logical indexing possible to increase speed?

Question

Bertil Veenstra 2022-12-31

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1886762-logical-indexing-possible-to-increase-speed

评论： Bertil Veenstra 2022-12-31

In this loop I convert a 26 Hz accelerometer signal to a 5 seconds interval array. The objective is to find the number of acc values for each 5 second interval. The reason for doing this is that the sample frequency is not exaclty 26, and sometimes there is some missing data in the acc signal. The code works, but takes up a lot of time when processing larger files. My feeling/hope is that creating this new vector can be faster using logical indexing. I just can't figure out how.... Any suggestions?

A = accTimeVector; % Timestamps of acc signal in seconds

length = fix(A(length(A))/5);

acc_5s_interval = zeros(1,length); % create vector with 5 sec intervals

j=0;

for i = 1:length(acc_5s_interval)

acc_5s_interval(i) = numel(A(A>= j & A < j+5));

j=j+5;

end

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

George Abrahams 2022-12-31

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1886762-logical-indexing-possible-to-increase-speed#answer_1138992

编辑：George Abrahams 2022-12-31

在 MATLAB Online 中打开

% Dummy timestamps for 32 second, 26 Hz signal starting at 31416 seconds.
signalDuration = 32;
hz = 26;
startTime = 31416;
A = (0:1/hz:signalDuration) + startTime;
n = length( A );
% Add dummy noise to timestamps.
rng( 1, "twister" )
noise = rand(1,n) * 1/(hz*2);
A = A + [ noise(1:end-1) min(0,noise(end)) ];
% Randomly remove 10% of readings.
keepIdx = randperm( n, floor(n*0.9) );
A = A( sort(keepIdx) );
% Remove all readings during the 2nd second.
A( A>=A(1)+1 & A<=A(1)+2 ) = [];
% Time intervals start from the first timestamp (rather than 0).
% Adding Inf to the edges means that we also count any fractional
% intervals at the end of the data, e.g. the last 0.5s in 10.5s of data 
% measured every 1s.
measurementInterval = 5;
edges = seconds( [ A(1):measurementInterval:A(end) Inf] );
histcounts( seconds(A), edges )
ans = 1×7
    93   117   112   113   122   124    45

I measured computation time for a 1 hour, 30 Hz, noiseless signal, measured at a time interval of 1 second. On my laptop, your for loop method took 123 milliseconds and the histcounts method took 0.5 milliseconds.

EDIT: Updated example to match question's 5 second interval. Added speed comparsion.