Resample/Replay for custom (extended) Feed

alexgorbachev

I extended Pandas feed to add a volume weighted average. I'm feeding 5 seconds bars and want to resample (replay, actually), 1 minute bars along with 5 seconds. However, cerebro.resampledata() only produces OHLC columns... Where do I add custom code to for an extra column/line aggregation? I saw a hint to add another filter on top of Replayer or Resampler and I could add a dummy filter just to see if it work (and it does - it's call method does execute and gets resampled 1m bar) but I can't figure out how to add another line/column to that feed.

run-out

@alexgorbachev Can you share your code please?

vladisld

@alexgorbachev I think it was already discussed here:
https://community.backtrader.com/topic/2812/resample-for-extra-information-in-lines

alexgorbachev

@vladisld Thanks for the reference. I may give it a try to create a custom Replayer then. Resampling workaround is easy - I can just resample myself and feed as a normal feed but I can't do that to replay incomplete bars...

alexgorbachev

@vladisld: Thanks so much for your help! I spend few hours trying to find examples until you pointed me to that post.

So here is what seems to work but I have couple questions if you could confirm... See the questions in the code comments.

class PandasDataVW(PandasData):
    """Pandas feed with additional columns `vw` for volume weighted average."""

    lines = ('vw',)
    params = (('vw', -1),)

class _MyBar(bt.dataseries._Bar):
    def bstart(self, maxdate=False):
        super().bstart(maxdate)
        self.vw = float('NaN')
        self.vw_vol = 0

    def bupdate(self, data, reopen=False):
        return_value = super().bupdate(data, reopen)
        self.vw_vol += data.vw[0] * data.volume[0]
        self.vw = self.vw_vol / self.volume if self.volume else data.vw[0]
        return return_value

class MyReplayer(bt.Replayer):

    def __init__(self, data):
        super().__init__(data)
        self.bar = _MyBar(maxdate=True)

cerebro.addstrategy(TestStrat)
data0 = PandasDataVW(dataname=df, timeframe=bt.TimeFrame.Seconds, compression=5)
cerebro.adddata(data0, name='s5')
# For some reason cloning doesn't work here - DataClone doesn't have my `vw` line
# so I can recreate the feed. I assume I shouldn't add filter to the same feed directly
# because I'm already using it as is. Do I do something wrong with cloning?
# data1 = data0.clone()
data1 = PandasDataVW(dataname=df, timeframe=bt.TimeFrame.Seconds, compression=5)
data1.addfilter(MyReplayer,
                timeframe=bt.TimeFrame.Minutes,
                compression=1)
cerebro.adddata(data1, name='m1')
# `_doreply=True` I saw in cerebro.resampledata() along with cloning so I replicated it.
# Is this required?
cerebro._doreplay = True
cerebro.run(preload=False)

Hope this comes useful to others.

invoker

@alexgorbachev at last ,it prove it is works,but there some thing to say,if your extend num>=2,
the extend in bstart and bupdate must be define and inplement by strict sort,it seem the backtrader recoginze them by order and not name