矢量/ 1D阵列的MATLAB索引约定

Question

Consider the preallocation of the following two vectors:

vecCol = NaN( 3, 1 );
vecRow = NaN( 1, 3 );

Now the goal is to assign values to those vectors (eg within a loop if vectorization is not possible). Is there a convention or best practice regarding the indexing?

Is the following approach recommended?

for k = 1:3
    vecCol( k, 1 ) = 1; % Row, Column
    vecRow( 1, k ) = 2; % Row, Column
end

Or is it better to code as follows?

for k = 1:3
    vecCol(k) = 1; % Element
    vecRow(k) = 2; % Element
end

Answer 1

It makes no difference functionally. If the context means that the vectors are always 1D (your naming convention in this example helps) then you can just use vecCol(i) for brevity and flexibility. However, there are some advantages to using the vecCol(i,1) syntax:

• It's explicitly clear which type of vector you're using. This is good if it matters, eg when using linear algebra, but might be irrelevant if direction is arbitrary.
• If you forget to initialise (bad but it happens) then it will ensure the direction is as expected
• It's a good habit to get into so you don't forget when using 2D arrays

• It appears to be slightly quicker. This will be negligible on small arrays but see the below benchmark for vectors with 10^8 elements, and a speed improvement of >10%.

   function benchie() % Benchmark. Set up large row/column vectors, time value assignment using timeit. n = 1e8; vecCol = NaN(n, 1); vecRow = NaN(1, n); f = @()fullidx(vecCol, vecRow, n); s = @()singleidx(vecCol, vecRow, n); timeit(f) timeit(s) end function fullidx(vecCol, vecRow, n) % 2D indexing, copied from the example in question for k = 1:n vecCol(k, 1) = 1; % Row, Column vecRow(1, k) = 2; % Row, Column end end function singleidx(vecCol, vecRow, n) % Element indexing, copied from the example in question for k = 1:n vecCol(k) = 1; % Element vecRow(k) = 2; % Element end end 

  Output (tested on Windows 64-bit R2015b, your mileage may vary!)

   % f (full indexing): 2.4874 secs % s (element indexing): 2.8456 secs 

  Iterating this benchmark over increasing n , we can produce the following plot for reference.

  

Answer 2

A general rule of thumb in programming is "explicit is better than implicit". Since there is no functional difference between the two, I'd say it depends on context which one is cleaner/better:

• if the context uses a lot of matrix algebra and the distinction between row and column vectors is important, use the 2-argument indexing to reduce bugs and facilitate reading

• if the context doesn't disciminate much between the two and you're just using vectors as simple arrays, using 1-argument indexing is cleaner