123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770 |
- function hd = coneAnalysis
-
- datachoice = 'visual field'; % for main figure
- % datachoice = 'retina'; % for supplementary figure
- warning('off','MATLAB:interp1:UsePCHIP')
- % deactivating the following cbrewer warning:
- % Warning: INTERP1(...,'CUBIC') will change in a future release. Use INTERP1(...,'PCHIP') instead.
- % > In interp1>sanitycheckmethod (line 265)
- % In interp1>parseinputs (line 401)
- % In interp1 (line 80)
- % In interpolate_cbrewer (line 31)
- % In cbrewer (line 101)
- % In heatmap20180823 (line 21)
-
- % global globalparam
- close all
-
- cd(['..\..\data\Fig6 cone data\'])
-
-
- % Define which character sequences identify the different types of data file.
- filestring.all = 'allCones';
- filestring.blue = 'Blue';
- filestring.green = 'Green';
- filestring.red = 'Red';
- filestring.uv = 'UV';
-
-
- % Manually set max. cone density per square mm, as in Zimmermann et al.
- maxdensity = 35000;
-
- % Manually select data files, and automatically check their names.
- switch datachoice
- case 'visual field'
- fileselection = {'VisualFieldProjection_Red_8dpf.txt', ...
- 'VisualFieldProjection_Green_8dpf.txt','VisualFieldProjection_Blue_8dpf.txt', ...
- 'VisualFieldProjection_UV_8dpf.txt','VisualFieldProjection_allCones_8dpf.txt'};
- case 'retina'
- fileselection = {'DensityinRetina_Red_8dpf.txt', ...
- 'DensityinRetina_Green_8dpf.txt','DensityinRetina_Blue_8dpf.txt', ...
- 'DensityinRetina_UV_8dpf.txt','DensityinRetina_allCones_8dpf.txt'};
- end
- occurrence = struct('all',[],'blue',[],'green',[],'red',[],'uv',[]);
- for file = 1:numel(fileselection)
- occurrence.all = [occurrence.all, ~isempty(strfind(fileselection{file},filestring.all))];
- occurrence.blue = [occurrence.blue, ~isempty(strfind(fileselection{file},filestring.blue))];
- occurrence.green = [occurrence.green, ~isempty(strfind(fileselection{file},filestring.green))];
- occurrence.red = [occurrence.red, ~isempty(strfind(fileselection{file},filestring.red))];
- occurrence.uv = [occurrence.uv, ~isempty(strfind(fileselection{file},filestring.uv))];
- end
-
-
- % Check whether each type of file was selected exactly once.
- if sum(occurrence.all) == 0
- disp(['None of the file names contain the characters ''',filestring.all,'''.'])
- elseif sum(occurrence.all) > 1
- error(['More than one file name contains the characters ''',filestring.all,'''!'])
- else
- datafile.all = fileselection{find(occurrence.all)}; %#ok<FNDSB>
- end
- if sum(occurrence.blue) == 0
- disp(['None of the file names contain the characters ''',filestring.blue,'''.'])
- elseif sum(occurrence.blue) > 1
- error(['More than one file name contains the characters ''',filestring.blue,'''!'])
- else
- datafile.blue = fileselection{find(occurrence.blue)}; %#ok<FNDSB>
- end
- if sum(occurrence.green) == 0
- disp(['None of the file names contain the characters ''',filestring.green,'''.'])
- elseif sum(occurrence.green) > 1
- error(['More than one file name contains the characters ''',filestring.green,'''!'])
- else
- datafile.green = fileselection{find(occurrence.green)}; %#ok<FNDSB>
- end
- if sum(occurrence.red) == 0
- disp(['None of the file names contain the characters ''',filestring.red,'''.'])
- elseif sum(occurrence.red) > 1
- error(['More than one file name contains the characters ''',filestring.red,'''!'])
- else
- datafile.red = fileselection{find(occurrence.red)}; %#ok<FNDSB>
- end
- if sum(occurrence.uv) == 0
- disp(['None of the file names contain the characters ''',filestring.uv,'''.'])
- elseif sum(occurrence.uv) > 1
- error(['More than one file name contains the characters ''',filestring.uv,'''!'])
