Revision fb79cc2fe753e26c55913cdf415442059a2eade7 authored by tyson aflalo on 11 December 2020, 19:26:23 UTC, committed by tyson aflalo on 11 December 2020, 19:26:23 UTC
1 parent d3f857c
LatencyAnalysis.m
classdef LatencyAnalysis < handle & Utilities.StructableHierarchy & Utilities.Structable
% Transform nsX file into wavelet features sampled at the task
% intervals.
% this uses the standard {nspIDX}{FeatureIDX} format for the output.
properties (GetAccess = 'public', SetAccess = 'public')
% parameters
dimred=[]
overwrite=1;
MaxNumChanges=2;
stati='linear';
end
properties (GetAccess = 'public', SetAccess = 'private')
% parameters
Lat;
Mu;
LatS;
MuLat1;
MuLat2;
CILat1
dPrime
timeWindow % CAR pca score for CARinds
end
methods
function obj=LatencyAnalysis(varargin)
[varargin]=Utilities.argobjprop(obj,varargin);
Utilities.argempty(varargin)
end
function [out,SaveID]=fit_transform(obj,basename,FRTrue,timeWindow,varargin)
% transform braodband data to features binned at the same time
% intervals as the task file.
SaveID=sprintf('%s_FaceScratch', basename);
Basedir=fullfile(env.get('result'),'FaceScratch','ImagExp');
if exist(fullfile(Basedir,[SaveID '.mat'])) && ~obj.overwrite
disp('file exists - loading')
tmp=load(fullfile(Basedir,SaveID))
out.cvAccuracy=tmp.cvAccuracy;
out.timeWindow=tmp.timeWindow;
return
end
for i=1:size(FRTrue,1)
TMP= FRTrue(i,:);
TMP=cellfun(@(x)cat(1,x{:}),TMP,'UniformOutput',false);
missingVals=~any(isnan(TMP{1}),1);
TMP=cellfun(@(x)(x(:,missingVals)),TMP,'UniformOutput',false);
FROut{i}=TMP;
%replace nans with means
A=cat(1,TMP{:});
obj.dimred.fit_transform(A);
for j=1:length(TMP)
[L{j}]=transform(obj.dimred,TMP{j});
end
for jj=1:size(L{1},2)
disp(jj)
tmp=cell2mat(cellfun(@(x)nanmean(x(:,jj),1),L,'UniformOutput',false));
if mean(tmp(1:floor(length(tmp)/2))) > mean(tmp(ceil(length(tmp)/2):end))
for kk=1:length(L)
L{kk}(:,jj)=-L{kk}(:,jj);
end
% L=cellfun(@(x)x(:,jj)=-(x(:,jj)),L,'UniformOutput',false);
end
end
Lat{i}=L;
Mu{i}=cell2mat(cellfun(@(x)nanmean(x(:)),FROut{i},'UniformOutput',false));
LatS{i}=cell2mat(cellfun(@(x)(x(:,1)),L,'UniformOutput',false));
MuLat1{i}=cell2mat(cellfun(@(x)nanmean(x(:,1),1),Lat{i},'UniformOutput',false));
CILat1{i}=cell2mat(cellfun(@(x)bootci(2000,@nanmean,x(:,1),1),Lat{i},'UniformOutput',false));
MuLat2{i}=cell2mat(cellfun(@(x)nanmean(x(:,2),1),Lat{i},'UniformOutput',false));
MuLat3{i}=cell2mat(cellfun(@(x)nanmean(x(:,3),1),Lat{i},'UniformOutput',false));
end
obj.Lat=Lat;
obj.Mu=Mu;
obj.LatS=LatS;
obj.MuLat1=MuLat1;
obj.MuLat2=MuLat2;
obj.CILat1=CILat1;
obj.timeWindow=timeWindow;
out.Lat=Lat;
out.Mu=Mu;
out.LatS=LatS;
out.MuLat1=MuLat1;
out.MuLat2=MuLat2;
out.CILat1=CILat1;
out.timeWindow=timeWindow;
save(fullfile(Basedir,SaveID),'-struct','out')
end
function [out,SaveID]=fit_DiscrimTransform(obj,basename,FRTrue,Labls,timeWindow,varargin)
% transform braodband data to features binned at the same time
% intervals as the task file.
