Function ecdf break down for large datasets

Question

Martin 2011-2-24

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1988-function-ecdf-break-down-for-large-datasets

Hi,

I have a very large vector x (around 130 million elements). When I try to find the empirical cumulative distribution function of the values from that vector using MATLAB's command "ecdf(x)" the function breaks down. Its plot shows the ECDF for only the smaller values of x and doesn't even exist for bigger values of x. When I try to run the ecdf command on only a part of the vector (say 10 million elements), the results seem OK. Does anyone know what could be wrong with the ecdf function so that it breaks down in this manner for very large datasets?

Thank you very much for you help.

Martin

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Martin 2011-3-8

Is there anyone who can help me with this issue? Thanks.

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Mathieu Boutin 2011-9-8

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/1988-function-ecdf-break-down-for-large-datasets#answer_20920

在 MATLAB Online 中打开

Hi Martin. You could try my new homemade function and see if it works fine:

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

function [v_f,v_x] = homemade_ecdf(v_data)
    nb_data = numel(v_data);
    v_sorted_data = sort(v_data);
    v_unique_data = unique(v_data);
    nb_unique_data = numel(v_unique_data);
    v_data_ecdf = zeros(1,nb_unique_data);
    for index = 1:nb_unique_data
        current_data = v_unique_data(index);
        v_data_ecdf(index) = sum(v_sorted_data <= current_data)/nb_data;
    end
    v_x = [v_unique_data(1) v_unique_data];
    v_f = [0 v_data_ecdf];
end