www.gusucode.com > 基于MATLAB色板调用源码程度 > 基于MATLAB色板调用源码程度/UnivarScatter_v1.0/UnivarScatter.m
function [xPositions, yPositions, Label, RangeCut] = UnivarScatter(data, varargin)
%--------------------------------------------------------------------------
% UnivariateScatter - Draw an Univariate Scatter plot out of a nx2 table with a
% categorical/string and a numerical variable, or out of a
% numerical array, with groups/categories corresponding
% to the columns of the numerical array.
% Many custom options are available as Name,Value
% pairs. For optimal visualization of your data I
% recommend to play with 'RangeCut', and also with pbaspect of the plot, it
% really changes the appearance.
% You need the following functions for this function to
% work:
%
% +CatTable2Array: included in the file
%
% Also, a very simple function to assign colors is
% provided, ColorCoder, you can see an example in the
% script attached
%
% Input: data - Can take two types of input:
%
% + a nx2 table with a categorical/string and a numerical
% variable, the groups will be made according to the
% categories
%
% + a numerical array in which the different columns
% correspond to the groups/categories of the Univariate
% Scatter plot in x axis. When plotting data with
% different number of points for the different
% groups, the empty elements of the array should be
% filled with nan. In order to do this much faster I
% recommend the function padcat(available in
% mathworks, Copyright (c) 2009, Jos van der Geest), which already
% concatenates row/column vectors of different
% lengths and fills with nans the missing values.
%
% Name,Value - Name-Value pairs arguments to customize the plot:
%
% +'Label' - Only when data is a numerical array
% A cell array of strings containing the
% labels that you want to give to categories/groups
% in the plot, by default they are 1,2,3...n The
% number of columns of this array must be the same as
% the number of columns in data, example:
% UniVarScatter(data,'Label',{'a','b','c'})
%
% +'Whiskers' - A string which can take 3 values
% *'lines': std, 95% SEM, mean are represented as
% lines
% *'boxes':std, 95% SEM, mean are represented as
% boxes. Default value
% *'none':none of these are shown
%
% +'PointStyle' - A String containing the different
% values that can be given as markertype for scatter,
% such as 'o','*''.', etc. Default is 'o'
%
% +'Width' - A number that determines the spread of
% each subset of points in x axis. Default is 0.4
%
% +'Compression' - A number that determines how
% separated from each other the points are in x
% dimension. Default is 5
%
% +'RangeCut' - A number or column/row vector with
% length= size(data,2) that determines how broad
% groups of points are made in y dimension, groups
% width in y dimension corresponds to 1/RangeCut of the
% 95% confidence interval of a normal distribution
% with std and mean of each subset of points.
% Default is 2+2*sqrt(number of points) for each subset
% of points, this has shown to be a good value, but
% it does not perform optimally for every dataset, so
% I strongly recommend not to discard the graph and
% try different combinations of values if you don't
% obtained a nice result.
%
% +'WhiskersWidthRatio' - Lines/boxes that represent
% the mean,SEM,std will be Width/WhiskersWidthRatio
% long in x dimension. Default is 4
%
% +'PointSize' - A number indicating the point size.
% Default is 36
%
% +'LineWidth' - A number indicating the width of the
% edge of the points. Default is 0.5
%
% +'MarkerEdgeColor' color(s) of the edges of the points; def. 'k'
% +'MarkerFaceColor' color(s) of the points; def. 'flat'
% +'MeanColor' color(s) of the mean line; def. black
% +'SEMColor' color(s) of the SEM box; def. dark gray
% +'StdColor' color(s) of the Std box; def. light gray
%
% All this last Color Properties behave similarly,
% they can take several types of values
%
% *RGB vector 1x3: all the representations
% are made in the same colour represented by that vector
%
% *A string: it may contain the name of the color
% for instance 'k' or 'black', also it can take
% the values 'none', or 'flat'
%
% *RGB array nx3 where n>=size(data,2): each row
% of this array corresponds to the color of the n
% group/category in x. You can use the function
% ColorCoder included in the zip to Easily create
% a custom RGB array by chosing the colors from
% the matlab palette
%
% Output: the figure
%
% xPositions - An Array with the x positions of the points, with
% the same size as yPositions
%
%
% yPositions - a numerical array in which the different columns
% correspond to the groups/categories of the Univariate
% Scatter plot in x axis, its the same as data if data was
% already a numerical array, and if it was a table,
% it generates a numerical array, and the columns
% correspond to the Label groups.