CondIDX=[1 2]
SaveID=sprintf('%s_FaceScratch', basename);
Basedir=fullfile(env.get('result'),'FaceScratch','Discrim');
if exist(fullfile(Basedir,[SaveID '.mat'])) && ~obj.overwrite
disp('file exists - loading')
tmp=load(fullfile(Basedir,SaveID))
out.cvAccuracy=tmp.cvAccuracy;
out.timeWindow=tmp.timeWindow;
return
end
for i=1:size(FRTrue,1)
TMP= FRTrue(i,:);
TMP=cellfun(@(x)cat(1,x{:}),TMP,'UniformOutput',false);
missingVals=~any(isnan(TMP{1}),1);
TMP=cellfun(@(x)(x(:,missingVals)),TMP,'UniformOutput',false);
FROut{i}=TMP;
%replace nans with means
for j=1:length(TMP)
j
N=cellfun(@(x)size(x,1),FRTrue{i});
A=[zeros(N(1),1); ones(N(2),1)];
% Results=Analyze.FitPLSLM(TMP{j},A,'Ncomp',5)
Results=Analyze.FitPLSLM([obj.Lat{i}{j}],A,'Ncomp',5);
dP(j)=Results.dP;
dPp(j)=Results.dPp;
end
DDPP{i}=dP;
end
plot(obj.timeWindow,DDPP{1})
obj.dPrime=DDPP;
out.dPrime=obj.dPrime;
% save(fullfile(Basedir,SaveID),'-struct','out')
end
function PlotData(obj,cvAccuracy,timeWindow,varargin)
plt.fig('units','inches','width',10,'height',7,'font','Helvetica','fontsize',16)
clrs=lines(length(obj.CILat1));
pnl = panel(); pnl.margin=20; pnl.pack(2,2); pnl.fontname='arial';pnl.fontsize=12;
pnl(1,1).select(); hold on
for i=1:length(obj.CILat1)
plot(obj.timeWindow,obj.Mu{i},'color',clrs(i,:),'linewidth',2)
end
xlabel('Time')
ylabel('Mean Rate (Hz)')
pnl(1,2).select(); hold on
% plot(timeWindow,-MuLat1{1},'r')
% plot(timeWindow,MuLat1{2},'k')
for i=1:length(obj.CILat1)
CI=obj.CILat1{i};
CI=CI-obj.MuLat1{i};
Utilities.boundedline(obj.timeWindow,obj.MuLat1{i},mean(abs(CI'),2),'alpha','cmap',clrs(i,:),'linewidth',2);
% Utilities.boundedline(timeWindow,MuLat1{1},abs(CI'),'alpha','cmap',clrs(1,:),'linewidth',2);
preIDX=find(obj.timeWindow<0);
preVal{i}=obj.MuLat1{i}(:,preIDX);
V95Mu(i)=prctile(preVal{i}(:),95)
plt.hline(V95Mu(i),{'color',clrs(i,:)});
end
xlabel('Time')
ylabel('Latent Proj')
axis tight
pnl(2,1).select(); hold on
for i=1:length(obj.CILat1)
CI=obj.CILat1{i};
CI=CI-obj.MuLat1{i};
Utilities.boundedline(obj.timeWindow,obj.MuLat1{i},mean(abs(CI'),2),'alpha','cmap',clrs(i,:),'linewidth',2);
% Utilities.boundedline(timeWindow,MuLat1{1},abs(CI'),'alpha','cmap',clrs(1,:),'linewidth',2);
end
xlim([0 .2])
pnl(2,2).select(); hold on
% plot(timeWindow,-MuLat1{1},'r')
% plot(timeWindow,MuLat1{2},'k')
for i=1:length(obj.CILat1)
plot(obj.timeWindow,obj.LatS{i}','color',clrs(i,:))
preIDX=find(obj.timeWindow<0);
preVal{i}=obj.LatS{i}(:,preIDX);
V95(i)=prctile(preVal{i}(:),95)
plt.hline(V95(i),{'color',clrs(i,:)});
end
axis tight
xlabel('Time')
ylabel('Latent Proj')
