gusucode.com > bigdata 工具箱 matlab源码程序 > bigdata/+matlab/+bigdata/+internal/+adaptors/TableAdaptor.m
%TableAdaptor Adaptor class for tabular tall arrays.
% Copyright 2016 The MathWorks, Inc.
classdef TableAdaptor < matlab.bigdata.internal.adaptors.AbstractAdaptor
properties (SetAccess = immutable)
VariableNames
end
properties (SetAccess = private)
VariableAdaptors
end
methods (Access = private)
function [varNames, varIdxs] = resolveVarNameSubscript(obj, subscript)
% Subscript type conversions - resolve certain types up-front.
if isa(subscript, 'matlab.bigdata.internal.util.EndMarker')
% For table indexing, we must resolve EndMarkers in the second subscript at the
% client right away.
szVec = [0, numel(obj.VariableNames)];
subscript = resolve(subscript, szVec, 2);
elseif islogical(subscript)
% Also convert logical to numeric up front
subscript = find(subscript);
elseif isstring(subscript)
% Treat string as cellstr to fit in with ismember etc.
subscript = cellstr(subscript);
end
if isnumeric(subscript)
if max(subscript(:)) > numel(obj.VariableNames)
error(message('MATLAB:table:VarIndexOutOfRange'));
end
% Note that we are implicitly relying on the indexing into VariableNames to
% perform validation of 'subscript' to check for negative / zero
% / NaN etc. values. But see g1368852 - table allows t.(0) and
% t.(-1)...
varNames = obj.VariableNames(subscript);
elseif iscellstr(subscript)
varNames = subscript;
elseif matlab.bigdata.internal.util.isColonSubscript(subscript)
varNames = obj.VariableNames;
else
if ~ischar(subscript)
error(message('MATLAB:table:InvalidVarSubscript'));
end
varNames = {subscript};
end
% Ensure we've got varNames as a row
varNames = reshape(varNames, 1, []);
% Find missing names
[tf, varIdxs] = ismember(varNames, obj.VariableNames);
if ~all(tf)
missingNames = varNames(~tf);
error(message('MATLAB:table:UnrecognizedVarName', missingNames{1}));
end
end
% Resolve a single dot-subscript
function varName = resolveDotSubscript(obj, subscript, allowMissing)
% For cases tt.Foo and equivalently tt.('Foo'), we allow the subscript to be
% only: scalar string, char-vector, and numeric integer scalar.
% Handle scalar strings by converting to char.
if isstring(subscript) && isscalar(subscript)
subscript = char(subscript);
end
if ischar(subscript) || ...
(isnumeric(subscript) && isscalar(subscript) && round(subscript) == subscript)
if allowMissing && ischar(subscript)
% Missing variables are allowed - check valid variable name though
if ~isvarname(subscript)
error(message('MATLAB:table:InvalidVariableName', subscript));
end
varName = subscript;
elseif allowMissing && isnumeric(subscript)
% Numeric integer scalar - must be in range, or one off the end.
if subscript <= numel(obj.VariableNames)
varName = obj.VariableNames{subscript};
elseif subscript == (1 + numel(obj.VariableNames))
% Appending a new variable with generated name
varName = sprintf('Var%d', subscript);
idx = 1;
while ismember(varName, obj.VariableNames)
varName = sprintf('Var%d_%d', subscript, idx);
idx = idx + 1;
end
else
% Outside allowed bounds
error(message('MATLAB:table:DiscontiguousVars'));
end
elseif matlab.bigdata.internal.util.isColonSubscript(subscript)
error(message('MATLAB:table:UnrecognizedVarName', subscript));
else
% Finally, get here to resolve numeric integer scalars or char-vectors against
% known variable names, re-use resolveVarNameSubscript.
varNames = obj.resolveVarNameSubscript(subscript);
assert(isscalar(varNames), ...
'Unexpectedly resolved dot-subscript to multiple variables.');
varName = varNames{1};
end
else
error(message('MATLAB:table:IllegalVarSubscript'));
end
end
end
methods
function obj = TableAdaptor(varargin)
narginchk(1,2);
obj@matlab.bigdata.internal.adaptors.AbstractAdaptor('table');
if nargin == 1
% preview data
previewData = varargin{1};
obj.VariableNames = previewData.Properties.VariableNames;
obj.VariableAdaptors = cellfun( ...