- else
- datafile.uv = fileselection{find(occurrence.uv)}; %#ok<FNDSB>
- end
- % % Fit centres from FitAzimElev20181217.txt:
- % leftnormmaxazim = 74.1675;
- % leftnormmaxelev = 5.9175;
- % leftnormmeanazim = 74.9488;
- % leftnormmeanelev = 5.7009;
- % leftnormratioazim = 74.9490;
- % leftnormratioelev = 5.7007;
- % rightnormmaxazim = -72.0633;
- % rightnormmaxelev = 7.3858;
- % rightnormmeanazim = -71.1931;
- % rightnormmeanelev = 7.5766;
- % rightnormratioazim = -71.1933;
- % rightnormratioelev = 7.5767;
- leftazim_body = -80.2647;
- leftelev_body = 6.0557;
- rightazim_body = +77.0306;
- rightelev_body = -1.9776;
- % Median eye position, pooled once, from positionAnalysis20181217.m:
- leftmedianazim = +6.2450;
- rightmedianazim = -9.8760;
- leftmeanazim = +5.2455;
- rightmeanazim = -9.8840;
- leftstdazim = 6.1703;
- rightstdazim = 6.5227;
- leftsemazim = 0.2472;
- rightsemazim = 0.2613;
- % Median vertical eye position, from figureVertical20190225.m:
- leftmeanelev = 3.4937; % meanmeanposition.UUH(:,:,1);
- rightmeanelev = 4.8735; % meanmeanposition.UUH(:,:,2);
-
- % Corrected to eye-centred visual field angles:
- leftazim_eye = abs(leftazim_body - leftmeanazim);
- leftelev_eye = leftelev_body - leftmeanelev;
- rightazim_eye = abs(rightazim_body - rightmeanazim);
- rightelev_eye = rightelev_body - rightmeanelev;
- % Create one figure panel per file, and all plots on it.
- datapresent = fieldnames(datafile);
- % Reorder field names to reflect desired arrangement of panels on figure.
- datapresent = {datapresent{[4 3 2 5 1]}};
-
- for typeindex = 1:numel(datapresent)
-
- conetype = datapresent{typeindex};
- textfile = datafile.(conetype);
- switch datachoice
- case 'visual field'
- conedata = importdata(textfile) * maxdensity;
- case 'retina'
- conedata = importdata(textfile);
- end
- switch datachoice
- case 'visual field'
- conedata = rot90(conedata,3); % so up is at top, anterior to the left
- case 'retina'
- conedata = rot90(conedata,1); % so up is at bottom, anterior to the right
- end
- conedata(conedata==0) = NaN; % consider nonzero data only, i.e. from the true area of the retina
-
- switch datachoice
- case 'visual field'
- % Undo approximate correction by Takeshi, then apply a more precise one:
- offsetAzimTakeshi = 8;
- offsetElevTakeshi = 5;
- offsetAzimAccurate = abs(leftmedianazim);
- offsetElevAccurate = leftmeanelev;
- resolutionAzim = size(conedata,2)/360;
- resolutionElev = size(conedata,1)/180;
- discreteShiftAzim = round((offsetAzimTakeshi-offsetAzimAccurate)/resolutionAzim);
- discreteShiftElev = round((offsetElevTakeshi-offsetElevAccurate)/resolutionElev);
- nandata = NaN(size(conedata));
- % Execute shift (min/max sanitise indices in case shift is negative):
- nandata(max(1,1-discreteShiftElev):min(end,end-discreteShiftElev), ...
- max(1,1-discreteShiftAzim):min(end,end-discreteShiftAzim)) = ...
- conedata(max(1,1+discreteShiftElev):min(end,end+discreteShiftElev), ...
- max(1,1+discreteShiftAzim):min(end,end+discreteShiftAzim));
- % Overwrite original data:
- conedata = nandata;
- case 'retina'
- % Cannot apply correction, as VF filter used by Takeshi is unknown.
- end
- % cd(['C:\Users\dehmelt\Dropbox\Documents\MATLAB\', ...
- % '201808 SphereAnalysis'])
- %
- % numelevation = size(conedata,1);
- % numazimuth = size(conedata,2);
- %
- % % The following four values represent the coordinates of the four corners
- % % of the 2D data sets provided by Baden lab (and are currently unknown).
- % elevation.max = 90;
- % elevation.min = -90;
- % azimuth.max = 180;
- % azimuth.min = -180;
- % % Build an (NX*NY x 2) array, where NX is the number of azimuth values,
- % % NY is the number of elevation values, NX*NY is the total number of data
- % % points, and for each data point (each row), the azimuth (column 1) and
- % % elevation (column 2) are listed.
