If I use IsValidDataPoint in Strategy or Indicator then this causes problems in Strategy Analyzer Optimization.
If I use BacktestType=Standard there is no problem but if I use BacktestType=Optimization then I get the error
This can also cause NT8 to stop (perhaps if the number of combinations to test is too large).
I presume the difference is that BacktestType=Standard probably runs on a single thread whereas BacktestType=Standard uses multiple threads. So perhaps the quoted error is due to NT8 not yet making IsValidDataPoint thread safe?
There are a few problems here:
1) Documentation for MaximumBarsLookBack does not match the current code:
The page currently says
However a simple test shows that is not true (see below).
2)The example code on IsValidDataPoint documentation is silly since it uses SMA which sets Value for every bar. In order to test IsValidDataPoint we need a new SMA indicator that only starts plotting after Period inputs.
Steps to reproduce problem:
a) Edit SMA indicator save as SMATest.
SMATest is minimally changed so it only starts setting Value after it gets Period inputs:
Also put this
in if (State == State.SetDefaults)
b) Edit strategy SampleMACrossOver, save as TestMACrossOver
Change SMA references to SMATest.
Also add a diagnostic Print
Add this line in OnBarUpdate
When you run TestMACrossOver in Strategy Analyzer BacktestType=Standard you see there is no error, also in the Output window you see
Which shows that standard NT8 B11 plots use 256 not infinite as MaximumBarsLookBack documentation says.
Now comment out the Print line.
Run TestMACrossOver in Strategy Analyzer BacktestType=Optimization (Optimizer=Default). Set a small number of combinations (here just 2) say Fast range 2 to 3, Slow range say 4.
On my PC (multi-core) this tends to keep going producing 2 error lines per run
But if I request a larger number of combinations (say fast range=2 to 10, slow range = 20 to 30) then Strategy Analyzer will freeze after a few repeats, probably displaying something like
with nothing happening in progress bar. I then have to press "Abort" button, but Strategy Analyzer has become unusable (have to close window and open new one). If I use a large range and also Optimize Data Series, then NT stops (get Windows message Ninja Trader has stopped working, windows is searching for solution"!). Log records "Unhandled exception: Object not set to an instance of an object".
Restarting NT8:
If I try the same thing but with Optimizer=Genetic, then I see I get 8 lines of the error (probably 1 per thread, or one per core) but Strategy Analyzer does not lock up, I dont have to Abort and I can keep repeating this without NT stopping.
So error handling in the Genetic Optimizer seems to be much better than the Default Optimizer.
If I switch to using SampleMACrossover: I get no operational problems with either Optimizer (there's just the mathematical problem that the built in SMA plots before it has Period inputs).
It looks like the current implementation of IsValidDataPoint has not been tested in a multi-threading context such as Optimization.
If I use BacktestType=Standard there is no problem but if I use BacktestType=Optimization then I get the error
.IsValidDataPoint cannot be used on a series running with MaximumBarsLookBack.TwoHundredFiftySix.
I presume the difference is that BacktestType=Standard probably runs on a single thread whereas BacktestType=Standard uses multiple threads. So perhaps the quoted error is due to NT8 not yet making IsValidDataPoint thread safe?
There are a few problems here:
1) Documentation for MaximumBarsLookBack does not match the current code:
The page currently says
Series<double> objects that hold indicator plot values are always set to MaximumBarsLookBack.Infinite
2)The example code on IsValidDataPoint documentation is silly since it uses SMA which sets Value for every bar. In order to test IsValidDataPoint we need a new SMA indicator that only starts plotting after Period inputs.
Steps to reproduce problem:
a) Edit SMA indicator save as SMATest.
SMATest is minimally changed so it only starts setting Value after it gets Period inputs:
Code:
sum = priorSum + Input[0] - (CurrentBar >= Period ? Input[Period] : 0); [COLOR="Green"]if (CurrentBar+1 < Period) return;//New line. Dont set Value until have Period values in sum[/COLOR] Value[0] = sum / (CurrentBar < Period ? CurrentBar + 1 : Period);
Code:
BarsRequiredToPlot = 0;//Value not plotted until have Period values in sum
b) Edit strategy SampleMACrossOver, save as TestMACrossOver
Change SMA references to SMATest.
Also add a diagnostic Print
Code:
else if (State == State.Configure)
{
smaFast = SMATest(Fast);//changed
smaSlow = SMATest(Slow);//changed
[COLOR="Green"]
BarsRequiredToTrade = 0;
Print(string.Format("smaSlow.MaximumBarsLookBack = {0}",smaSlow.MaximumBarsLookBack));//reports TwoHundredFiftySix[/COLOR]
Code:
[COLOR="Green"]if (!smaSlow.IsValidDataPoint(0)) return;//new line[/COLOR] if (CrossAbove(smaFast, smaSlow, 1))
When you run TestMACrossOver in Strategy Analyzer BacktestType=Standard you see there is no error, also in the Output window you see
smaSlow.MaximumBarsLookBack = TwoHundredFiftySix
Now comment out the Print line.
Run TestMACrossOver in Strategy Analyzer BacktestType=Optimization (Optimizer=Default). Set a small number of combinations (here just 2) say Fast range 2 to 3, Slow range say 4.
On my PC (multi-core) this tends to keep going producing 2 error lines per run
Strategy 'TestMACrossOver': Error on calling 'OnBarUpdate' method on bar 0: .IsValidDataPoint cannot be used on a series running with MaximumBarsLookBack.TwoHundredFiftySix.
calculating Iteration 1/99
Restarting NT8:
If I try the same thing but with Optimizer=Genetic, then I see I get 8 lines of the error (probably 1 per thread, or one per core) but Strategy Analyzer does not lock up, I dont have to Abort and I can keep repeating this without NT stopping.
So error handling in the Genetic Optimizer seems to be much better than the Default Optimizer.
If I switch to using SampleMACrossover: I get no operational problems with either Optimizer (there's just the mathematical problem that the built in SMA plots before it has Period inputs).
It looks like the current implementation of IsValidDataPoint has not been tested in a multi-threading context such as Optimization.
Comment