conversion from mathematica to matlab --> (appendto)

Question

i have the following in mathematica and want to use it in matlab.I tried but i have mistakes and can't fixed them.It is that i don't get yet the matlab philosophy! So,

intMC = {}; sigmat = {};
Do[np1 = np + i*100;
  xpoints = Table[RandomReal[], {z1, 1, np1}];
  a1t = Table[f[xpoints[[i2]]], {i2, 1, np1}];
  a12 = StandardDeviation[a1t]/Sqrt[Length[a1t]];
  AppendTo[intMC, {np1, Mean[a1t], a12}];
  AppendTo[sigmat, {np1, a12}],
  {i, 1, ntr}];

I did this:

  fx=@ (x) exp(-x.^2);

    intmc=zeros();
    sigmat=zeros();

    for i=1:ntr
        np1=np+i*100;
        xpoints=randn(1,np1);
        for k=1:np1
        a1t=fx(xpoints(k))
        end                   %--> until here it prints the results,but in the 
                              %end it gives 
                              % me a message " Attempted to access xpoints(2,:);
                              %index out of bounds because size(xpoints)=[1,200]
                              %and stops executing.

    %a1t=fx(xpoints(k,:))    %  -->I tried this instead of the above but
    %a1t=bsxfun(@plus,k,1:ntr)   % it doesn't work

        a12=std(a1t)/sqrt(length(a1t))
        intmc=intmc([np1 mean(a1t) a12],:) %--> i can't handle these 3 and
        sigmat=sigmat([np1 a12 ],:)        %as i said it stopped executing

    end

score 2 · Accepted Answer · answered Jan 25 '11 at 16:18

2

In order to append a scalar to a Matlab array, you can call either array(end+1) = value or array = [array;value] (replace the semicolon with a comma if you want a 1-by-n array). The latter also works for appending arrays; to append arrays with the former, you'd call array(end+1:end:size(newArray,1),:) = newArray in case you want to catenate along the first dimension.

However, appending in loops that do more than, say, 100 iterations, is a bad idea in Matlab, because it is slow. You're better off pre-assigning the array first - or even better, vectorizing the computation so that you don't have to loop at all.

If I understand you correctly, you want to calculate mean and SEM from a growing number of samples from a normal distribution. Here's how you can do this with a loop:

intmc = zeros(ntr,3); %# stores, on each row, np1, mean, SEM
sigmat = zeros(ntr,2); %# stores, on each row, np1 and SEM

for i=1:ntr
%# draw np+100*i normally distributed random values
np1 = np+i*100;
xpoints = randn(np1,1);
%# if you want to use uniform random values (as in randomreal), use rand
%# Also, you can apply f(x) directly on the array xpoints

%# caculate mean, sem
m = mean(xpoints);
sem = std(xpoints)/sqrt(np1);

%# store
intmc(i,:) = [np1, m, sem];
sigmat(i,:) = [np1,sem];

end %# loop over i

answered Jan 25 '11 at 16:18

Jonas

74,690
10
137
177

Thanks for your answer Jonas!You were very helpful.I can't understand one part though.How we apply directly f(x) on the array xpoints.I can't understand why we don't use sth like :fx(xpoints(k)). – George Jan 25 '11 at 16:45
@George: xpoints is an array. Mathematical operations are applied element-wise on an array (note that if an operation has a special meaning lin linear algebra, such as multiplication, you need to use `.*` instead of `*` to get element-wise behavior). `exp(-x.^2)` works on each element of the array, so you can just pass an array for `x`, and don't need a loop. – Jonas Jan 25 '11 at 16:51
yes,but in order for exp(-x.^2) to work on each element of the array xpoints,don't we have somehow to put the exp(-x.^2) on the array?Instead,we are just manipulating xpoints.Where is the fx to take action on the xpoints?Sorry if i can't be understood.. – George Jan 25 '11 at 17:00
@George: You can either write `exp(-xpoints.^2)`, or you can write `f=@(x)exp(-x.^2);` and then call `f(xpoints)`. – Jonas Jan 25 '11 at 17:33
Thats right.In the code you gave me there is nowhere this.It's missing a line? – George Jan 25 '11 at 17:38
@George: In the Mathematica code, you create a uniform random variable and then transform it with f(x). In my code, I generate normally distributed variables directly. To reproduce the Mathematica code, create `xpoints` by calling `rand`, and then apply `f` to it. – Jonas Jan 25 '11 at 17:44
Alright ,i did this and i think its ok : a1t=fx(xpoints(1:np1)) , m = mean(a1t); sem = std(a1t)/sqrt(length(a1t)) – George Jan 25 '11 at 17:55
@George: Yes, that looks fine. Though you don't need to specify `1:np1` when passing `xpoints`, since `xpoints` has `np1` elements, anyway. – Jonas Jan 25 '11 at 17:56
Hello, how can i plot " sigmat" (simple plot)?And also "intmc" with errorbars?Because i can't figure a way to handle plot.I have problems with the dimensions. – George Jan 26 '11 at 12:56
@George: It's easiest to just call `plot(sigmat)`. And for errobars, you do `plot(intmc),hold on,errorbar(intmc,sigmat)` – Jonas Jan 26 '11 at 13:58
If i do "plot(sigmat)"it gives me wrong results because it gives as x axis the ntr=20points and as y axis the np1 points.I want as x axis the ntr points and as y the value of sigmat.---- As for errorbar(intmc,sigmat) it gives me error ->X, Y and error bars must all be the same length – George Jan 26 '11 at 14:05
@George: Ah, yes. I forgot they weren't vectors. `plot(sigmat(:,1),sigmat(:,2))` and `plot(intmc(:,1),intmc(:,2)),hold on, errobar(intmc(:,1),intmc(:,2),intmc(:,3))` – Jonas Jan 26 '11 at 14:11
The first worked great!But at the plot woth the errorbar ,it gives me the right plot(i used 'r.',to plot points) but without the error bars.And it gives me a message "Undefined function or method 'errobar' for input arguments of type 'double'." – George Jan 26 '11 at 14:20
@George: Yes, it's a typo. It should say `errorbar`. I strongly suggest you search the documentation for `plot` and `errorbar` to find out more about the commands. – Jonas Jan 26 '11 at 15:20
It seems to be ok though...Anyway ,i will search it more.You have helped me a lot Jonas and i thank you very much! – George Jan 26 '11 at 15:40
Ok,it was typing ! I wrote errobar instead of errorbar.Now,it works fine!Thanks again Jonas! – George Jan 26 '11 at 15:47
@Jonas: Jonas,if you could help me with this http://stackoverflow.com/questions/5032029/map-command-and-few-other-from-mathematica-to-matlab because i couldn't found a solution yet... – George Feb 22 '11 at 09:50

conversion from mathematica to matlab --> (appendto)

1 Answers1