- % elevation.step = (elevation.max-elevation.min)/(numelevation-1);
- % azimuth.step = (azimuth.max-azimuth.min)/(numazimuth-1);
- % elevationlist = elevation.max:-elevation.step:elevation.min;
- % azimuthlist = azimuth.max:-azimuth.step:azimuth.min;
- % predictor = [];
- % for k = 1:numazimuth
- % predictor = [predictor; ...
- % repmat(azimuthlist(k),[numelevation 1]), elevationlist'];
- % end
- % % One raw data value per data point, might have to be reformatted to match
- % % the order in which the list of data azimuths/elevations in "predictor" is
- % % being created. Predictor goes top-to-bottom, left-to-right.
- % response = conedata(:);
- %
- % offset = 0;
- % amplitude = 1;
- % meanazimuth = 90;
- % meanelevation = 0;
- % concentration = 1;
- %
- % globalparam.offset = offset;
- % globalparam.amplitude = amplitude;
- % globalparam.concentration = concentration;
- %
- %
- % figure(1)
- % % imagesc(conedata)
- % surf(conedata)
- % shading flat
- % axis square
- % colormap(cbrewer('div','PRGn',100))
- % hold on
- % plot([1,numazimuth],[numelevation/2 numelevation/2],'--w','LineWidth',2)
- % plot([numazimuth/2, numazimuth/2],[1,numelevation],'--w','LineWidth',2)
- % hold off
- % set(gca,'XTick',[1,numazimuth*[1/4,1/2,3/4,1]],'XTickLabel',{180 90 0 -90 -180})
- % set(gca,'YTick',[1,numelevation*[1/4,1/2,3/4,1]],'YTickLabel',{90 45 0 -45 -90})
- % xlabel('azimuth'),ylabel('elevation'),set(gcf,'Color',[1 1 1])
- % dataclim = get(gca,'CLim');
- hd.cone.fg = figure(1);
- hd.cone.fg.Position = [400 160 1000 700];
- % hd.cone.fg.Position = [500 50 540 380];
- hd.cone.fg.Color = [1 1 1];
- hd.cone.ax(typeindex,1) = axes;
- if typeindex < 4
- hd.cone.ax(typeindex,1).Position = [.11+(typeindex-1)*.29 .61 hd.cone.fg.Position([4 3])/3500];
- else
- hd.cone.ax(typeindex,1).Position = [.19+(typeindex-4)*.42 .11 hd.cone.fg.Position([4 3])/3500];
- end
-
- az = 0:180/(size(conedata,1)-1):180;
- el = -90:180/(size(conedata,2)-1):90;
- % conedatanonnan = fliplr(flipud(conedata));
- conedatanonnan = conedata;
- conedatanonnan(isnan(conedatanonnan)) = 0;
- density = conv2(conedatanonnan,fspecial('gaussian',[100 100],3),'same');
- % surf(az,el,conv2(conedatanonnan,fspecial('gaussian',[100 100],10),'same'))
- % shading flat
- hold on
- if strcmp(datachoice,'retina') && typeindex == 5 % FD20190829 added as simple correction
- density = fliplr(density);
- end
- imagesc(az,el,density)
- % if strcmp(conetype,'all')
- % contour(az,el,density,'LevelStep',1e-1,'Color',.5*[1 1 1])
- % else
- % contour(az,el,density,'LevelStep',3e-2,'Color',.5*[1 1 1])
- % end
- contourlevels = (.1:.1:1) * max(max(density)); % increments = 10%-of-max.
- contour(az,el,density,contourlevels,'Color',.5*[1 1 1]);
- hold off
- % axis square
- % uistack(hd.cone.ax(typeindex,2),'bottom')
- % hd.cone.ax(2,2) = copyobj(hd.cone.ax(2,1),hd.cone.fg)
- hd.cone.ax(typeindex,1).XLim = [0,180];
- hd.cone.ax(typeindex,1).YLim = [-90,90];
- oldclim = hd.cone.ax(typeindex,1).CLim;
- % climcutoff = .8;
- climcutoff = .60;
- % climcutoff = 0;
- hd.cone.ax(typeindex,1).CLim = [min(oldclim)+climcutoff*(max(oldclim)-min(oldclim)), ...
