Multiday simulation checkpoint & possibility to run failed test from the last successful checkpoint

Imagine that I have a multiday simulation which has 300 days and takes 24 hours to complete. And somehow it failed after 12 hours.
Is it possible to create a mechanics which allows me to run backtest from the last successful checkpoint to save computational time?



  • ptunneyptunney Posts: 246

    it is not a bad idea, for a parallelized backtest it is possible, for a single multiday job it would not be possible.
    Even though it is possible, it would take some work, to re-run the failed days and combine the results.
    Also it would have to just re-submit the same script, you could not edit to fix issues for particular days, so it would only be useful when the failure was at our end, which we always aim to keep to a minumum.
    I will pass the suggestion to the head of development.

  • The idea was for the multiday backtests. For the parallel one, it is much easier to just re-submit the backtest, because it is already pretty fast. Things are different with multiday backtests. One case in which it constantly fails -- using vwap/twap for execution.

  • ptunneyptunney Posts: 246

    pp794378 , please PM me or email me at about the VWAP/TWAP issue. If you can pin it down to a specific date, symbol order and reproduce with a simple backtest I can have the devs dig into it.

Sign In or Register to comment.