@(vn) matlab.bigdata.internal.adaptors.getAdaptor(previewData{[],vn}), ...
obj.VariableNames, 'UniformOutput', false);
else
% names and adaptors
assert(iscellstr(varargin{1}) && ...
numel(varargin{1}) == numel(varargin{2}));
obj.VariableNames = reshape(varargin{1}, 1, []);
obj.VariableAdaptors = reshape(varargin{2}, 1, []);
end
obj = setSmallSizes(obj, length(obj.VariableNames));
for idx = 1:numel(obj.VariableAdaptors)
obj.VariableAdaptors{idx} = copyTallSize(obj.VariableAdaptors{idx}, obj);
end
end
function names = getProperties(obj)
names = [obj.VariableNames, 'Properties'];
end
function idxs = resolveVarNamesToIdxs(obj, namesOrIdxs)
[~, idxs] = obj.resolveVarNameSubscript(namesOrIdxs);
end
function obj = resetSizeInformation(obj)
% Overloaded for TableAdaptor - NDims and num variables don't change.
obj = resetTallSize(obj);
end
function obj = resetTallSize(obj, varargin)
% Overloaded for TableAdaptor - need to propagate new tall size to contained
% adaptors.
obj = resetTallSize@matlab.bigdata.internal.adaptors.AbstractAdaptor(...
obj, varargin{:});
for idx = 1:numel(obj.VariableAdaptors)
obj.VariableAdaptors{idx} = copyTallSize(obj.VariableAdaptors{idx}, obj);
end
end
function displayImpl(obj, context, ~)
if context.IsPreviewAvailable
doDisplay(context);
else
% Fabricate a preview table
var = ['?'; '?'; '?'];
vars = repmat({var}, 1, numel(obj.VariableNames));
previewData = table(vars{:}, 'VariableNames', obj.VariableNames);
doDisplayWithFabricatedPreview(context, previewData, obj.NDims, obj.Size);
end
end
function varargout = subsrefDot(obj, pa, szPa, s)
if isequal(s(1).subs, 'Properties')
rowName = getString(message('MATLAB:table:uistrings:DfltRowDimName'));
varName = getString(message('MATLAB:table:uistrings:DfltVarDimName'));
out = struct('Description', {''}, ...
'UserData', {[]}, ...
'DimensionNames', {{rowName, varName}}, ...
'VariableNames', {obj.VariableNames}, ...
'VariableDescriptions', {{}}, ...
'VariableUnits', {{}}, ...
'RowNames', {{}});
else
allowMissing = false;
varName = obj.resolveDotSubscript(s(1).subs, allowMissing);
% Extract the variable from the table
out = subsrefBraces(obj, pa, szPa, ...
struct('type', '{}', 'subs', {{':', varName}}));
end
[varargout{1:nargout}] = iRecurseSubsref(out, s(2:end));
end
function varargout = subsrefBraces(obj, pa, ~, s)
if numel(s(1).subs) ~= 2
error(message('MATLAB:table:NDSubscript'));
end
[firstSub, secondSub] = deal(s(1).subs{:});
[~, secondSubNumeric] = obj.resolveVarNameSubscript(secondSub);
shouldDereference = true;
outValue = slicefun(iExtractVariableFunctor(secondSubNumeric, shouldDereference), pa);
dim = 2;
if isempty(secondSubNumeric)
% table brace indexing selecting an empty list of variables returns Nx0 double.
adaptor = matlab.bigdata.internal.adaptors.getAdaptorForType('double');
else
adaptor = matlab.bigdata.internal.adaptors.combineAdaptors(...
dim, obj.VariableAdaptors(secondSubNumeric));
end
out = tall(outValue, adaptor);
% Use tall subsref to select rows. Note that this is a rather imperfect
% implementation as it presumes no more than 3 non-tall dimensions.
if ~matlab.bigdata.internal.util.isColonSubscript(firstSub)
newSubs = cell(1,4);
newSubs{1} = firstSub;
newSubs(2:end) = {':'};
out = subsref(out, substruct('()', newSubs));
end
[varargout{1:nargout}] = iRecurseSubsref(out, s(2:end));
end
function obj = subsasgnBraces(~, ~, ~, ~, ~) %#ok<STOUT>
error(message('MATLAB:bigdata:table:SubsasgnBracesNotSupported'))
end
function out = subsrefParens(obj, pa, szPa, s)
if numel(s(1).subs) ~= 2
error(message('MATLAB:table:NDSubscript'));
end
[firstSub, secondSub] = deal(s(1).subs{:});
% First off, subselect the columns specified by secondSub
[varNames, varIdxs] = obj.resolveVarNameSubscript(secondSub);
if ~isequal(varNames, obj.VariableNames)
shouldDereference = false;
selectedColumnsPa = slicefun(iExtractVariableFunctor(varIdxs, shouldDereference), pa);
else
selectedColumnsPa = pa;
end
selectedAdaptors = obj.VariableAdaptors(varIdxs);
newAdaptor = matlab.bigdata.internal.adaptors.TableAdaptor(varNames, selectedAdaptors);
% Next, perform the row selection
selectedRowsAndColumnsPa = subsrefParensImpl(selectedColumnsPa, szPa, ...
