Allocating data into a structure faster

Question

Chris 2011-9-9

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/15527-allocating-data-into-a-structure-faster

I'm trying to improve the efficiency of my code. Essentially I'm importing data into matlab every couple seconds, the data is a cell array with 3 columns for each data point [time, name, data], like this:

A =    [7.3468e+005]    'M001'    [0.5000]
  [7.3468e+005]    'M016'    [0.0861]
  [7.3468e+005]    'M002'    [0.2693]
  [7.3468e+005]    'M009'    [0.7381]
  [7.3468e+005]    'M004'    [    10]

And I want to organize it based on the data names. Currently I'm using a for loop and dynamic field names to put the data into a structure like this:

for xx = 1:1:A_length
    data.(A{xx,2}).time = [data.(A{xx,2}).time([2:2500]); A{xx,1}];
    data.(A{xx,2}).data = [data.(A{xx,2}).data([2:2500]); A{xx,3}];
end

So the newest data is always at the end and the structure is always preallocated to 2500 data points.

I'm looking for ideas to speed this up. I was hoping to avoid the for loop all together, like this:

      data.(A{:,2}).time = [data.(A{:,2}).time([2:2500]); A{:,1}];
      data.(A{:,2}).data = [data.(A{:,2}).data([2:2500]); A{:,3}];

But I've been told that's a no go. Any ideas?

2 个评论
显示无隐藏无

Fangjun Jiang 2011-9-9

Do you have freedom to re-design your structure? I would use A.M016=[7.3468e+005,0.0861]; for example to reduce the level of structure.

Chris 2011-9-15

I might be able to, I'd have to look into it. But even if I could put A into a better format I still have to sort it into the data structure using a for loop.

I'm not sure that

data.([data_name]).time = A.([data_name])(2);

is any faster than

data.([data_name]).time = A{xx,3};

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Jan 2011-9-9

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/15527-allocating-data-into-a-structure-faster#answer_21109

在 MATLAB Online 中打开

Under MATLAB 2009a time([2:2500]) is 10 times slower than time(2:2500). Actaully [a:b] creates the vector a:b and converts it to a vector again, which does not have such a big overhead. But internally time(a:b) checks only for a and b if they exceed the dimensions, and for time([a:b]) this check is performed for each element of the index vector.

Another explanation: in time([a:b]) the vector a:b is created explictely, but in time(a:b) it is not.

Getting A{xx,2} repeatedly wastes time. Better create a temporary variable:

for xx = 1:A_length
  tmp = A{xx,2};
  data.(tmp).time = [data.(tmp).time(2:2500); A{xx,1}];
  data.(tmp).data = [data.(tmp).data(2:2500); A{xx,3}];
end

Nevertheless, using a different data structure would be much faster, e.g. collecting time and data in a matrix and the names 'M001' etc in a cell string. Then the vectorization would be possible and easy.