pnl(1,2).select(); hold on
% plot(timeWindow,-MuLat1{1},'r')
% plot(timeWindow,MuLat1{2},'k')
%
% if ~isempty(basename)
% SaveID=sprintf('%s_FaceScratch', basename);
% Basedir=fullfile(env.get('result'),'FaceScratch','ImagExp','Figs');
% colormap(jet)
% plt.SaveFigure(1,Basedir,SaveID,'PDF')
% end
end
function BootStrapPWFit(obj,varargin)
% plt.fig('units','inches','width',11,'height',5,'font','Helvetica','fontsize',16)
clrs=lines(length(obj.CILat1));
%piecewise linear
tmp=find(obj.timeWindow>.1);
P0=[0 0 0 tmp(1) 30 0];
lb = -inf*[1 1 1 1 1 1];
ub = inf*[1 1 1 1 1 1];
plusfun = @(x) max(x,0);
model = @(P,x) P(1) + P(6)*x + (P(6)-P(2))*plusfun(P(4)-x) + (P(3)-P(6))*plusfun(x-(P(4)+P(5)));
options = optimoptions('lsqcurvefit','Display','off');
for i=1:length(obj.CILat1)
for j=1:500
RandSamp=obj.LatS{i};
RandSamp=mean(RandSamp(randi(size(RandSamp,1),size(RandSamp,1),1),:),1);
% tmp=my_findchangepts( RandSamp,'Statistic',obj.stati,'MaxNumChanges',obj.MaxNumChanges,'MinDistance',10);
% LatSamp{i}(j)=obj.timeWindow(tmp(1));
Pfit = lsqcurvefit(model,P0,1:length(RandSamp),RandSamp,lb,ub,options);
modelpred = model(Pfit,sort(1:length(RandSamp)));
% plot(1:length(RandSamp),RandSamp,'o',sort(1:length(RandSamp)),modelpred,'r-')
tmp=find(modelpred>prctile( RandSamp(obj.timeWindow<0),95));
% [obj.timeWindow(tmp(1)) obj.timeWindow(round(Pfit(4))) obj.timeWindow(tmp(1))]
LatSampA{i}(j)=obj.timeWindow(tmp(1));
LatSampB{i}(j)=obj.timeWindow(round(Pfit(4)));
end
LatCIA{i}=prctile(LatSampA{i},[25 75]);
LatCIB{i}=prctile(LatSampB{i},[25 75]);
disp(LatCIA{i})
disp(LatCIB{i})
a=histogram(LatSampA{i}*1000,1000*(0.01:.005:.1),'Normalization','probability')
text(10,0+i/10,sprintf('%0.0f ',1000*LatCIA{i}))
end
xlabel('Time of significant modulation (ms)')
ylabel('Probability')
end
function PlotPWFit(obj,cvAccuracy,timeWindow,varargin)
plt.fig('units','inches','width',6,'height',3,'font','Helvetica','fontsize',16)
pnl = panel(); pnl.margin=15; pnl.pack('h',{2/3 1/3}); pnl.fontname='arial';pnl.fontsize=12;
pnl(1).select()
clrs=lines(length(obj.CILat1));
tmp=find(obj.timeWindow>.08);
P0=[0 0 0 tmp(1) 30 0];
lb = -inf*[1 1 1 1 1 1];
ub = inf*[1 1 1 1 1 1];
plusfun = @(x) max(x,0);
model = @(P,x) P(1) + P(6)*x + (P(6)-P(2))*plusfun(P(4)-x) + (P(3)-P(6))*plusfun(x-(P(4)+P(5)));
options = optimoptions('lsqcurvefit','Display','off');
for i=1:length(obj.CILat1)
CI=obj.CILat1{i};
CI=CI-obj.MuLat1{i};
% my_findchangepts( obj.MuLat1{i},'Statistic',obj.stati,'MaxNumChanges',obj.MaxNumChanges,'color',clrs(i,:));
tmp=findchangepts( obj.MuLat1{1},'Statistic',obj.stati,'MaxNumChanges',obj.MaxNumChanges);
