How to determine if Optimizer stuck?

[sajosh]

I have 8 strategy analyzer windows running batch optimizations on the same 6 stocks each. This has been running for over 24 hours now.

How can I determine if it's all still running ok?

It's running the brute force Opt on 2 variables each with 100 range in 1 increments. The strategies are simple crossovers. The Windows task manager shows NT process fluctuating around the 90% CPU usage. The Running Optimizer Elapsed has been at 01:02:49 / Remaining 00:00:00 for all of today. The log tab shows nothing. All the SA windows say it's optimizing. I looked in all folders for any temporary files with current times but didn't find any in Program Files\NT or Documents\NT. Next time I'll put print().

I have Vista, Intel i7 with 4 processors, 4Gb RAM, plenty of HD space, and nothing else is running on that machine.

Is it possible that they could be colliding in trying to access the same data? Could there be corruption?

[NinjaTrader_JoshP]

sajosh,

If your CPU is still churning along then it is still thinking. Unfortunately since you are running so many at the same time it is very hard to know when exactly they will all finish. We do not recommend testing it in this fashion since if there were any errors it would take a long time for you to get any feedback from it.

[sajosh]

The SA was stuck. I let it run for 72 hours with no change other than the cpu usage flucutated. So I closed (x) the last SA (#8) and all the others unfroze. The results were garbage and there was a lot of cross SA curruption.

So lesson here is not to run multiple SA optimizations.

I ran a few sequentially and it runs fine. It's just parallel doesn't run fine. It would be nice to be able to queue up the SA so they run after another. It would be great to let them run unattended overnight or when away.

[NinjaTrader_JoshP]

Thank you for the suggestion. I have added it to our feedback list as item #760.

[sajosh]

Another suggestion for future release is to show results in the Instrument and Optimizer tabs. This has two benefits. One, we know that the process is or isn't stuck. Two, we can see some preliminary best results and determine if the code is working correctly.

[NinjaTrader_JoshP]

Thanks sajosh. Being able to view results while they are still forming will be added into our feedback system as #791.

[billr]

I would like to add my vote to an indicator that would let us know the SA is stuck. Mine usually runs one test quickly (suspiciously quick sometimes). The next run sticks. Every time. And it won't stop after closing NT. I have to force quit process.

Suggestion: Why not build a more robust backtesting program. Fix tick driven charts, handle data more accurately using time stamped tick, make it more reliable, etc. i.e make it useful. Charge extra. I would certainly pay for a working version. (why wait for Multi Charts to beat you to it?)

[NinjaTrader_JoshP]

billr,

Do you have a reproducible scenario? We will need one to be able to analyze what is happening on your particular install.

I am not sure what you mean by "fix tick driven charts". Data is already handled by the timestamp of the tick. The issue with timestamps is that they simply do not have enough granularity resulting in the possibility of many ticks with the same timestamp.

[billr]

Thanks for your response Josh.

I will try to reproduce it without having to shut down. If not, maybe I'll let it crash and send you the log/trace afterwards. In this case, when I think I have shut down, NT is usually still running and I have to shut it down via Task Mgr. Should I send you something prior to End Task or after? (hope that makes sense)

Regarding my whining on the backtesting, my main problem is with how NT handles Renko charts. It is so fraught with design issues it should not really be used. I understand Renko charts and how they cause problems with your backtesting. But you have similar problems with backtesting intrabar executions as well. When I mentioned the time stamp, what I meant was allowing the live or replay sequence of ticks to be recognized inside of a bar. My understanding is you only chart using OHLC and there is no recognition of sequence between these points and any entry/exit occurring in a bar is subject to a rules based, artificial sequencing vs. the actual sequence of ticks. I may not have this exactly right, but I do know that many of the backtesting assumptions 'inserted' into the building of charts for backtesting biases the results positive. The extent of this bias for Renko is really discouraging, if not dangerous, for the novice. I currently use a work-around with a custom programmed renko, but its use really slows NT down and may have something to do with my SA getting stuck.

[NinjaTrader_JoshP]

Backtesting only has the granularity of the series you are testing based on the known data points of OHLC. If you wanted more granularity you needed to make the strategy a multi-series strategy that utilizes a higher granularity resolution to get the information you are looking for. There is unfortunately no way around this and no product out there can do any better. This includes MultiCharts. Consider 1min bars. There is no way NT or any platform will ever know how exactly that 1min bar was built with intrabar price movement. Only way to know is to add more resolution which can be done through additional series. To do this in NT please see here: http://www.ninjatrader.com/support/f...ead.php?t=6652