- max(oldclim)];
- switch datachoice
- case 'visual field'
- hd.cone.ax(typeindex,1).XLabel.String = 'azimuth';
- hd.cone.ax(typeindex,1).YLabel.String = 'elevation';
- case 'retina'
- hd.cone.ax(typeindex,1).XLabel.String = 'x (a.u.)';
- hd.cone.ax(typeindex,1).YLabel.String = 'y (a.u.)';
- end
- textfilesplit = strsplit(textfile,'_');
- switch textfilesplit{2}
- case 'allCones'
- densitylabel = 'all cones combined';
- case 'Green'
- densitylabel = 'green cones (mm^{-2})';
- case 'Red'
- densitylabel = 'red cones (mm^{-2})';
- case 'Blue'
- densitylabel = 'blue cones (mm^{-2})';
- case 'UV'
- densitylabel = 'ultraviolet cones';
- end
- % switch textfilesplit{2}
- % case 'allCones'
- % densitylabel = {'retinal density','(any cones)'};
- % case 'Green'
- % densitylabel = {'retinal density / mm^{-2}','(green cones)'};
- % case 'Red'
- % densitylabel = {'retinal density','(red cones)'};
- % case 'Blue'
- % densitylabel = {'retinal density','(blue cones)'};
- % case 'UV'
- % densitylabel = {'retinal density','(ultraviolet cones)'};
- % end
- % switch textfilesplit{2}
- % case 'allCones'
- % densitylabel = 'density (any cones)';
- % case 'Green'
- % densitylabel = 'density (green cones)';
- % case 'Red'
- % densitylabel = 'density (red cones)';
- % case 'Blue'
- % densitylabel = 'density (blue cones)';
- % case 'UV'
- % densitylabel = 'density (ultraviolet cones)';
- % end
- hd.cone.ax(typeindex,1).Title.String = densitylabel;
- hd.cone.ax(typeindex,1).Box = 'on';
- % hd.cone.ax(typeindex,1).XTick = 0:15:180;
- % hd.cone.ax(typeindex,1).YTick = -90:15:90;
- % hd.cone.ax(typeindex,1).XTickLabel = {0,[],[],45,[],[],90,[],[],135,[],[],180};
- % hd.cone.ax(typeindex,1).YTickLabel = {-90,[],[],-45,[],[],0,[],[],45,[],[],90};
- % hd.cone.ax(typeindex,1).XTick = 0:45:180;
- % hd.cone.ax(typeindex,1).YTick = -90:45:90;
- % hd.cone.ax(typeindex,1).XTickLabel = {0,45,90,135,180};
- % hd.cone.ax(typeindex,1).YTickLabel = {-90,-45,0,45,90};
- hd.cone.ax(typeindex,1).XTick = 0:30:180;
- hd.cone.ax(typeindex,1).YTick = -90:30:90;
- switch datachoice
- case 'visual field'
- hd.cone.ax(typeindex,1).XTickLabel = {0,[],[],90,[],[],180};
- hd.cone.ax(typeindex,1).YTickLabel = {-90,[],[],0,[],[],90};
- case 'retina'
- hd.cone.ax(typeindex,1).XTickLabel = {0,[],[],[],[],[],1};
- hd.cone.ax(typeindex,1).YTickLabel = {0,[],[],[],[],[],1};
- end
- hd.cone.ax(typeindex,1).FontSize = 14;
- hd.cone.ax(typeindex,1).TitleFontSizeMultiplier = 1;
- hd.cone.ax(typeindex,1).TitleFontWeight = 'bold';
- % hd.cone.ax(typeindex,1).Title.Interpreter = 'none';
- hd.cone.ax(typeindex,1).Title.Interpreter = 'tex';
- % [hd.cone.ax(typeindex,1).XLabel.FontSize, ...
- % hd.cone.ax(typeindex,1).YLabel.FontSize, ...
- % hd.cone.ax(typeindex,1).Title.FontSize, ...
