type类型有误！") def getNextDate(self, type): """ 返回回测下一日下标的日期 :return: """ if type == 'int': tradeDate = self.getAllData("trade_date_int") tradeDateHistory = self.getAllData("calendar_date_int") elif type == 'datetime': tradeDate = self.getAllData("trade_date_datetime") tradeDateHistory = self.getAllData("calendar_date_datetime") else: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "type类型有误！") index = np.where(tradeDateHistory == tradeDate[self.lpp])[0][0] return tradeDateHistory[index + 1][0] def getHalfTimeWeight(halflife, window): ''' 得到半衰期权重,WLS回归用 :param halflife: 半衰期时长，int类型 :param window: 观察期窗口，int类型 :return: 半衰期标准化后的权重，np.array类型，一行 ''' sigma = 0.5 ** (1 / halflife) weight = np.empty([1, window]) for i in range(window): weight[0, i] = sigma ** (window - i) weight = weight / np.sum(weight) return weight.reshape(1, -1) # 转化为1Darray 方便 def myFiltering(self, inputValue): """ 将非交易日的因子值设置为np.nan用函数,同时将inf转化为nan :param inputValue: np.array类型，输入的数据 :param data: float类型，填充缺失值用参数 :return: np.array类型，填充缺失值后的数据 """ stockTradeDayCount = self.getFieldData('stock_on_list_duration', 1) stockTradeDayCount = stockTradeDayCount.astype(np.float32) stockTradeDayCount[stockTradeDayCount > 0] = 1 stockTradeDayCount[stockTradeDayCount == 0] = np.nan ret = inputValue * stockTradeDayCount ret[np.isinf(ret)] = np.nan return ret def myRank(self, data, type): """ 将因子截面数据做排名返回 :param data: np.array类型，因子截面数据的输入值 :param type: char类型，降序（descend）、升序（ascend） :return: np.array类型，data的排名返回值 """ if type == '降序' or type == 'descend': rev = -1 elif type == '升序' or type == 'ascend': rev = 1 data_ = copy.deepcopy(data) data_ = np.array(data_) Site = np.where(~np.isnan(data_)) data_[Site] = np.argsort(data_[Site] * rev) data_[Site] = np.argsort(data_[Site]) return data_.reshape(1, -1) def mySort(self, data, type): """ 将因子排序下标返回 :param data: np.array类型，因子截面数据的输入值 :param type: char类型，降序（descend）、升序（ascend） :return: np.array类型，data的排名下标返回值 """ if type == '降序' or type == 'descend': rev = -1 elif type == '升序' or type == 'ascend': rev = 1 data_ = copy.deepcopy(data) Site = np.where(~np.isnan(data_)) data_ = np.array(data_) data_[Site] = np.argsort(data_[Site] * rev) return data_.reshape(1, -1) def quoteCorrection(self, inputvalue): ''' 纠正getFieldDailyData获得数据列数错误用函数 :param inputValue: np.array类型，输入的数据 :return: np.array类型，列数正确的函数 ''' column_num = len(self.getAllData('stock_code')) value_num = inputvalue.shape[1] if column_num > value_num: ret = np.hstack( (inputvalue, np.nan * np.zeros([inputvalue.shape[0], column_num - value_num]))) elif column_num <= value_num: ret = inputvalue[:, :column_num] return ret def regForNeutralizeFactor(self, factor, list1): ''' 中性化因子用函数 :param factor: 被中性化的因子,二维数组(1,) :param list: 包含中性化因子数组的列表，列表内部数据格式：二维数组(1,) :return: ndarray类型，回归后的残差，二维数组(1,) ''' factor = factor.T neutralizeList = np.zeros((factor.shape[0], len(list1))) * np.nan result = np.zeros((1, factor.shape[0])) * np.nan for i in range(len(list1)): neutralizeList[:, i] = list1[i][0, :] y = factor x = neutralizeList for i in range(x.shape[1]): if i == 0: y_not_empty = np.where(~np.isnan(y))[0].tolist() x_not_empty = np.where(~np.isnan(x[:, i]))[0].tolist() not_empty = list(set(x_not_empty).intersection(set(y_not_empty))) else: y_not_empty = not_empty x_not_empty = np.where(~np.isnan(x[:, i]))[0].tolist() not_empty = list(set(x_not_empty).intersection(set(y_not_empty))) y = y[not_empty] x = x[not_empty] m = LinearRegression() m.fit(x, y) residual = y - m.predict(x) for i in range(len(not_empty)): result[0, not_empty[i]] = residual[i, 0] return result def myNeutralizeCalc(self, y, x): ''' :param x: :param y: :return: ''' fit_data = np.vstack((x, y)) site = np.where(~np.isnan(np.sum(fit_data, axis=0)))[0] x_ = x[:, site] y_ = y[:, site] resid = np.nan * np.zeros(y.shape).reshape(1, -1) if x_.size and y_.size: m = LinearRegression() m.fit(x_.T, y_.T) resid[0, site] = (y_.T - m.predict(x_.T)).reshape(1, -1) return resid def getReportPeriodsList(self, modified): ''' 获取最新报告期列表用函数 :param modified: bool类型，确定是否需要扩展 :return: np.array类型，报告期数组 ''' report_periods_list = self.getFieldData('reportPeriodsList') report_periods_list = report_periods_list[~np.isnan(report_periods_list )] report_periods_list = np.array(list(set(report_periods_list.tolist()))) report_periods_list = np.flipud(np.sort(report_periods_list)) if modified == False: return report_periods_list else: if report_periods_list[0] % 10000 == 1231: report_periods_list = np.insert(report_periods_list, 0, report_periods_list[0] + 9100) elif report_periods_list[0] % 10000 == 930: report_periods_list = np.insert(report_periods_list, 0, report_periods_list[0] + 301) elif report_periods_list[0] % 10000 == 630: report_periods_list = np.insert(report_periods_list, 0, report_periods_list[0] + 300) elif report_periods_list[0] % 10000 == 331: report_periods_list = np.insert(report_periods_list, 0, report_periods_list[0] + 299) return report_periods_list def getFinancialAllData(self, key): """ 导入数据包用函数 :param key: char类型，数据包名 :return: tuple类型，包含数据包数据值和数据名称的元组 """ if key in self.allData: return self.allData[key] npzFile = key + '.npz' for dataPath in self.dataPathList: if npzFile in os.listdir(dataPath): dataFile = dataPath + "\\\\" + npzFile break try: data = np.load(dataFile, allow_pickle=True) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "数据包导入失败！找不到数据包" + key + "，请更新数据包！") else: value = data['value'] fieldName = data['fieldname'] v = (value, fieldName) self.allData[key] = v return self.allData[key] def getFinancialFieldData(self, tableName, fieldName, reportDate): """ 获取金融数据 :param tableName: char类型，数据包名称 :param fieldName: char/list/tuple类型，数据包中具体数据名称 :param reportDate: int类型，报告期 :return: np.array类型，具体的金融数据 """ nextDate = self.getNextDate() tableData = self.getFinancialAllData(tableName) fieldData = tableData[0] fieldNameList = tableData[1] if isinstance(fieldName, str): try: site = np.where(fieldNameList == fieldName)[0][0] except: p.error_print( sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "在" + tableName + "数据包里找不到对应的" + fieldName + "字段！") else: fieldData = fieldData[:, [0, 1, 2, site]] stockCode = self.getStockCode() IA_ret = np.ones([1, len(stockCode)]) fieldData = fieldData[np.where( fieldData[:, 1] <= nextDate)[0], :] ret = np.nan * np.zeros([1, len(stockCode)]) fieldData = fieldData[np.where( fieldData[:, 2] == reportDate)[0], :] if fieldData.size == 0: return ret else: for i in range(len(stockCode)): dataTemp = fieldData[np.where( fieldData[:, 0] == stockCode[i])[0]] if dataTemp.size == 0: IA_ret[0, i] = np.nan continue if np.shape(dataTemp)[0] > 1: dataTemp = dataTemp[np.argsort(dataTemp[:, 1])] ret[0, i] = dataTemp[-1, -1] if tableName != 'ASHARE_PROFITNOTICE': ret[np.where(np.isnan(ret))] = 0 ret = ret * IA_ret elif isinstance(fieldName, (np.ndarray, list, tuple)): try: site = list( map(lambda x: np.where(fieldNameList == x)[0][0], fieldName)) except: var_str = fieldName[0] for i in list(range(1, len(fieldName))): var_str += ',' + fieldName[i] p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "在" + tableName + "数据包里找不到对应的" + var_str + "字段！") else: tmp = [0, 1, 2] tmp.extend(site) fieldData = fieldData[:, tmp] stockCode = self.getStockCode() IA_ret = np.ones([1, len(stockCode)]) fieldData = fieldData[np.where( fieldData[:, 1] <= nextDate)[0], :] ret = np.nan * np.zeros([len(site), len(stockCode)]) fieldData = fieldData[np.where( fieldData[:, 2] == reportDate)[0], :] if fieldData.size == 0: return ret