Optimizing backtest

[Lionel92]

Hello

I am facing poor performance while using the optimzer to back test a strategie (it takes 1 minutes per run for 5 years of data in 1 minute bars. 2000 runs = 1,5 days ...) 4 threads on the computer.

The strategy is complex 2 or 3 time frames and using indicators which can either plot data on a chart or feed dataseries. The strategie itself is not ploting anything.

I have seen in a forum that using indicators with no plot (just data series) could have a huge impact on performance.

could you confirm it's true because creating new versions of all the indicators will take some time

Many thanks in advance

Lionel

[NinjaTrader_Bertrand]

Hi Lionel, but the script would only use minute based data or higher, correct? So no tick based decisions involved here?

Are there any drawings / debug prints done via the strategy or called indicators?

For your custom series, which lookback is being used? Infinite vs 256 would have an impact on performance though most memory usage.

[Lionel92]

The script would only use minute based data or higher, correct?
Yes. In addition "Calculate on bar close is also activated.

Are there any drawings / debug prints done via the strategy or called indicators?
the strategy is not printing anything but the summary of the run in a text file (strategy summary info). no drawing by the strategy.

Indicator used by the strategy are drawing many things if they are added to a chart (same version for chart ploting and strategy). I don't know where they are printing, if they do, when a backtest is run.

This point is really the heart of my question, will i see an improvement if i create new version of indicators dedicated to backtesting (ie with no plot)

For your custom series, which lookback is being used? Infinite no other choice

Lionel

[NinjaTrader_Bertrand]

Hi Lionel, for the direct plots values I would not expect to see a difference then, what would be likely taxing is drawing objects, i.e. a line, dots or rectangles. Those and the custom stats you do I would completely disable in backtesting for now and retest to find the bottleneck.

You could try to force all series to 256 as well and see what outcome you would get in your sceneario - http://www.ninjatrader.com/support/h...htsub=lookback

Same would be for the adding timeframes, try to test how your handling here affects performance, i.e. comment things this way that would have a single series only run for comparison.