www.gusucode.com > datafun 工具箱matlab源码程序 > datafun/rmmissing.m
function [B,I] = rmmissing(A,varargin)
%RMMISSING Remove rows or columns with missing entries
% First argument must be numeric, datetime, duration, calendarDuration,
% string, categorical, character array, cell array of character vectors,
% a table, or a timetable.
% Standard missing data is defined as:
% NaN - for double and single floating-point arrays
% NaN - for duration and calendarDuration arrays
% NaT - for datetime arrays
% <missing> - for string arrays
% <undefined> - for categorical arrays
% blank character [' '] - for character arrays
% empty character {''} - for cell arrays of character vectors
%
% B = RMMISSING(A) removes missing entries from a vector, or rows with
% missing entries from a matrix or table.
%
% B = RMMISSING(A,DIM) reduces the size of A along the dimension DIM.
% DIM = 1 removes rows, and DIM = 2 removes columns with missing entries.
% If A is a table, DIM = 2 removes table variables. By default, RMMISSING
% reduces the size of A along its first non-singleton dimension: it
% removes rows from matrices and tables.
%
% B = RMMISSING(A,...,'MinNumMissing',N) removes rows (columns) that
% contain at least N missing entries. N must be an integer. By default,
% N = 1.
%
% [B,I] = RMMISSING(A,...) also returns a logical row (column) vector I
% indicating which rows (columns) of A were removed.
%
% Arguments supported only for table inputs:
%
% B = RMMISSING(A,...,'DataVariables',DV) removes rows according to
% missing data in table variables DV. The default is all table variables
% in A. DV must be a table variable name, a cell array of table variable
% names, a vector of table variable indices, a logical vector, or a
% function handle that returns a logical scalar (such as @isnumeric).
%
% Examples:
%
% % Remove NaN entries from a vector
% a = [NaN 1 2 NaN NaN 3]
% b = rmmissing(a)
%
% % Remove only rows that contain at least 2 missing entries
% A = [NaN(1,3); 13 1 -20; NaN(4,1) ones(4,2); -1 7 -10; NaN(1,3)]
% B = rmmissing(A,'MinNumMissing',2)
%
% % Remove table rows that contain standard missing data
% v1 = {'AB'; ''; ''; 'XYZZ'; 'CDE'};
% v2 = [NaN; -1; 8; 10; 4];
% v3 = categorical({'yes'; '--'; 'yes'; 'no'; 'yes'},{'yes' 'no'});
% T = table(v1,v2,v3)
% U = rmmissing(T)
%
% % Ignore rows with NaN entries when using sortrows
% a = [ [20 10 NaN 30 -5]', [1:5]' ]
% [b,ia] = rmmissing(a)
% a(~ia,:) = sortrows(b)
%
% See also ISMISSING, STANDARDIZEMISSING, FILLMISSING, ISNAN, ISNAT
% Copyright 2015-2016 The MathWorks, Inc.
[A,AisTable,byrows,kount,dataVars,allVars] = parseInputs(A,varargin{:});
if ~AisTable
I = ismissing(A);
I = cumputeIndex(I,byrows,kount);
else
if ~all(varfun(@ismatrix,A,'OutputFormat','uniform'))
error(message('MATLAB:rmmissing:NDArrays'));
end
if byrows
if allVars
I = ismissing(A);
else
I = ismissing(A(:,dataVars));
end
I = cumputeIndex(I,byrows,kount);
if isa(A,'timetable')
% Also remove the rows that correspond to missing RowTimes
I = I | ismissing(A.Properties.RowTimes);
end
else
I = false(1,width(A));
for vj = dataVars
Ivj = ismissing(A(:,vj));
I(vj) = cumputeIndex(Ivj,byrows,kount);
end
end
end
B = reduceSize(A,I,byrows);
%--------------------------------------------------------------------------
function I = cumputeIndex(I,byrows,kount)
if byrows
I = sum(I,2) >= kount;
else
I = sum(I,1) >= kount;
end
end
%--------------------------------------------------------------------------
function B = reduceSize(A,I,byrows)
% Keep non-missing
if byrows
B = A(~I,:);
else
B = A(:,~I);
end
end
%--------------------------------------------------------------------------
function [A,AisTable,byrows,kount,dataVars,allVars]=parseInputs(A,varargin)
% Parse RMMISSING inputs
AisTable = isa(A,'table') || isa(A,'timetable');
if ~isnumeric(A) && ~islogical(A) && ...
~ischar(A) && ~iscategorical(A) && ~iscellstr(A) && ~isstring(A) && ...
~isdatetime(A) && ~isduration(A) && ~iscalendarduration(A) && ...
~AisTable
error(message('MATLAB:rmmissing:FirstInputInvalid'));
end
if ~ismatrix(A)
error(message('MATLAB:rmmissing:NDArrays'));
end
% Defaults
kount = 1;
byrows = true;
allVars = true; % use all table variables
if ~AisTable
if isrow(A) && ~isscalar(A)
byrows = false;
end
dataVars = []; % not supported for arrays
else
dataVars = 1:width(A);
end
if nargin > 1
input2 = varargin{1};
offsetNV = 1; % N-V pairs start at 3rd and 4th inputs
if ischar(input2) && nargin > 2
% rmmissing(A,'MinNumMissing',M)
offsetNV = 0; % N-V pairs start at 2nd and 3rd inputs
else
% rmmissing(A,DIM,...)
if (isnumeric(input2) || islogical(input2)) && isscalar(input2)
if input2 == 1
byrows = true;
elseif input2 == 2
byrows = false;
else
error(message('MATLAB:rmmissing:DimensionInvalid'));
end
else
error(message('MATLAB:rmmissing:DimensionInvalid'));
end
end
% Parse N-V pairs
if nargin > 2
indNV = (1+offsetNV):numel(varargin);
if rem(length(indNV),2) ~= 0
error(message('MATLAB:rmmissing:NameValuePairs'));
end
for i = indNV(1:2:end)
opt = varargin{i};
if matlab.internal.math.checkInputName(opt,'MinNumMissing')
kount = varargin{i+1};
if (~isnumeric(kount) && ~islogical(kount)) || ~isscalar(kount) || ~isreal(kount) || fix(kount) ~= kount || ~(kount >= 0)
error(message('MATLAB:rmmissing:MinNumMissing'));
end
elseif matlab.internal.math.checkInputName(varargin{i},'DataVariables')
allVars = false;
if AisTable
dataVars = matlab.internal.math.checkDataVariables(A,varargin{i+1},'rmmissing');
else
error(message('MATLAB:rmmissing:DataVariablesArray'));
end
else
error(message('MATLAB:rmmissing:NameValueNames'));
end
end
end
end
end % parseInputs
end % rmmissing