Programming with Matlab

• OOP in Matlab, apparently faster nowadays?

> I get results similar to http://stackoverflow.com/questions/1693429/is-matlab-oop-slow-or-am-i-doing-something-wrong. At my old work (2012-2013) it skipped running the JIT compiler on the entire script-only (no parameter passing, all global variables, but several for-loop sections) codebase just because I added one top-level function call that passed in a string. 10X slower, if not more. Nowhere was this documented or indicated in the profiler, which was really frustrating. My boss and I were astonished and then switched to Python/Numpy for further algorithm development since we had a lot of work to do and needed OOP and speed.

• Add JIT information to profiler here

• Yet Another Guide to Matlab Great Google Wiki site for everyone. Haven't gone through it all yet but I'd point people here.

• I learned mostly from looking at example code by Googling things I needed to do. Indexing is a little weird to get used to.
  • Row/column both start at 1 in the top left corner of the matrix.
  • mat[1,:] –> returns first row of matrix
  • mat[1:4,1] –> returns rows 1 thru 4 of column 1

• A' –> A transposed

• Also, a common hitch for operating on matrices is that you usually want to do element by element operations. For example:
  • A*B does matrix multiplication
  • A.*B does each element in matrix multiplied by each other (same size matrix of course)

• If you're doing large data sets, you'll see that any sort of for-looping operation takes a long time! Matlab is optimized for matrix operations and sucks at for loops (something about re-allocating each time). Look up “vectorization” for more where that came from.

• meshgrid is really helpful for generating image/matrix coordinates
• Many are from Richard Socher's website

%Set value at many index pairs in matrix
a = zeros(3,3);
rows = [1 2 3];
cols = [3 1 2];
a(sub2ind(size(a),rows,cols)) = [1 2 3]

%Moving average for large N
y = cumsum(x)/N;
y(N+1:end) = y(N+1:end) - y(1:end-N);

• Legend markers come out as same color:
  • Bug is that matlab burps on plotting matrices!

a = [1,2,3; 4,5,6; 7,8,9];
b = rand(3,3);
c = rand(3,3);

plot(a,b,'b-',a,c,'g-')
%Burps on legend
legend('1','2');

%Reshaping to vectors fixes it
plot(reshape(a,1,[]),reshape(b,1,[]),'b-',reshape(a,1,[]),reshape(c,1,[]),'g-');
legend('1','2');

• Batch Processing. Dealing with reading in all the files in a directory and doing something with them.

• Matlab Tech-Note
  

• Absolute pixel:

get(0,'PointerLocation')

• Relative to axes

a = axes;
get(a,'CurrentPoint');

• Don't use for loops
  • Writing Fast Matlab Code by Pascal Getreuer
• Might need to use C/Mex code (really simple too!)
  • Another excellent PDF by Pascal Getreuer

• Help document here
• Turn off wifi?! Crazy…but it seems to work.