在【答读者问23】计算指标的时候是直接使用pandas计算好指标加载进去速度快,还是在backtrader中计算指标速度快?经过一个简单的测试,发现在pandas中计算指标然后加载开高低收等数据及指标值到backtrader中和在backtrader直接计算指标,如果都是用向量的方法,速度是差不多的。今天正好有读者咨询,如果把pandas中计算好的指标改造成backtrader的指标,本篇文章就做一个示范。在开始之前,backtrader的教程部分,已经讲解过好几篇关于backtrader技术指标的相关问题,以下文章供参考:
47、backtrader的一些基本概念—技术指标(indicator)的使用教程
48、backtrader的一些基本概念----如何创建一个新的技术指标(indicator)-(2021-10-17更新)
49、【backtrader股票策略】如何实现跨周期调用技术指标的策略?
75 [backtrader期货策略]十大经典策略-分时均线交叉策略
好了,开始今天的主题~
问题:有一个pandas计算的指标,希望能够转换到backtrader中,如何实现呢?pandas计算的公式主要内容如下:
def get_mtm_atr(df): # 对mtm计算atr指标,在原有数据的基础上,输出增加了一个列,mtm_atr col_list = list(df.columns) df['mtm_l'] = df['low'] / df['low'].shift(n1) - 1 df['mtm_h'] = df['high'] / df['high'].shift(n1) - 1 df['mtm_c'] = df['close'] / df['close'].shift(n1) - 1 df['mtm_c1'] = df['mtm_h'] - df['mtm_l'] df['mtm_c2'] = abs(df['mtm_h'] - df['mtm_c'].shift(1)) df['mtm_c3'] = abs(df['mtm_l'] - df['mtm_c'].shift(1)) df['mtm_tr'] = df[['mtm_c1', 'mtm_c2', 'mtm_c3']].max(axis=1) df['mtm_atr'] = df['mtm_tr'].rolling(window=n1, min_periods=1).mean() return df[col_list+['mtm_atr']]
我们继续接着【答读者问23】的代码,测试一下在pandas中实现mtm_atr指标与在backtrader中实现mtm_atr指标。
import numpy as npimport pandas as pdimport random import datetimeimport backtrader as bt import plotly.offline as pyimport plotly.graph_objs as go""" # 生成一个1万,10万,100万,1000万的随机数,保存本地,行程csv文件。 # 然后分别使用两种方法读取,计算指标 # 方法一:使用pandas读取数据,并计算mtm_atr指标,分别保存为两个新的列:"mtm_atr"并adddata到cerebro中,写策略的时候,就不再计算mtm_atr指标 # 方法二:读取csv文件形成feed,然后adddata到cerebro中,然后在策略里面计算mtm_atr指标 # 测试在不同的数据量级别下使用的时间 """def generate_random_n_bar_df(n): start_datetime = datetime.datetime(1990,1,1,9,0,0) # bar的数据和时间都是乱生成的,估计没有那种行情是这种,但是应该是不影响测试结果的可靠性 result=[[random.randint(100,200),random.randint(100,200),random.randint(100,200),random.randint(100,200),random.randint(100,200),random.randint(100,200)] for i in range(n)] result_df = pd.DataFrame(result,columns=['open',"high","low","close","volume","openinterest"]) result_df.index=pd.to_datetime([start_datetime+datetime.timedelta(seconds=i) for i in list(range(n))]) return result_df# 从1000到100万的bar的数目模拟生成for i in range(1,21): n=i*1000 data = generate_random_n_bar_df(n) data.to_csv(f"data_{n}.csv") print(f"{n}个bar的模拟数据成功保存到工作目录") # 直接使用pandas的方法,非直接使用backtrader计算指标class ExtendPandasFeed(bt.feeds.PandasData): params = ( ('datetime', 0), ('open', 1), ('high', 2), ('low', 3), ('close', 4), ('volume', 5), ('openinterest', 6), ("mtm_atr",7), ) lines = ('mtm_atr',) datafields = [ 'datetime', 'open', 'high', 'low', 'close', 'volume', 'openinterest',"mtm_atr" ] def start(self): super(ExtendPandasFeed, self).start() # reset the iterator on each start self._rows = self.p.dataname.itertuples() def _load(self): try: row = next(self._rows) except StopIteration: return False # Set the standard datafields - except for datetime # print(self.getlinealiases()) for datafield in self.getlinealiases(): if datafield == 'datetime': continue # get the column index colidx = getattr(self.params, datafield) # print(datafield,colidx) if colidx < 0: # column not present -- skip continue # get the line to be set line = getattr(self.lines, datafield) # print(colidx) # indexing for pandas: 1st is colum, then