%
% Label - A string cell array with the names of the groups
% specially useful when using a table as input, so
% you know the order in which the scatter plots are
% showed.
%
% RangeCut - The value of RangeCut(explained at the Name,Values
% section), specially useful to optimize the
% representation of your data if the RangeCut value
% generated by the function does not perform well for
% your data.
%
% version: <1.00> from 30/11/2015
%
% Manuel Lera Ram韗ez: manulera14@gmail.com
%
% Developed during Master Thesis Internship in Marcos Gonz醠ez-Gait醤's Lab
% Departments of Biochemistry and Molecular Biology, Sciences II,
% 30 Quai Ernest-Ansermet, CH-1211 Geneva 4, Switzerland
%--------------------------------------------------------------------------
%% Variable handling and default values generation
if nargin==0
error(sprintf('There\''s no input'))
end
if mod(length(varargin),2)~=0
error(sprintf('The arguments should be in pairs,in which the first one is a string, as in (data, \''Label\'',{\''A\'',\''B\''})'))
end
stringVars=varargin(1:2:end);
valueVars=varargin(2:2:end);
if any(~cellfun(@isstr,stringVars))
error(sprintf('The arguments should be in pairs,in which the first one is a string, as in (data, \''Label\'',{\''A\'',\''B\''})'))
end
%data_istable is a logical value used to suppress the default Label value
%assignation
if istable(data)
data_istable=true;
[data,Label]=CatTable2Array(data);
warning('The label will be taken from the table, any label input will be ignored');
else
data_istable=false;
end
%% String Variables
if ~data_istable
LabelInd=strmatch('Label',stringVars);
if ~isempty(LabelInd) && data_istable==false
%Check that the Label is a string cell array and has a propper length
if iscellstr(valueVars{LabelInd}) && length(valueVars{LabelInd})==size(data,2)
Label=valueVars{LabelInd};
else
error('wrong Label, Label should be a linear cell array of strings, its length should be the same as the number of columns of data')
end
else
Label=[];
end
end
Ind=strmatch('Whiskers',stringVars);
if ~isempty(Ind)
%Check that Whiskers is a string with a valid value
if ischar(valueVars{Ind}) && any([strcmp('box',valueVars{Ind}),strcmp('none',valueVars{Ind}),strcmp('lines',valueVars{Ind})])
Whiskers=valueVars{Ind};
else
error(sprintf('wrong Whiskers, it should be a string saying \''none\'', \''box\'' or \''lines\'''))
end
else
Whiskers='box';
end
Ind=strmatch('PointStyle',stringVars);
if ~isempty(Ind)
%Check that PointStyle is a string
if ischar(valueVars{Ind})
PointStyle=valueVars{Ind};
else
error(sprintf('wrong PointStyle, it should be a string defining the shape of the point \''o\'', \''.\'', \''*\'', etc.'))
end
else
PointStyle='o';
end
%% Numeric Variables
Ind=strmatch('Width',stringVars);
if ~isempty(Ind)
%Check that Width is a number
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind})== [1,1])
Width=valueVars{Ind};
else
error(sprintf('wrong Width, it should be a number'))
end
else
Width=0.4;
end
Ind=strmatch('Compression',stringVars);
if ~isempty(Ind)
%Check that Compression is a number
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind})== [1,1])
Compression=valueVars{Ind};
else
error(sprintf('wrong Compression, it should be a number'))
end
else
Compression=5;
end
Ind=strmatch('RangeCut',stringVars);
if ~isempty(Ind)
%Check that RangeCut is a number
if isnumeric(valueVars{Ind})
if size(valueVars{Ind})==1
RangeCut=valueVars{Ind}*ones(size(data,2),1);
elseif all(size(valueVars{Ind})== [size(data,2),1]) || all(size(valueVars{Ind}) == [1,size(data,2)])
RangeCut=valueVars{Ind};
else
error(sprintf('wrong RangeCut, it should be a number, or a column/row vector with length=size(data,2)'))
end
else
error(sprintf('wrong RangeCut, it should be a number, or a column/row vector with length=size(data,2)'))
end
else
numPoints=sum(~isnan(data));
RangeCut=(log(numPoints)/log(150)+3*(numPoints).^(1/3))';
end
Ind=strmatch('WhiskersWidthRatio',stringVars);
if ~isempty(Ind)
%Check that WhiskersWidthRatio is a number
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind})== [1,1])
WhiskersWidthRatio=valueVars{Ind};
else
error(sprintf('wrong WhiskersWidthRatio, it should be a number'))
end
else
WhiskersWidthRatio=4;
end
Ind=strmatch('PointSize',stringVars);
if ~isempty(Ind)
%Check that PointSize is a number
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind})== [1,1])
PointSize=valueVars{Ind};
else
error(sprintf('wrong PointSize, it should be a positive number'))
end
else
PointSize=36;
end
Ind=strmatch('LineWidth',stringVars);
if ~isempty(Ind)
%Check that LineWidth is a number
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind})== [1,1])
LineWidth=valueVars{Ind};
else
error(sprintf('wrong LineWidth, it should be a positive number'))
end
else
LineWidth=0.5;
end
%% Mixed Variables
%Some of the following variables are the classic scatter personalizing tools,
%please revisit the scatter documentation for deeper understanding.