- % ha.cone.ax(typeindex,1).FontSize,
- hd.cone.ax(typeindex,1).XGrid = 'on';
- hd.cone.ax(typeindex,1).YGrid = 'on';
- hd.cone.ax(typeindex,1).Layer = 'top';
-
-
- hold on
- plot([0,180],[0,0],'--k')
- plot([90,90],[-90,90],'--k')
- hold off
-
-
- % % leftnormmaxazimVF = abs(leftnormmaxazim - leftmedianazim);
- % % leftnormmeanazimVF = abs(leftnormmeanazim - leftmedianazim);
- % % leftnormratioazimVF = abs(leftnormratioazim - leftmedianazim);
- % % rightnormmaxazimVF = abs(rightnormmaxazim - rightmedianazim);
- % % rightnormmeanazimVF = abs(rightnormmeanazim - rightmedianazim);
- % % rightnormratioazimVF = abs(rightnormratioazim - rightmedianazim);
- hold on
- hd.cone.ob(1,1) = plot([153,165],[-65,-65],'k');
- hd.cone.ob(1,2) = plot([159,159],[-59,-71],'k');
- switch datachoice
- case 'visual field'
- hd.cone.ob(1,3) = text(153,-64.5,'A','HorizontalAlignment','right','VerticalAlignment','middle');
- hd.cone.ob(1,4) = text(166,-64.5,'P','HorizontalAlignment','left','VerticalAlignment','middle');
- hd.cone.ob(1,5) = text(159.2,-59.5,'U','HorizontalAlignment','center','VerticalAlignment','bottom');
- hd.cone.ob(1,6) = text(159.2,-69.5,'D','HorizontalAlignment','center','VerticalAlignment','top');
- case 'retina'
- hd.cone.ob(1,3) = text(153,-64.5,'P','HorizontalAlignment','right','VerticalAlignment','middle');
- hd.cone.ob(1,4) = text(166,-64.5,'A','HorizontalAlignment','left','VerticalAlignment','middle');
- hd.cone.ob(1,5) = text(159.2,-59.5,'D','HorizontalAlignment','center','VerticalAlignment','bottom');
- hd.cone.ob(1,6) = text(159.2,-69.5,'U','HorizontalAlignment','center','VerticalAlignment','top');
- end
- hold off
-
- % circle.radius = 40/2;
- % circle.xcentre = leftnormmaxazimVF; % pick preferred measure here
- % circle.ycentre = leftnormmaxelev; % pick preferred measure here
- % circle.x = -circle.radius:.1:circle.radius;
- % circle.y = sqrt(circle.x(1)^2-circle.x.^2);
- % hold on
- % plot([circle.xcentre + circle.x, fliplr(circle.xcentre + circle.x)], ...
- % [circle.ycentre + circle.y, circle.ycentre - circle.y],'-','LineWidth',3,'Color',0*[1 1 1])
- % plot([circle.xcentre + circle.x, fliplr(circle.xcentre + circle.x)], ...
- % [circle.ycentre + circle.y, circle.ycentre - circle.y],'-','LineWidth',2,'Color',1*[1 1 1])
- % hold off
-
- % boundary.angle = 64;
- % % centre.azimuth = leftnormmaxazimVF;
- % % centre.elevation = leftnormmaxelev;
- % centre.azimuth = rightazim;
- % centre.elevation = rightelev;
- % boundary.resolution = .01;
- % [boundary.azimuth,boundary.elevation] = ...
- % diskBoundary(centre.azimuth,centre.elevation,boundary.angle,boundary.resolution);
- % hold on
- % plot(boundary.azimuth,boundary.elevation,'-','LineWidth',3,'Color',0*[1 1 1])
- % plot(boundary.azimuth,boundary.elevation,'-','LineWidth',2,'Color',1*[1 1 1])
- % hold off
- boundary.angle = 40;
- % centre.azimuth = leftnormmaxazimVF;
- % centre.elevation = leftnormmaxelev;
- % centre.azimuth = leftazim_body;
- % centre.elevation = leftelev_body;
- centre.azimuth = abs(leftazim_body);
- centre.elevation = leftelev_body;
- switch datachoice
- case 'retina'
- centre.azimuth = 180-centre.azimuth;
- centre.elevation = -centre.elevation;
- end
- switch datachoice
- case 'visual field'
- boundary.resolution = .01;
- [boundary.azimuth,boundary.elevation] = ...
- diskBoundary(centre.azimuth,centre.elevation,boundary.angle,boundary.resolution);
- hold on
- plot(boundary.azimuth,boundary.elevation,'-','LineWidth',3,'Color',0*[1 1 1])
- plot(boundary.azimuth,boundary.elevation,'-','LineWidth',2,'Color',1*[1 1 1])
- hold off
- end
- mtp = 'o';
- msz = 100;
- mec = [0 0 0];
- % mfc = [.2 .8 .2];
- % mfa = .2;
- mfc = 1*[1 1 1];
- mfa = .5;
- hold on
- % hd.cone.ob(1,7) = scatter(rightazim,rightelev,msz,mtp, ...
- % 'MarkerEdgeColor',mec,'MarkerFaceColor',mfc,'MarkerFaceAlpha',mfa);
- hd.cone.ob(1,8) = scatter(centre.azimuth,centre.elevation,msz,mtp, ...
- 'MarkerEdgeColor',mec,'MarkerFaceColor',mfc,'MarkerFaceAlpha',.8);
- % hd.cone.ob(1,9) = scatter(leftnormratioazimVF,leftnormratioelev,msz,mtp, ...