substruct('()', {firstSub, ':'}));
% Build the tall table
tmp = tall(selectedRowsAndColumnsPa, newAdaptor);
% and then recurse
out = iRecurseSubsref(tmp, s(2:end));
end
function pa = subsasgnParens(obj, pa, szPa, s, b) %#ok<INUSD>
error(message('MATLAB:bigdata:table:SubsasgnParensNotSupported'));
end
function out = subsasgnParensDeleting(obj, pa, szPa, s)
import matlab.bigdata.internal.util.isColonSubscript
% The language front-end should not permit expressions where there is any form
% of indexing following parens.
assert(numel(s) == 1);
if numel(s(1).subs) ~= 2
error(message('MATLAB:table:NDSubscript'));
end
[firstSub, secondSub] = deal(s(1).subs{:});
if isColonSubscript(secondSub)
% Delete whole slices
if isColonSubscript(firstSub)
error(message('MATLAB:bigdata:table:DeleteWholeTableUsingIndexing'));
elseif ~istall(firstSub)
error(message('MATLAB:bigdata:table:FirstSubscriptColonOrTallVariable'));
end
% Here we know we're left with a tall subscript in first place, we need to
% negate it (providing it's logical)
firstSub = tall.validateType(firstSub, 'subsasgn', {'logical'}, 1);
out = obj.subsrefParens(pa, szPa, substruct('()', {~firstSub, secondSub}));
else
if matlab.bigdata.internal.util.isColonSubscript(firstSub)
% Deleting whole variables - negate the variable list
deleteNames = obj.resolveVarNameSubscript(secondSub);
keepNames = setdiff(obj.VariableNames, deleteNames, 'stable');
out = obj.subsrefParens(pa, szPa, substruct('()', {firstSub, keepNames}));
else
error(message('MATLAB:table:InvalidEmptyAssignment'));
end
end
end
function out = subsasgnDot(obj, pa, szPa, s, b)
if isequal(s(1).subs, 'Properties')
error(message('MATLAB:bigdata:table:SetPropertiesUnsupported'));
end
allowMissing = true;
varName = obj.resolveDotSubscript(s(1).subs, allowMissing);
if numel(s) == 1
% Adding or updating a whole variable.
if ~istall(b)
% Note there's no scalar expansion for "t.x = b".
error(message('MATLAB:bigdata:table:AssignVariableMustBeTall'));
end
bAdaptor = matlab.bigdata.internal.adaptors.getAdaptor(b);
if ~isnan(obj.Size(1)) && ~isnan(bAdaptor.Size(1)) && obj.Size(1) ~= bAdaptor.Size(1)
error(message('MATLAB:bigdata:array:IncompatibleTallStrictSize'));
end
% Build a new adaptor
names = obj.VariableNames;
adaptors = obj.VariableAdaptors;
if ~ismember(varName, names)
names{end+1} = varName;
end
idx = find(strcmp(varName, names));
assert(isscalar(idx));
adaptors{idx} = bAdaptor;
newAdaptor = matlab.bigdata.internal.adaptors.TableAdaptor(...
names, adaptors);
outPa = strictslicefun(@(t, v) iUpdateWholeVariable(t, varName, v), ...
pa, hGetValueImpl(b));
out = tall(outPa, newAdaptor);
else
% Replacing part of variable - extract, update, replace.
tallVar = obj.subsrefDot(pa, szPa, s(1));
tallVar = subsasgn(tallVar, s(2:end), b);
out = obj.subsasgnDot(pa, szPa, substruct('.', varName), tallVar);
end
end
function out = subsasgnDotDeleting(obj, pa, ~, S)
if numel(S) > 1
error(message('MATLAB:bigdata:table:DotDeletingSingleLevelIndexing'));
end
allowMissing = false;
deletingName = obj.resolveDotSubscript(S(1).subs, allowMissing);
% Need to work out which index we're removing
deletingTF = strcmp(deletingName, obj.VariableNames);
outPa = slicefun(@(x) iRemoveVariable(x, deletingName), pa);
newAdaptor = matlab.bigdata.internal.adaptors.TableAdaptor(...
obj.VariableNames(~deletingTF), ...
obj.VariableAdaptors(~deletingTF));
out = tall(outPa, newAdaptor);
end
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function functor = iExtractVariableFunctor(vars, shouldDereference)
functor = @fcn;
function out = fcn(t)
out = t(:, vars);
if shouldDereference
out = out{:,:};
end
end
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
function t = iRemoveVariable(t, varName)
t.(varName) = [];
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Simply apply a new variable into the table, ensuring the new data is the
% correct size. The check here is only needed in rare cases (i.e. where the
% table is completely empty) - otherwise the table assignment itself actually
% throws this error. See g1367363.
function t = iUpdateWholeVariable(t, varName, v)
if size(t,1) ~= size(v,1)
error(message('MATLAB:table:RowDimensionMismatch'));
end
t.(varName) = v;
end
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Apply remaining indexing expressions
function varargout = iRecurseSubsref(data, S)
if isempty(S)
varargout = {data};
else
[varargout{1:nargout}] = subsref(data, S);
end
end