else: for i in list(range(len(stockCode))): dataTemp = fieldData[np.where( fieldData[:, 0] == stockCode[i])[0]] if dataTemp.size == 0: IA_ret[0, i] = np.nan continue if np.shape(dataTemp)[0] > 1: dataTemp = dataTemp[np.argsort(dataTemp[:, 1])] for j in list(range(3, np.shape(dataTemp)[1])): ret[j - 3, i] = dataTemp[-1, j] if tableName != 'ASHARE_PROFITNOTICE': ret[np.where(np.isnan(ret))] = 0 ret = ret * IA_ret return ret def getFinancialFieldDataMatrix(self, tablename, fieldname, report_periods_list, datalen): ''' 获取财务数据矩阵用函数 :param tablename: char类型，表名 :param fieldname: char类型，字段名 :param report_periods_list: 1darray数组，报告期列表 :param datalen: int类型，数据行数 :return: np.array类型，财务数据矩阵 ''' ret = np.nan * np.zeros((int(datalen), len(self.getStockList()))) for i in range(datalen): try: ret[i] = self.getFinancialFieldData(tablename, fieldname, report_periods_list[i]) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "数据包导入失败！找不到数据包" + tablename + "，请更新数据包！") return ret def getFinancialFieldDataMatrixModified(self, value, tablename, fieldname, report_periods_list): ''' 注意：此函数对value进行操作，返回的是value的修正值，内存占用不变 获取财务数据修正矩阵用函数 :param tablename: char类型，修正表名 :param fieldname: char类型，具体修正字段名 :param report_periods_list: 1darray数组，报告期列表 :return: np.array类型，财务数据矩阵 ''' datalen = min(3, value.shape[0]) if fieldname != 'NET_PROFIT_EXCL_MIN_INT_INC': ret = np.nan * np.zeros((datalen, len(self.getStockList()))) for i in range(datalen): try: ret[i] = self.getFinancialFieldData(tablename, fieldname, report_periods_list[i]) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "逐行报告期数据提取失败！") value[:datalen][np.where(np.isnan( value[:datalen]))] = ret[np.where(np.isnan(value[:datalen]))] return value else: ret_1 = np.nan * np.zeros((datalen, len(self.getStockList()))) for i in range(datalen): try: ret_1[i] = self.getFinancialFieldData(tablename, fieldname, report_periods_list[i]) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "逐行报告期数据提取失败！") ret_2 = np.nan * np.zeros((datalen, len(self.getStockList()))) for i in range(datalen): try: upDown = self.getFinancialFieldData('ASHARE_PROFITNOTICE', [ 'S_PROFITNOTICE_NETPROFITMIN', 'S_PROFITNOTICE_NETPROFITMAX' ], report_periods_list[i]) * 10000 ret_2[i] = np.nanmean(upDown, axis=0) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "字段为净利润，业绩预告逐行报告期数据提取失败！") value[:datalen][np.where(np.isnan( value[:datalen]))] = ret_1[np.where(np.isnan(value[:datalen]))] value[:datalen][np.where(np.isnan( value[:datalen]))] = ret_2[np.where(np.isnan(value[:datalen]))] return value def getFinancialFieldDataMatrixModified_gta(self, value, tablename, fieldname, report_periods_list): ''' 函数 注意：此函数对value进行操作，返回的是value的修正值，内存占用不变 获取财务数据修正矩阵用函数 :param tablename: char类型，修正表名 :param fieldname: char类型，具体修正字段名 :param report_periods_list: 1darray数组，报告期列表 :return: np.array类型，财务数据矩阵 ''' dic_fieldname = {'B0011': 'GROSSREVENUE', 'B0013': 'OPERATEPROFIT', 'B001': 'TOTALPROFIT', 'A001': 'ASSET' , 'A0031': 'EQUITYPARENT', 'A003': 'EQUITY', 'B0024': 'PROFITPARENT', 'B001101': 'GROSSREVENUE', 'B002': 'PROFIT'} fieldname = dic_fieldname[fieldname] datalen = min(3, value.shape[0]) if fieldname != 'PROFITPARENT' and 'PROFIT': ret = np.nan * np.zeros((datalen, len(self.getStockList()))) for i in range(datalen): try: ret[i] = self.getFinancialFieldData(tablename, fieldname, report_periods_list[i]) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "逐行报告期数据提取失败！") value[:datalen][np.where(np.isnan( value[:datalen]))] = ret[np.where(np.isnan(value[:datalen]))] return value elif fieldname == 'PROFITPARENT': # 归母净利润 ret_1 = np.nan * np.zeros((datalen, len(self.getStockList()))) for i in range(datalen): try: ret_1[i] = self.getFinancialFieldData(tablename, fieldname, report_periods_list[i]) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "逐行报告期数据提取失败！") ret_2 = np.nan * np.zeros((datalen, len(self.getStockList()))) for i in range(datalen): try: upDown = self.getFinancialFieldData('STK_FIN_FORECFIN', [ 'PROFITPARENTFLOOR', 'PROFITPARENTCEILING' ], report_periods_list[i]) ret_2[i] = np.nanmean(upDown, axis=0) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "字段为归母净利润，业绩预告逐行报告期数据提取失败！") value[:datalen][np.where(np.isnan( value[:datalen]))] = ret_1[np.where(np.isnan(value[:datalen]))] value[:datalen][np.where(np.isnan( value[:datalen]))] = ret_2[np.where(np.isnan(value[:datalen]))] return value elif fieldname == 'PROFIT': # 净利润 ret_1 = np.nan * np.zeros((datalen, len(self.getStockList()))) for i in range(datalen): try: ret_1[i] = self.getFinancialFieldData(tablename, fieldname, report_periods_list[i]) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "逐行报告期数据提取失败！") ret_2 = np.nan * np.zeros((datalen, len(self.getStockList()))) for i in range(datalen): try: upDown = self.getFinancialFieldData('STK_FIN_FORECFIN', [ 'PROFITFLOOR', 'PROFITCEILING' ], report_periods_list[i]) ret_2[i] = np.nanmean(upDown, axis=0) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "字段为净利润，业绩预告逐行报告期数据提取失败！") value[:datalen][np.where(np.isnan( value[:datalen]))] = ret_1[np.where(np.isnan(value[:datalen]))] value[:datalen][np.where(np.isnan( value[:datalen]))] = ret_2[np.where(np.isnan(value[:datalen]))] return value def getFinancialFieldDataMatrix_Q(self, value, report_periods_list): ''' 获取财务单季度数据矩阵用函数 :param value: ndarray类型，getFieldDataMatrix获得的财务数据矩阵 :param report_periods_list: 1darray数组，报告期列表 :return: np.array类型，财务单季度数据矩阵 ''' datalen = value.shape[0] ret = np.nan * np.zeros((datalen - 1, len(self.getStockList()))) for i in range(datalen - 1): try: if report_periods_list[i] % 10000 == 331: ret[i] = value[i] else: ret[i] = value[i] - value[i + 1] except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "财务数据单季度计算失败！") return ret def getFinancialFieldDataMatrix_TTM(self, value, n): ''' 获取财务TTM数据矩阵用函数 :param value: ndarray类型，getFieldDataMatrix_Q获得的财务单季度数据矩阵 :param n: int类型，期数，默认为4 :return: np.array类型，财务TTM数据矩阵 ''' datalen = value.shape[0] ret = np.nan * np.zeros((datalen - (n - 1), len(self.getStockList()))) for i in range(datalen - (n - 1)): try: ret[i] = np.sum(value[i:(i + n), :], axis=0) except: p.error_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "财务数据TTM计算失败！") return ret def getFirstNonnan(arr, axis, invalid_val): ''' 获得矩阵每行或者每列第一个非空值所在位置用函数 :param arr: ndarray类型 :param axis: 0/1 0为每列提取，1为每行提取 :param invalid_val: int/float，若行/列全空返回的数字 :return: 1darray数组 ''' mask = ~np.isnan(arr) return np.where(mask.any(axis=axis), mask.argmax(axis=axis), invalid_val) def getNewestData(arr): ''' 获取矩阵每一列第一个非空值用函数 :param arr: ndarray类型 :return: ndarray类型，一行矩阵 ''' ret = np.nan * np.zeros((1, arr.shape[1])) for i in range(arr.shape[0]): ret[0, np.where(np.isnan(ret[0]))[0]] = arr[i][np.where(np.isnan(ret[0]))[0]] return ret def getGrowthData(self, factor_list, dataLen, type): ''' 计算growth因子用函数 :param factor_list: ndarray类型，growth因子基础值矩阵 :param dataLen: int类型，默认为1，前后两期的间隔 :param type: char类型，默认为rate，计算的growth因子种类 :return: ndarray类型，一行矩阵 ''' factor_list = np.flipud(factor_list) if type == 'rate': growth = (factor_list[dataLen:] - factor_list[:-dataLen]) / np.abs(factor_list[:-dataLen]) elif type == 'value': growth = factor_list[dataLen:] - factor_list[:-dataLen] growth = growth.astype('float') else: p.warning_print(sys._getframe().f_code.co_filename, sys._getframe().f_code.co_name, "传入的type类型有误") return None growth[np.where(growth == np.inf)] = np.nan growth[np.where(growth == -np.inf)] = np.nan growth = np.flipud(growth) growth = self.getNewestData(growth) return growth