MathGroup Archive 2011

[Date Index] [Thread Index] [Author Index]

Search the Archive

within ROOT(similar with c++), put big data into mathematica then get

  • To: mathgroup at smc.vnet.net
  • Subject: [mg115663] within ROOT(similar with c++), put big data into mathematica then get
  • From: junhui Liao <junhui.liaoforothers at gmail.com>
  • Date: Tue, 18 Jan 2011 05:46:40 -0500 (EST)

Dear all,

I want to interpolate my experimental signals by Mathematica "Sinc"
function.
My data is in ROOT(OO language similar with C++, actually combined C+
+).
There are lots of signals(20k+), and each signal has 4k points. What I
want to
do is to each signal, do interpolation so I can increase it to 4k*n
points(n= ~20).

After reading some documents, I still not sure how to organize it.
This following is get from mathematica example.

############# Original Mathematica code (Contributed by: Nasser M.
Abbasi) #################

(* This is the sinc interpolation formula *)

sincInterpolate[Tau_, T_, samples_] := Module[{k},
   Sum[samples[[ k ]]*Sinc[Pi/T (Tau - (k - 1) T)], {k, 1,
     Length[samples]}]
   ];
(* Do the sinc interpolation sum *)
xa = Table[{\[Tau],
    sincInterpolate[Tau, T, samples]}, {Tau, 0, maxt,
    sincDeltaTime}
   ];
########## End of original Mathematica code ############


Following is my idea, please comment it OK or not.
########## My idea on C++ code ##################
Within ROOT(C++).

// There are(at least) two strategies to do my interpolation.

//1st strategy, put the original data input Mathematica then get
results(interpolated ) back to ROOT.

SincInterpolate(double* time, double* amp, double*& time_interpolated,
double*& amp_interpolated) {
//link, env, arg define .....
// Put each signal's 4k points into Mathematica
MLPutFunction(lp, "Tranpose", 1);
    MLPutFunction(lp, "List", 2);
         MLPutReal64List(lp,time,n+1); // n = length of input array
(list).
         MLPutReal64List(lp, amp, n+1);
MLEndPacket(lp);

MLPutFunction(lp, "EvaluatePacket", 1);
MLPutFunction(lp, "ToExpression",1);

//*****Don't know how to code this part yet, right now :-).
(* Do the sinc interpolation sum *)
xa = Table[{\[Tau],
    sincInterpolate[Tau, T, samples]}, {Tau, 0, maxt,
    sincDeltaTime}
   ];
(* This is the sinc interpolation formula *)
MLPutFunction(lp, "sincInterpolate[Tau_, T_, samples_] := Module[{k},
   Sum[samples[[ k ]]*Sinc[Pi/T (Tau - (k - 1) T)], {k, 1,
     Length[samples]}]",3);

//End of *****

MLGetReal64List(lp, time_interpolated);
MLGetReal64List(lp, amp_interpolated);



// 2nd strategy, make a few loops in ROOT, to each element, call
mathematica function only when needed.
for(i = 0; i< 4k; i++) {
    for(j = 0; j<J ; j ++) { // J = constant ~= 20.
          for(k = 0; k< 4k; k++) {
               amp_interpolated = amp[i]*Sinc((time[i]+T*j/J-time[k])/
T); //J = constant, Sinc is Mathematica funciton ;
               push_back [amp_interpolated];
          }
    }
}
But in this case, the loop is huge, 4k*20*4k = 3.2e8. And there are 4k
vectors and each vector with length = 4k* 20 = 8.2e4.
And these data need to transfer from Mathematic to ROOT. Plus, this is
only one signal, I totally have 20k+ signals.
So, I worried a little bit this could be very low efficiency.


Question 1,  I don't know which strategy is better( in efficiency) and
is there any better one than these two(like code directly in c++)?
Question 2,  Could you show some close source data or give some
directive on how to code the mathlink part?

Many thanks in advance!
Junhui


  • Prev by Date: Re: question about CUDA
  • Next by Date: Re: question about CUDA
  • Previous by thread: Re: Problems Exporting to PDF
  • Next by thread: Do I need MathLink to run finite-difference fast enough for Manipulate?