Ind=strmatch('MarkerEdgeColor',stringVars);
EdgeIsArray=false;
if ~isempty(Ind)
%Check that MarkerEdgeColor takes valid arguments
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) >= [size(data,2),3])
MarkerEdgeColor=valueVars{Ind};
EdgeIsArray=true;
elseif (isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) == [1,3])) || ischar(valueVars{Ind})
MarkerEdgeColor=valueVars{Ind};
else
error(sprintf('wrong MarkerEdgeColor input'))
end
else
MarkerEdgeColor='k';
end
Ind=strmatch('MarkerFaceColor',stringVars);
%This will be used later when plotting
FaceIsArray=false;
if ~isempty(Ind)
%Check that MarkerFaceColor takes valid arguments
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) >= [size(data,2),3])
MarkerFaceColor=valueVars{Ind};
FaceIsArray=true;
elseif (isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) == [1,3])) || ischar(valueVars{Ind})
MarkerFaceColor=valueVars{Ind};
else
error(sprintf('wrong MarkerFaceColor input'))
end
else
MarkerFaceColor='flat';
end
Ind=strmatch('MeanColor',stringVars);
MeanColorIsArray=false;
if ~isempty(Ind)
%Check that MeanColor takes valid arguments
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) >= [size(data,2),3])
MeanColor=valueVars{Ind};
MeanColorIsArray=true;
elseif (isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) == [1,3])) || ischar(valueVars{Ind})
MeanColor=valueVars{Ind};
else
error(sprintf('wrong MeanColor input'))
end
else
MeanColor='k';
end
Ind=strmatch('SEMColor',stringVars);
SEMColorIsArray=false;
if ~isempty(Ind)
%Check that SEMColor takes valid arguments
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) >= [size(data,2),3])
SEMColor=valueVars{Ind};
SEMColorIsArray=true;
elseif (isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) == [1,3])) || ischar(valueVars{Ind})
SEMColor=valueVars{Ind};
else
error(sprintf('wrong SEMColor input'))
end
elseif strcmp(Whiskers,'box')
SEMColor=[1 1 1]*0.95;
else
SEMColor='k';
end
Ind=strmatch('StdColor',stringVars);
StdColorIsArray=false;
if ~isempty(Ind)
%Check that StdColor takes valid arguments
if isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) >= [size(data,2),3])
StdColor=valueVars{Ind};
StdColorIsArray=true;
elseif (isnumeric(valueVars{Ind}) && all(size(valueVars{Ind}) == [1,3])) || ischar(valueVars{Ind})
StdColor=valueVars{Ind};
else
error(sprintf('wrong StdColor input'))
end
elseif strcmp(Whiskers,'box')
StdColor=[1 1 1]*0.85;
else
StdColor='k';
end
%% Plotting data
HoldWasOn=ishold;
%This is just a trick to not change anything if there was a plot with hold on already,
%to create a new figure if there was none, and to clear the figure if it
%had hold off
plot(nan,nan)
hold on
xPositions=nan(size(data));
yPositions=data;
for i=1:size(data,2)
%% Sort the data and determine the limits of the different groups in which we divide the points
% yValues is a column vector containing one of the columns of the matrix
% data, it contains the y values of one of the subset of points we want to
% represent.
