Number of realizations in NExpectation, Method Monte Carlo

Question

I have been looking in NExpectation about how large the sample size is when Mathematica performs Monte Carlo simulation. For example, if we write

NExpectation[x Sin[x] + 1, x \[Distributed] NormalDistribution[], Method -> "MonteCarlo"]

which is the number of realizations of $x$ that Mathematica makes?

Also, I do not understand well how you can specify the number of realizations you want. For instance, if I want $1000$ realizations of $x$, what should I write?

If you have looked at the documentation, you already know that NExpectation does integration of expression if the distribution is continuous, or discrete sum if this is a discrete one. In your case, it will do integration of $(x \sin(x)+1)$ with the NormalDistribution — José Antonio Díaz Navas, Dec 19 '17 at 08:58
@JoséAntonioDíazNavas But Monte Carlo is based on sampling: http://mathworld.wolfram.com/MonteCarloIntegration.html. I would like to know which $N$ Mathematica takes. — user54284, Dec 19 '17 at 09:01

score 1 · Answer 1 · answered Mar 19 '18 at 11:52

The number of samples taken by the "MonteCarlo" Method of NExpectation[] is controlled through an undocumented option. To see what happens, let's use a function with Sow[]:

f[x_?NumericQ] := (Sow[x]; x Sin[x] + 1)

and then evaluate

tst = Reap[NExpectation[f[x], x \[Distributed] NormalDistribution[], 
                        Method -> "MonteCarlo"]];

We then find that

Length[tst[[-1, 1]]]
   1012001

Now, let's adjust that undocumented option to use $10^7$ samples:

tst2 = Reap[NExpectation[f[x], x \[Distributed] NormalDistribution[], 
                         Method -> {"MonteCarlo", "SamplingIncrement" -> 1*^7}]];

and we find that

Length[tst2[[-1, 1]]]
   10012001

so, $10^7$ samples plus a little extra was used.

To see all the other adjustable options:

 Options[Statistics`Library`NExpectationMonteCarloMethod]
   {PrecisionGoal -> 2, AccuracyGoal -> 3, MaxIterations -> 50000, 
    "RandomSeed" -> Automatic, EvaluationMonitor -> None, 
    ConfidenceLevel -> 19/20, WorkingPrecision -> MachinePrecision, 
    "ReportingMethod" -> Automatic, "SamplingIncrement" -> 1000000}

Anton Antonov · Answer 2 · 2018-03-19T15:16:52.477

You can use NIntegrate and specify for the method "MonteCarlo" the maximum number of points:

NIntegrate[(x*Sin[x] + 1) PDF[NormalDistribution[], 
   x], {x, -Infinity, Infinity}, 
 Method -> {"MonteCarlo", "MaxPoints" -> 999}]

(* 1.63975 *)

This shows that 1000 points were used:

res =
  Reap@NIntegrate[(x*Sin[x] + 1) PDF[NormalDistribution[], 
      x], {x, -Infinity, Infinity}, 
    Method -> {"MonteCarlo", "MaxPoints" -> 999}, 
    EvaluationMonitor :> Sow[x]];
Length@res[[2, 1]]

(* 1000 *)

Also, you might want to take a look into this answer of "Monte Carlo integration with random numbers generated from a Gaussian distribution".

Picaud Vincent · Answer 3 · 2017-12-19T09:44:26.140

0

Looking at official doc and NExpectation options:

Options[NExpectation]

{AccuracyGoal -> \[Infinity], Compiled -> Automatic, 
 Method -> Automatic, PrecisionGoal -> Automatic, 
 WorkingPrecision -> MachinePrecision, TargetUnits -> Automatic}

it is true that we do not find how to have a fine control (number of realizations etc.).

Concerning the number of realizations, here 100, you can use this trick:

NExpectation[Block[{$IterationLimit = 100}, x*Sin[x] + 1], 
 x \[Distributed] NormalDistribution[], Method -> "MonteCarlo"]

1.60641

to be compared with the more accurate result:

1.60624

edited Dec 19 '17 at 09:44

answered Dec 19 '17 at 09:24

Picaud Vincent

2,463
13
20

1

Since NExpectation does not hold its arguments, Block[{$IterationLimit = 100}, x*Sin[x] + 1] evaluates to x*Sin[x] + 1 before it is passed to NExpectation, and the change to $IterationLimit is lost before it can have any effect. (I don't think $IterationLimit works in the way implied here. I'd be surprised if it is used in any other way than what is described in this tutorial.) – Michael E2 Mar 19 '18 at 12:36

Number of realizations in NExpectation, Method Monte Carlo

3 Answers3