- % 'MarkerEdgeColor',mec,'MarkerFaceColor',mfc,'MarkerFaceAlpha',mfa);
- % hd.cone.ob(1,10) = scatter(rightnormmaxazimVF,rightnormmaxelev,msz,mtp, ...
- % 'MarkerEdgeColor',mec,'MarkerFaceColor',mfc,'MarkerFaceAlpha',mfa);
- % hd.cone.ob(1,11) = scatter(rightnormmeanazimVF,rightnormmeanelev,msz,mtp, ...
- % 'MarkerEdgeColor',mec,'MarkerFaceColor',mfc,'MarkerFaceAlpha',mfa);
- % hd.cone.ob(1,12) = scatter(rightnormratioazimVF,rightnormratioelev,msz,mtp, ...
- % 'MarkerEdgeColor',mec,'MarkerFaceColor',mfc,'MarkerFaceAlpha',mfa);
- hold off
-
-
-
- if typeindex == 2 || typeindex == 3
- hd.cone.ax(typeindex,1).YLabel.String = [];
- hd.cone.ax(typeindex,1).YTickLabel = [];
- end
- switch conetype
- case 'blue'
- colormap(hd.cone.ax(typeindex,1),cbrewer('seq','Blues',100))
- case 'green'
- colormap(hd.cone.ax(typeindex,1),cbrewer('seq','Greens',100))
- case 'red'
- colormap(hd.cone.ax(typeindex,1),cbrewer('seq','Reds',100))
- case 'uv'
- colormap(hd.cone.ax(typeindex,1),cbrewer('seq','Purples',100))
- case 'all'
- colormap(hd.cone.ax(typeindex,1),cbrewer('seq','Greys',100))
- end
-
- hd.cone.ax(typeindex,2) = axes;
- hd.cone.ax(typeindex,2).Position = hd.cone.ax(typeindex,1).Position + ...
- hd.cone.ax(typeindex,1).Position(3) * .28*[1 0 0 0];
- hd.cone.cbar(typeindex) = colorbar;
- hd.cone.ax(typeindex,2).Colormap = hd.cone.ax(typeindex,1).Colormap;
- hd.cone.ax(typeindex,2).CLim = hd.cone.ax(typeindex,1).CLim;
- hd.cone.ax(typeindex,2).FontSize = hd.cone.ax(typeindex,1).FontSize;
- hd.cone.ax(typeindex,2).Color = 'none';
- [hd.cone.ax(typeindex,2).XAxis.Visible,hd.cone.ax(typeindex,2).YAxis.Visible] = deal('off');
-
-
- % colormap(cbrewer('div','PRGn',100))
- end
-
- hd.cone.ax(1,3) = axes;
- hd.cone.ax(1,3).Position = [8 0 1 1];
- hd.cone.ax(1,3).Color = 'none';
- [hd.cone.ax(1,3).XAxis.Visible,hd.cone.ax(1,3).YAxis.Visible] = deal('off');
- for typeindex = 1:numel(datapresent)
- parentpos = hd.cone.ax(typeindex,1).Position;
- xpos = parentpos(1)+parentpos(3)/2;
- ypos = parentpos(2)+parentpos(4)*1.128;
- hd.cone.ob(3,typeindex) = text(xpos, ypos, 'retinal density / mm^{-2}');
- % hd.cone.ob(3,typeindex) = text(xpos, ypos, 'retinal density (µm^{-2})');
- hd.cone.ob(3,typeindex).FontWeight = 'normal';
- hd.cone.ob(3,typeindex).Interpreter = 'tex';
- hd.cone.ob(3,typeindex).FontSize = hd.cone.ax(typeindex,1).FontSize;
- hd.cone.ob(3,typeindex).HorizontalAlignment = 'center';
- hd.cone.ob(3,typeindex).VerticalAlignment = 'bottom';
- end
- uistack(hd.cone.ax(1,3),'bottom')
- hd.cone.ax(1,4) = axes;
- hd.cone.ax(1,4).Position = [0 0 1 1];
- hd.cone.ax(1,4).Color = 'none';
- [hd.cone.ax(1,4).XAxis.Visible,hd.cone.ax(1,4).YAxis.Visible] = deal('off');
- for typeindex = 1:numel(datapresent)
- parentpos = hd.cone.ax(typeindex,1).Position;
- % xpos = parentpos(1)+parentpos(3)*-.18;
- xpos = parentpos(1)+parentpos(3)*-.23;
- ypos = parentpos(2)+parentpos(4)*1.30;
- % hd.cone.ob(3,typeindex) = text(xpos, ypos, 'retinal density / mm^{-2}');
- switch typeindex
- case 1
- letterstring = 'a';
- case 4
- letterstring = 'b';
- case 5
- letterstring = 'c';
- otherwise
- letterstring = '';
- end
- hd.cone.ob(4,typeindex) = text(xpos, ypos, letterstring);
- hd.cone.ob(4,typeindex).FontWeight = 'bold';
- hd.cone.ob(4,typeindex).Interpreter = 'tex';
- hd.cone.ob(4,typeindex).FontSize = hd.cone.ax(typeindex,1).FontSize * 18/14;
- hd.cone.ob(4,typeindex).HorizontalAlignment = 'right';
- hd.cone.ob(4,typeindex).VerticalAlignment = 'top';
- end
- uistack(hd.cone.ax(1,3),'bottom')
- [hd.cone.ax(:,1).TickLength] = deal([0 0]);
-
- % for k = 1:5, hd.cone.cbar(k,1).Position(1) = hd.cone.ax(k,1).Position(1) + .83*hd.cone.ax(k,1).Position(3); end
-
- runtime = clock;
- timestring = ['_', ...