row line[0] = row[colidx] # datetime colidx = getattr(self.params, 'datetime') tstamp = row[colidx] # convert to float via datetime and store it dt = pd.to_datetime(tstamp) dtnum = bt.date2num(dt) # get the line to be set line = getattr(self.lines, 'datetime') line[0] = dtnum # Done ... return return Trueclass NotDirectStrategy(bt.Strategy): # params = (('short_window',10),('long_window',60)) params = {"n1":20} def log(self, txt, dt=None): ''' log信息的功能''' dt = dt or bt.num2date(self.datas[0].datetime[0]) print('%s, %s' % (dt.isoformat(), txt)) def __init__(self): # 一般用于计算指标或者预先加载数据,定义变量使用 pass def next(self): # Simply log the closing price of the series from the reference # self.log(f"工商银行,{self.datas[0].datetime.date(0)},收盘价为:{self.datas[0].close[0]}") # self.log(f"short_ma:{self.short_ma[0]},long_ma:{self.long_ma[0]}") # 得到当前的size data = self.datas[0] self.log(f"close:{data.close[0]},mtm_atr:{data.mtm_atr[0]}") # def notify_order(self, order):# if order.status in [order.Submitted, order.Accepted]:# # order被提交和接受# return# if order.status == order.Rejected:# self.log(f"order is rejected : order_ref:{order.ref} order_info:{order.info}")# if order.status == order.Margin:# self.log(f"order need more margin : order_ref:{order.ref} order_info:{order.info}")# if order.status == order.Cancelled:# self.log(f"order is concelled : order_ref:{order.ref} order_info:{order.info}")# if order.status == order.Partial:# self.log(f"order is partial : order_ref:{order.ref} order_info:{order.info}")# # Check if an order has been completed# # Attention: broker could reject order if not enougth cash# if order.status == order.Completed:# if order.isbuy():# self.log("buy result : buy_price : {} , buy_cost : {} , commission : {}".format(# order.executed.price,order.executed.value,order.executed.comm)) # else: # Sell# self.log("sell result : sell_price : {} , sell_cost : {} , commission : {}".format(# order.executed.price,order.executed.value,order.executed.comm)) # def notify_trade(self, trade):# # 一个trade结束的时候输出信息# if trade.isclosed:# self.log('closed symbol is : {} , total_profit : {} , net_profit : {}' .format(# trade.getdataname(),trade.pnl, trade.pnlcomm))# if trade.isopen:# self.log('open symbol is : {} , price : {} ' .format(# trade.getdataname(),trade.price))def get_mtm_atr(df,n1): # 对mtm计算atr指标,在原有数据的基础上,输出增加了一个列,mtm_atr col_list = list(df.columns) df['mtm_l'] = df['low'] / df['low'].shift(n1) - 1 df['mtm_h'] = df['high'] / df['high'].shift(n1) - 1 df['mtm_c'] = df['close'] / df['close'].shift(n1) - 1 df['mtm_c1'] = df['mtm_h'] - df['mtm_l'] df['mtm_c2'] = abs(df['mtm_h'] - df['mtm_c'].shift(1)) df['mtm_c3'] = abs(df['mtm_l'] - df['mtm_c'].shift(1)) df['mtm_tr'] = df[['mtm_c1', 'mtm_c2', 'mtm_c3']].max(axis=1) df['mtm_atr'] = df['mtm_tr'].rolling(n1,min_periods=1).mean() # df = df.dropna() # print(df[df.index>=pd.to_datetime("1990-01-01 09:00:19")]) return df[col_list+['mtm_atr']] def run_not_direct_data(n): data_name = f"data_{n}.csv" df = pd.read_csv(data_name,index_col=0) df.index = pd.to_datetime(df.index) # 计算指标 df = get_mtm_atr(df,20) # print(df) datetime_list = list(df.index) # 添加cerebro cerebro = bt.Cerebro() # 