% We sort it in order to make groups according to its value, however, we
% want to keep the information of the order of the yvalues, just in case
% its needed for something, we used SortingIndex for that
% eliminate the nans and keep the information to later put it in
% xPositions, this info is kept in yValues_nanIndex
yValues_nanIndex=~isnan(data(:,i));
yValues=data(yValues_nanIndex,i);
[yValues,y_SortingIndex]=sort(yValues);
[~,x_SortingIndex]=sort(y_SortingIndex);
%If one of the columns is empty, we don't plot anything
if isempty(yValues)
continue
end
% xValues is a column vector as big as yValues, but with all values equal to i. If we
% represented plot(xValues, yValues), we would get all the points with the
% same x,a dot blot, and therefore there would be a massive overlapping
% depending on how many points we had. Thats why the while loops that we
% find afterwards change the x of each point, so they spread and do not
% overlap.
xValues=ones(size(yValues))*i;
%% Plotting the error bars and the standard deviation bars
%Color type managing
if MeanColorIsArray
MeanIndex=i;
else
MeanIndex=1;
end
if SEMColorIsArray
SEMIndex=i;
else
SEMIndex=1;
end
if StdColorIsArray
StdIndex=i;
else
StdIndex=1;
end
%If we want boxes for the mean and std
if strcmp(Whiskers,'box')
yMean=mean(yValues);
yStd=std(yValues);
ySem=yStd/sqrt(size(yValues,1));
yCI=ySem*1.96;
%plot the standard deviation box
rectangle('Position',[i-Width/WhiskersWidthRatio,yMean-yStd,2*Width/WhiskersWidthRatio,2*yStd ],'FaceColor',StdColor(StdIndex,:),'EdgeColor', StdColor(StdIndex,:),'LineWidth',0.1);
%plot the mean+-SEM box
rectangle('Position',[i-Width/WhiskersWidthRatio,yMean-yCI,2*Width/WhiskersWidthRatio,2*yCI ],'FaceColor',SEMColor(SEMIndex,:),'EdgeColor', SEMColor(SEMIndex,:),'LineWidth',0.1);
%plot the mean line
plot([i-Width/WhiskersWidthRatio i+Width/WhiskersWidthRatio],[yMean yMean],'Color',MeanColor(MeanIndex,:),'LineWidth',2)
end
%% Changing the xValue of each point
% This is done the following way: The points are sorted in equidistant
% groups according to yValues, this distance is stablished as 1/RangeCut of the
% width of the 95% interval of the data. Of course this makes sense for
% more or less normally distributed data, but it can be changed easily, and also it can
% be changed easily changing RangeCut value
range_val=norminv([0.05 0.95],mean(yValues),std(yValues));
%range_val=quantile(yValues,[0.05 0.95]);
cuts=abs(range_val(1)-range_val(2))/RangeCut(i);
% In case one of the variables has equal values for all its points, the
% norminv will return a nan, since std(yValues) will be 0, in that case we
% arbitrarily give cuts the value of cuts=mean(yValues)/2, this value does not
% really matter because there will only be one group anyway
if isnan(cuts)
cuts=mean(yValues)/2;
end
% The "seed" group contains the values which are in the range of the
% mean+-1/2cuts. cutUp is the higher border of the interval. So we
% take this value, and from it we move up and down in this two loops(steps is a variable
% that allows us to use the same while loop to move in both directions,
% selecting the subset of yValues that are in each group range, and modifying
% their xValues, so that they are spread and do not overlap.
for steps=[-1,1]
cutUp=mean(yValues)+cuts/2;
subsetind= yValues<=cutUp & yValues>cutUp-cuts;
keep_going=true;
while keep_going
% subsetind is a logical variable that represents how many points are
% in that group
if all(sum(subsetind)~=[1 0]) %If there's only one point, we represent it in the middle, and xValues remains equal to i
% distmaker is a variable that is equal to Width when sum(subsetind)=1 , and gets smaller when the number of points
% gets bigger in a group, exponentially, and it tends to zero in
% the infinite. I don't have strong arguments for the choice of
% this particular expression, but for me it works very well, and
% you can tune with Width and Compression very well the appereance of
% the plot
distmaker=Width./(exp((sum(subsetind).^2-1)./(sum(subsetind)*Compression)));
% xSubset is a column vector with all values equal to i, and as
% long as the subset of the number of values in yValues that belong
% to this particular group range
xSubset=ones(sum(subsetind),1)*i;
%oddie is a variable that indicates whether the number of points in
%the group is odd or even, it's very important for the indexing
%afterwards
oddie=mod(size(xSubset,1),2);
% xb is a symmetrical vector with mean=0, and the values in it have
% the displacements in x that we want to apply to xValues, but if
% we just applied it increasingly, the smaller yValue would get the most negative
% xValues displacement,etc. and what we would have is a series of lines of points, and
% in publications, what you usually get is that the points with
% either the highest or
% lowest values get the positions that are more outside, and then the lowest or
% the highest get the central position. In this function, the lowest values of yValues
% get the external positions, and the highest values get the central positions, making
% some sort of eyebrow or sad mouth shape)
% The range of xb gets bigger as the number of points inside the
% group increases due to the action of distmaker, which gets
% smaller as number of points increases.