- num2str(runtime(1),'%.4u'), ...
- num2str(runtime(2),'%.2u'), ...
- num2str(runtime(3),'%.2u'), ...
- '_', ...
- num2str(runtime(4),'%.2u'), ...
- num2str(runtime(5),'%.2u'), ...
- num2str(runtime(6),'%02.0f')];
-
- % savefig(2,['conefigure/',textfilesplit{2},timestring])
- % figure(2),print(['conefigure/',textfilesplit{2},timestring],'-depsc')
- % % figure(2),print(['conefigure/',textfilesplit{2},timestring],'-dtiff')
- % figure(2),print(['conefigure/',textfilesplit{2},timestring],'-dtiffn')
- % figure(2),print(['conefigure/',textfilesplit{2},timestring],'-dpng')
-
- end
- function [azimuthOut, elevationOut] = diskBoundary(azimuthIn, elevationIn, angleIn, resolutionIn)
- centre.azimuth = azimuthIn;
- centre.elevation = elevationIn;
- border.angle = angleIn / 2; % convert solid angle ("diameter") to "radius"
- border.resolution = resolutionIn;
- % The following equation describes orthodromic (or "great circle") distance as an angle:
- % arccos(sin(centre.elevation)*sin(border.elevation) + ...
- % cos(centre.elevation)*cos(border.elevation)*cos(border.azimuth-centre.azimuth)) = angle;
- % border.elevation = centre.elevation-border.angle/2 : .1 : centre.elevation+border.angle/2;
- intended = -90 : border.resolution : 90;
- % intended = (centre.elevation-border.angle) : border.resolution : (centre.elevation+border.angle);
- % "auxiliary" elevations are explicitly added to guarantee coverage near top/bottom of circle:
- auxiliary = centre.elevation + border.angle*[-1 1];
- % border.elevation = unique([intended,auxiliary]);
-
- limit = cosd([centre.elevation-border.angle centre.elevation+border.angle]);
- % intended = acosd(min(limit) : border.resolution : max(limit));
- % % XXX Not sure why the following line leads to half of the curve segments not being computed...
- % intended = [-acosd(0:border.resolution:1),acosd(1-border.resolution:-border.resolution:0)];
- % auxiliary = centre.elevation + border.angle*[-1 1];
- border.elevation = unique([intended,auxiliary]);
- % figure(6)
- % plot(intended)
- border.azimuth = centre.azimuth + ...
- acosd((cosd(border.angle) - sind(centre.elevation)*sind(border.elevation)) ./ ...
- (cosd(centre.elevation)*cosd(border.elevation)));
- % Eliminate complex elements...
- border.azimuth(imag(border.azimuth)~=0) = NaN; % This statement sometimes fails horribly.
- % border.azimuth(imag(border.azimuth)>1e-5) = NaN; % This statement is silly, but more stable.
- % ...and compute corresponding values for the second half of the boundary.
