How do I make conditional code more efficient

7 次查看（过去 30 天）

William Green 2019-3-25

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/452127-how-do-i-make-conditional-code-more-efficient

编辑： William Green 2019-3-25

I currently have a section of a function that is very slow (when length(data) >> 200). I've been struggling to make use of matlab's speedy matrix manipulation, particularly with the conditional statement that has a variable in the condition (u).

data = cumsum(randi(20,1,200)); % An ordered vector of data with no repetitions
t0 = data(end);
tau = 5; % Some arbitrary num
r = zeros(1,50)
for k = 0:49
    uvec = data(data<= t0 - k*tau - tau); % All data that meets a condition
    inner_int = zeros(1,length(uvec));
    for i = 1:length(uvec)
        u = uvec(i);
        inner_int(i) = sum(data > (k*tau + u) & data <= (k*tau + u + tau) ); % Conditional statement
    end
    r(k+1) = sum(inner_int);
end

I've been tried various solutions including bsxfun, histcounts and using integrals of matlab's dirac function to reduce my number of for loops, but cannot seem to get anything faster than the above.

For reference this is the maths of what I'm implementing

2 个评论
显示无隐藏无

Walter Roberson 2019-3-25

In that context, where you have sum() of the result of logical expressions, then you are asking to count the number of items that match. In that context, nnz() should be faster than sum()

However, it looks to me as if that formula might be using dirac delta, which corresponds to equality tests rather than inequality tests.

William Green 2019-3-25

nnz() seems a little faster, thank you! My main issue is the for loops though.

Yes the formula is using dirac delta, so the inner integral is essentially "How many points in the data are between ktau+u and (k+1)tau+u" and the outer integral is 1 if u is in data (and u < the upper bound). It's similar to binning, but again I'm not really sure how to implement it efficiently

请先登录，再进行评论。

请先登录，再回答此问题。