添加策略 cerebro.addstrategy(NotDirectStrategy) # 准备数据 params = dict( fromdate = datetime_list[0], todate = datetime_list[-1], timeframe = bt.TimeFrame.Minutes, compression = 1, ) feed = ExtendPandasFeed(dataname=df,**params) # 添加合约数据 cerebro.adddata(feed, name = "xxx") cerebro.broker.setcommission(commission=0.0005) # 添加资金 cerebro.broker.setcash(100000.0) # 开始运行 cerebro.run()run_not_direct_data(1000) # 直接加载数据到backtrader中,然后计算相应的指标,由于backtrader自带的指标库中没有mtm_atr指标,所以,需要我们自建一个指标class MtmAtr(bt.Indicator): # 需要在lines里面声明指标带的名称,line的名称,可以使用self.lines.xxx或者self.l.xxx或者甚至使用self.xxx lines = ('mtm_atr','mtm_tr') # 可能需要的参数值,可以不需要 params = (("n1",20),) # 可以在init里面计算相应的逻辑,能够在init实现,就可以只使用init,如果在init里面不能够完全实现,那么,就可以考虑使用next和once # 另外,如果想要避免因为数据不足导致计算指标不准,希望等到数据充足之后在计算,可以增加一个self.addminperiod def __init__(self): self.addminperiod(self.p.n1-1) # 保存今日的价格 pre_low = self.data.low(-1*self.p.n1) pre_high = self.data.high(-1*self.p.n1) pre_close = self.data.close(-1*self.p.n1) self.mtm_l = self.data.low/pre_low - 1 self.mtm_h = self.data.high/pre_high - 1 self.mtm_c = self.data.close/pre_close -1 # self.mtm_tr = bt.Max(mtm_c1,mtm_c2,mtm_c3) # 向量法计算会忽略一部分pandas计算的时候时间不够就开始计算平均值的问题,min_periods,和原先的pandas计算相比, # 忽略了一部分数据,直到全部数据够平均才会开始计算 # self.lines.mtm_atr = bt.indicators.SMA(self.mtm_tr,period = self.p.n1) # 保存mtm_tr不为0的bar的根数 self.mtm_tr_bar_num = None def next(self): # 在next中计算的时候,在数据不够的时候,如果求最大值,bt.Max需要三个值都存在了,才不会是nan, # df.max(axis=1)只要有一列不为nan,就会出现不为nan的值,这导致了如果使用bt.Max会和原先pandas计算的结果不一样 # 所以,mtm_c1,mtm_c2,mtm_c3,mtm_atr也需要在向量中计算 # 下面的方法可以和pandas计算的结果一致。 # 不太建议使用这种方法去实现,更好的方法是用上面的向量法 try: mtm_c1 = self.mtm_h[0] - self.mtm_l[0] except: mtm_c1=-np.inf try: mtm_c2 = abs(self.mtm_h[0] - self.mtm_c[-1]) except: mtm_c2=-np.inf try: mtm_c3 = abs(self.mtm_l[0]-self.mtm_c[-1]) except: mtm_c3=-np.inf if mtm_c1!= -np.inf or mtm_c2!= -np.inf or mtm_c3!= -np.inf: mtm_tr_value = self.mtm_tr[0] = max(mtm_c1,mtm_c2,mtm_c3) else: mtm_tr_value = np.NaN if self.mtm_tr_bar_num is not None: self.mtm_tr_bar_num+=1 if self.mtm_tr_bar_num is None and not np.isnan(mtm_tr_value): self.mtm_tr_bar_num = 1 # print(bt.num2date(self.data.datetime[0]),mtm_tr_value,self.mtm_tr_bar_num) if self.mtm_tr_bar_num>=self.p.n1: self.lines.mtm_atr[0]=sum(self.mtm_tr.get(size = self.p.n1))/self.p.n1 if self.mtm_tr_bar_num>0 and self.mtm_tr_bar_num<self.p.n1: self.lines.mtm_atr[0]=sum(self.mtm_tr.get(size = self.mtm_tr_bar_num))/ self.mtm_tr_bar_num class DirectStrategy(bt.Strategy): # params = (('short_window',10),('long_window',60)) params = {"n1":20} def log(self, txt, dt=None): ''' log信息的功能''' dt = dt or bt.num2date(self.datas[0].datetime[0]) print('%s, %s' % (dt.isoformat(), txt)) def __init__(self): # 一般用于计算指标或者预先加载数据,定义变量使用 self.mtm_atr = MtmAtr(self.datas[0],n1 = self.p.n1) def next(self): # Simply log the closing price of the series from the reference # self.log(f"工商银行,{self.datas[0].datetime.date(0)},收盘价为:{self.datas[0].close[0]}") # self.log(f"short_ma:{self.short_ma[0]},long_ma:{self.long_ma[0]}") # 得到当前的size data = self.datas[0] self.log(f"close:{data.close[0]},mtm_atr:{self.mtm_atr[0]}") # def notify_order(self, order):# if order.status in [order.Submitted, order.Accepted]:# # order被提交和接受# return# if order.status == order.Rejected:# self.log(f"order is rejected : order_ref:{order.ref} order_info:{order.info}")# if order.status == order.Margin:# self.log(f"order need more margin : order_ref:{order.ref} order_info:{order.info}")# if order.status == order.Cancelled:# self.log(f"order is concelled : order_ref:{order.ref} order_info:{order.info}")# if order.status == order.Partial:# self.log(f"order is partial : order_ref:{order.ref} order_info:{order.info}")# # Check if an order has been completed# # Attention: broker could reject order if not enougth cash# if order.status == order.Completed:# if order.isbuy():# self.log("buy result : buy_price : {} , buy_cost : {} , commission : {}".format(# order.executed.price,order.executed.value,order.executed.comm)) # else: # Sell# self.log("sell result : sell_price : {} , sell_cost : {} , commission : {}".format(# order.executed.price,order.executed.value,order.executed.comm)) # def notify_trade(self, trade):# # 一个trade结束的时候输出信息# if trade.isclosed:# self.log('closed symbol is : {} , total_profit : {} , net_profit : {}' .format(# trade.getdataname(),trade.pnl, trade.pnlcomm))# if trade.isopen:# self.log('open symbol is : {} , price : {} ' .format(# trade.getdataname(),trade.price)) def run_direct_data(n): data_name = f"data_{n}.csv" df = pd.read_csv(data_name,index_col=0) df.index = pd.to_datetime(df.index) datetime_list = list(df.index) # 添加cerebro cerebro = bt.Cerebro() # 添加策略 cerebro.addstrategy(DirectStrategy) # 准备数据 params = dict( fromdate = datetime_list[0], todate = datetime_list[-1], timeframe = bt.TimeFrame.Minutes, compression = 1, dtformat=('%Y-%m-%d %H:%M:%S'), # 日期和时间格式 tmformat=('%H:%M:%S'), # 时间格式 ) feed = bt.feeds.PandasDirectData(dataname=df,**params) # 添加合约数据 cerebro.adddata(feed, name = "xxx") cerebro.broker.setcommission(commission=0.0005) # 添加资金 cerebro.broker.setcash(100000.0) # 开始运行 cerebro.run() run_direct_data(1000)# 测试运行时间direct_time_list =[]not_direct_time_list =[]bar_num_list =[]for i in range(1,21): bar_num=i*1000 begin_time = datetime.datetime.now() run_direct_data(bar_num) end_time = datetime.datetime.now() consume_time = (end_time-begin_time).seconds direct_time_list.append(consume_time) begin_time = datetime.datetime.now() run_not_direct_data(bar_num) end_time = datetime.datetime.now() consume_time = (end_time-begin_time).seconds not_direct_time_list.append(consume_time) bar_num_list.append(bar_num) # 画图data = [ go.Scatter( x=bar_num_list, y=not_direct_time_list, name = '提前使用pandas计算指标消耗的时间' ), go.Scatter( x=bar_num_list, y=direct_time_list, name = '在backtrader中计算指标消耗的时间' )] layout = go.Layout( title = '随着K线数目增加,两种计算指标的方式消耗的时间') fig = go.Figure(data = data)# 步骤四fig.update_layout( title= '随着K线数目增加,两种计算指标的方式消耗的时间', xaxis_title="bar_num", yaxis_title="消耗时间(s)", # xaxis = {"type":"log"})fig.show()
意料之中的backtrader会跑输,毕竟在next中计算了好多的东西,降低了速度。
智慧、心灵、财富,总要有一个在路上,愿我们能在人生的道路上,不断成长、不断成熟~~~
感兴趣可以关注我的专栏:
my_quant_study_note:分享一些关于量化投资、量化交易相关的思考
backtrader量化投资回测与交易:本专栏免费,分享backtrader相关的内容。
量化投资神器-backtrader源码解析-从入门到精通:本专栏目前收费299元,预计更新100篇策略(更新中)+36篇backtrader讲解(已完成)+backtrader源码分析。
系统当前共有 427 篇文章