- border.azimuth = [border.azimuth, 2*centre.azimuth-border.azimuth];
- border.elevation = [border.elevation, border.elevation];
-
- border.azimuth = mod(border.azimuth+180,360)-180;
- % jump = diff(border.azimuth) > 10*border.resolution;
- % border.azimuth(jump) = NaN;
- % border.elevation(jump) = NaN;
-
- azimuthOut = border.azimuth;
- elevationOut = border.elevation;
- end
- % function combinedvalue = vonmisesfisher2param_centre(param,predictor)
- %
- % global globalparam
- %
- % convention = 'CCW';
- % dimension = 3;
- % radius = 1;
- % if size(predictor,2) == 2
- % predictor(:,3) = radius*ones(size(predictor(:,1)));
- % end
- % [x,y,z] = geo2car(predictor(:,1),predictor(:,2),predictor(:,3),convention);
- % direction = [x,y,z];
- %
- % numberofdistributions = 1; % <---
- % value = NaN(size(predictor,1),numberofdistributions);
- %
- % for k = 1:numberofdistributions
- %
- % meanazimuth = param(2*k-1);
- % meanelevation = param(2*k);
- %
- % offset = globalparam.offset(k);
- % amplitude = globalparam.amplitude(k);
- % concentration = globalparam.concentration(k);
- %
- % [meanx,meany,meanz] = geo2car(meanazimuth,meanelevation,radius,convention);
- % meandirection = [meanx,meany,meanz];
- %
- % bessel = concentration / ...
- % (2*pi * (exp(concentration) - exp(-concentration)));
- %
- % normalisation = concentration^(dimension/2 - 1) / ...
- % ((2*pi)^(dimension/2) * bessel);
- %
- % value(:,k) = normalisation * exp(concentration*meandirection*direction')';
- % value(:,k) = amplitude * value(:,k) + offset;
- %
- % badness = fitpenalty(meanazimuth,-180,180) + ...
- % fitpenalty(meanelevation,-90,90) + ...
- % fitpenalty(amplitude,0,Inf) + ...
- % fitpenalty(concentration,0,Inf);
- %
- % value(:,k) = value(:,k) + 1e3*badness;
- %
- % end
- %
- % combinedvalue = sum(value,2);
- %
- % end
- % function penalty = fitpenalty(value,lowerlimit,upperlimit)
- %
- % factor = 1;
- %
- % if value < lowerlimit
- %
- % penalty = factor*(lowerlimit-value).^1;
- %
- % elseif value > upperlimit
- %
- % penalty = factor*(value-upperlimit).^1;
- %
- % else
- %
- % penalty = 0;
- %
- % end
- %
- % end
- % function combinedvalue = oneVonMisesFisher(params,predictor)
- % %TWOVONMISESFISHER Evaluate two overlapping von Mises-Fisher distributions.
- % %
- % % VALUE = VONMISESFISHER computes the value of only one(!)
- % % spherical (p=3) von Mises-Fisher distributions at the chosen location.
- % %
- % % It is formatted to match the input/output structure required by the
- % % built-in non-linear regression function nlinfit.
- %
- % convention = 'CW';
- % dimension = 3;
- % radius = 1;
- % if size(predictor,2) == 2
- % predictor(:,3) = radius*ones(size(predictor(:,1)));
- % end
- % [x,y,z] = geo2car(predictor(:,1),predictor(:,2),predictor(:,3),convention);
- % direction = [x,y,z];
- %
- % numberofdistributions = 1; % <-- This is the only difference to "twoVonMisesFisher.m".
- % value = NaN(size(predictor,1),numberofdistributions);
- %
- % for k = 1:numberofdistributions
- %
- % offset = params(5*k-4);
- % amplitude = params(5*k-3);
- % meanazimuth = params(5*k-2);
- % meanelevation = params(5*k-1);
- %
- % meanelevation = 25; % HACK!
- %
- %
- % [meanx,meany,meanz] = geo2car(meanazimuth,meanelevation,radius,convention);
- % meandirection = [meanx,meany,meanz];
- % % meandirection = geo2car([params((-2:-1)+4*k),radius]);
- % concentration = params(5*k);
- %
- % bessel = concentration / ...
- % (2*pi * (exp(concentration) - exp(-concentration)));
- %
- % normalisation = concentration^(dimension/2 - 1) / ...
- % ((2*pi)^(dimension/2) * bessel);
- %
- % value(:,k) = normalisation * exp(concentration*meandirection*direction')';
- % value(:,k) = amplitude * value(:,k) + offset;
- %
- % % HACK to constrain values
- % if abs(meanazimuth) > 180 || ...
- % abs(meanelevation) > 90 || ...
- % amplitude < 0 || ...
- % concentration < 0
- %
- % value(:,k) = 1e10*ones(size(value(:,k)));
- %
- % end
- %
- % end
- %
- % combinedvalue = sum(value,2);
- %
- % end
|