xb=linspace(1-Width+distmaker,1+Width-distmaker,size(yValues(subsetind),1))-1;
% Since xb is symmetrical it's easy to add the more extreme values
% of xb to the xValues with the lower yValues. oddie is needed to
% index properly depending on wether the number of elements is even
% or odd. If we sorted the values of xb by their absolute value,
% their index would be (1,end,2,end-1,3,end-2,etc.) If you think of
% an example or you debug the function and observe the values of xb
% it's very easy to understand the indexing done here. Also, it's
% here where oddie is useful
xSubset(1:2:end-oddie)=xSubset(1:2:end-oddie)+xb(1:round(end/2)-oddie)';
xSubset(2:2:end)=xSubset(2:2:end)-xb(1:round(end/2)-oddie)';
xValues(subsetind)= xSubset;
%This following code line is very useful to see how the groups are made,
%for watching it, comment the scatter lines in the end, and
%uncomment this line, however groups with one point won't be shown
%scatter(xValues(subsetind),yValues(subsetind),PointStyle)
end
% advance in the while loop as long as the cutUp value is out of the
% range of the points
keep_going=~(cutUp>max(yValues) || cutUp<min(yValues));
cutUp=cutUp+steps*cuts;
subsetind= yValues<cutUp & yValues>cutUp-cuts;
end
end
%% Drawing each subset of points
% This is to overcome the fact that MarkerEdgeColor and MarkerFaceColor can
% be either strings,1x3 vectors or nx3 arrays
if EdgeIsArray
EdgeIndex=i;
else
EdgeIndex=1;
end
if FaceIsArray
FaceIndex=i;
else
FaceIndex=1;
end
scatter(xValues,yValues,PointSize,PointStyle,'MarkerEdgeColor', MarkerEdgeColor(EdgeIndex,:),'MarkerFaceColor',MarkerFaceColor(FaceIndex,:),'LineWidth',LineWidth)
%If we want lines to represent the SEM and the std
if strcmp(Whiskers,'lines')
%plot the mean line
yMean=mean(yValues);
plot([i-Width/WhiskersWidthRatio i+Width/WhiskersWidthRatio],[yMean yMean],'Color',MeanColor(MeanIndex),'LineWidth',1.5)
%plot the sd
yStd=std(yValues);
plot([i-Width/WhiskersWidthRatio*0.5 i+Width/WhiskersWidthRatio*0.5],[yMean+yStd yMean+yStd],'Color',StdColor(StdIndex,:),'LineWidth',1.5)
plot([i-Width/WhiskersWidthRatio*0.5 i+Width/WhiskersWidthRatio*0.5],[yMean-yStd yMean-yStd],'Color',StdColor(StdIndex,:),'LineWidth',1.5)
plot([i i],[yMean-yStd yMean+yStd],'Color',StdColor(StdIndex,:))
%plot the conf. interval of the mean
ySem=yStd/sqrt(size(yValues,1));
yCI=ySem*1.96;
plot([i-Width/WhiskersWidthRatio*0.7 i+Width/WhiskersWidthRatio*0.7],[yMean+yCI yMean+yCI],'Color',SEMColor(SEMIndex,:),'LineWidth',1.5)
plot([i-Width/WhiskersWidthRatio*0.7 i+Width/WhiskersWidthRatio*0.7],[yMean-yCI yMean-yCI],'Color',SEMColor(SEMIndex,:),'LineWidth',1.5)
plot([i i],[yMean-yCI yMean+yCI],'Color',SEMColor(SEMIndex,:))
end
xPositions(yValues_nanIndex,i)=xValues(x_SortingIndex);
end
set(gca,'xtick',1:size(data,2))
if ~isempty(Label)
set(gca,'xticklabel',Label)
pbaspect([size(data,2)+1,4,1])
end
if ~HoldWasOn
hold off
end
end