Incorrect `_timeframe` after loading Pandas dataframe to a data feed

soulmachine

I've built 1-minute time bars from tick-by-tick trading data of Bitcoin, which is loaded into a Pandas dataframe first, then converted to a data feed using the following code:

class CryptoPandasData(bt.feeds.PandasData):
    lines = ('mean', 'median', 'volume_sell', 'volume_buy',
             'volume_quote', 'volume_quote_sell', 'volume_quote_buy',
             'count', 'count_sell', 'count_buy')
    params = (
        ('datetime', 'timestamp'),
        ('openinterest',None),
        ('mean', -1),
        ('median', -1),
        ('volume_sell', -1),
        ('volume_buy', -1),
        ('volume_quote', -1),
        ('volume_quote_sell', -1),
        ('volume_quote_buy', -1),
        ('count', -1),
        ('count_sell', -1),
        ('count_buy', -1),
    )

Then print data_feed._timeframe:

bt.TimeFrame.Names[data_feed._timeframe]
'Days'

We can see that the timeframe of the data feed is bt.TimeFrame.Days, which is apparently not correct, because my time bars are precisely 1-minute bars.

Any ideas? Thanks!

Full Jupyter notebook: https://github.com/soulmachine/crypto-notebooks/blob/master/backtest/pandas-dataframe-as-a-backtrader-data-feed.ipynb

dasch

When you create the datafeed, you need to specify the timeframe and compression your data is using. Backtrader will not identify it by itself.

So when you do:

 data_feed = CryptoPandasData(dataname=time_bars)

you would do it that way:

 data_feed = CryptoPandasData(dataname=time_bars, timeframe=bt.TimeFrame.Minutes, compression=1)

soulmachine

@dasch Nice, problem solved, thanks a lot!

soulmachine

@dasch What if my bars are volume bars(from Chapter 2 "Financial Data Structures" of the book "Advances in Financial Machine Learning" by Marcos Prado), which don't have a fixed timeframe, use timeframe=bt.TimeFrame.Ticks?

dasch

I am not really sure, since I never used such a usecase. Just give it a try.