Hi @jeff-lee

I had the issue

min() arg is an empty sequence

when I forgot during resampling to specify the format of the original (not resampled) data.

So I did not write:

originalData = bt.feeds.PandasData(dataname=df, timeframe=bt.TimeFrame.Minutes, compression=1, datetime='Date', fromdate=startDate, todate=endDate, )

but:

originalData = bt.feeds.PandasData(dataname=df, # timeframe=bt.TimeFrame.Minutes, # compression=1, datetime='Date', fromdate=startDate, todate=endDate, )

Under which conditions did you get the error that motivated you to do your bug fix?

@leggomaeggo I have tried this, still this problem, when there is a large dataset

@vladisld Thanks. That makes sense. Much cleaner to do it in __init__().

@run-out In fact, never mind, managed to figure it out a different way using the bt docs. Thanks for the help!

I run my own optimization outside of backtrader. You could have a look at the run_backtest and backtest_controller_multi methods.

https://github.com/neilsmurphy/backtrader_template/blob/main/main.py

@run-out Yup you are right, it is trading in decimals which is not possible. Let me do something to fix this.

@braan Not sure about that.

@kjiessar
Aaand again. After long debugging session I found the culprit and it wasn't even in the code I had shown.

Errorness:

self.sell(data=self.data0, exectype=bt.Order.Limit, price=self.datas[0].close, size=entry["size"], valid=datetime.timedelta(minutes=self.p.tis))

correct one:

self.sell(data=self.data0, exectype=bt.Order.Limit, price=self.datas[0].close[0], size=entry["size"], valid=datetime.timedelta(minutes=self.p.tis))

Note the proce parameter and the last [0]

@kjiessar
Its a bit embarassing to say but I found the issue

data = store.getdata(dataname=symbol, name=symbol, timeframe=bt.TimeFrame.Minutes, compression=1, ohlcv_limit=70 ) #fromDate is missing

I had the parameter but with the wrong name, becuase of all the dynamic handling of the parameters, no error was raised.

Sorry for the trouble.

@dasch said in data feeds with different start date:

you could provide some code to see what issue you have, its better than just guesswork what your issues may be.

The trade just happened, but no plot.
I think it is too little data in C98, if I change to another data not so little (with BTC data 90% of the length), then the graph will show normal. There should be some problem with the logic of plot.
You can try to reproduce this whit data feeds of different start time. But need to override the prenext method to make trade happen when there is only one data feeds have valid data

@frenzyy Please disregard - found a solution here: https://community.backtrader.com/topic/3797/opening-range-breakout-strategy-for-multiple-stocks

@anarchy89 Did you run it?

@backtrader said in optimization on Strategy Selection example:

You then need to have 5 optimizations (the sample shows 2) and the StFetcher instantiating the appropriate strategy with the appropriate parameters. In the example idx is used as the key to decide which strategy to return. The same idx can be used to decide which strategy to instantiate and with which parameters.

Can you give me an example code? thank you

@longhngo This solved the problem in my case

@anarchy89

The standard TimeReturn analyzer has been extended to support tracking a data feed. The 2 major parameters invoved: timeframe (default: None) If None then the complete return over the entire backtested period will be reported Pass TimeFrame.NoTimeFrame to consider the entire dataset with no time constraints data (default: None) Reference asset to track instead of the portfolio value.

So the top analyzer tracks the return for the specific data and gives it the name datareturns while the next line tracks the return for the whole portfolio and gives it the name timereturns, at least that's how i'm reading it.

Wow, thanks so much @run-out !

I found the problem in data sample. Because the sample is not big enough for a indicator parameter. I used a indicator HullMovingAverage with a big period, so with the small timeframe it is normal, but in the large timeframe it need a bigger size of input data.

That's why cerebro plots running result, but not print log.

P/S: I will add entire code next question. Thanks for your advice!

Thanks @vladisld so much!

I read your guide very carefully. And I think there is only one way to do:

Re-implementing _Bar class. Re-implementing Resample/Replay class to use new _Bar class. Then re-implementing resample/replay in the feed class to use customized Resample/Replay.