Utilities.boundedline(1:length(obj.timeWindow),obj.MuLat1{i},mean(abs(CI'),2),'alpha','cmap',clrs(i,:),'linewidth',2);
% Utilities.boundedline(timeWindow,MuLat1{1},abs(CI'),'alpha','cmap',clrs(1,:),'linewidth',2);
Pfit = lsqcurvefit(model,P0,1:length( obj.MuLat1{i}), obj.MuLat1{i},lb,ub,options);
modelpred = model(Pfit,sort(1:length( obj.MuLat1{i})));
plot(sort(1:length( obj.MuLat1{i})),modelpred,'color',clrs(i,:),'linewidth',2)
tmp=find(obj.timeWindow<0);
tmp=find(modelpred>prctile( obj.MuLat1{i}(obj.timeWindow<0),95));
plt.vline([(tmp(1))],{'--','color',clrs(i,:),'linewidth',2})
% plt.hline(prctile( obj.MuLat1{i}(obj.timeWindow<0),95),{'--','color',clrs(i,:),'linewidth',2})
% plt.vline([Pfit(4) Pfit(4)+Pfit(5)],{'color',clrs(i,:)},'linewidth',2)
end
%%
set(gca,'XTick',1:25:length(obj.timeWindow))
obj.timeWindow( cellfun(@(x)str2num(x),get(gca,'XTickLabel')))
% set(gca,'XTickLabel',obj.timeWindow( 1:25:length(obj.timeWindow)))
set(gca,'XTickLabel',1000*obj.timeWindow( cellfun(@(x)str2num(x),get(gca,'XTickLabel'))))
axis tight
xlabel('Time relative to probe contact (ms)')
ylabel('First Latent Dim')
pnl(2).select(); hold on
obj.BootStrapPWFit()
%%
%
% pnl = panel(); pnl.margin=20; pnl.pack('h',length(obj.CILat1)); pnl.fontname='arial';pnl.fontsize=12;
%
% for i=1:length(obj.CILat1)
% pnl(i).select(); hold on
%
% MaxNumChanges=2;
% stati='linear'
% CI=obj.CILat1{i};
% CI=CI-obj.MuLat1{i};
% my_findchangepts( obj.MuLat1{i},'Statistic',stati,'MaxNumChanges',MaxNumChanges);
% tmp=findchangepts( obj.MuLat1{1},'Statistic',stati,'MaxNumChanges',MaxNumChanges);
% Utilities.boundedline(1:length(obj.timeWindow),obj.MuLat1{i},mean(abs(CI'),2),'alpha','cmap',clrs(1,:),'linewidth',2);
% % Utilities.boundedline(timeWindow,MuLat1{1},abs(CI'),'alpha','cmap',clrs(1,:),'linewidth',2);
%
% end
%
% if ~isempty(basename)
% SaveID=sprintf('%s_FaceScratch', basename);
% Basedir=fullfile(env.get('result'),'FaceScratch','ImagExp','Figs');
% colormap(jet)
% plt.SaveFigure(1,Basedir,SaveID,'PDF')
% end
end
function PlotDiscrimLat(obj,FRTrue,timeWindow,varargin)
plt.fig('units','inches','width',11,'height',5,'font','Helvetica','fontsize',16)
clrs=lines(length(obj.CILat1));
tmp=find(obj.timeWindow>.08);
P0=[0 0 0 tmp(1) 30 0];
lb = -inf*[1 1 1 1 1 1];
ub = inf*[1 1 1 1 1 1];
plusfun = @(x) max(x,0);
model = @(P,x) P(1) + P(6)*x + (P(6)-P(2))*plusfun(P(4)-x) + (P(3)-P(6))*plusfun(x-(P(4)+P(5)));
options = optimoptions('lsqcurvefit','Display','off');
for i=1:length(obj.CILat1)
CI=obj.CILat1{i};
CI=CI-obj.MuLat1{i};
% my_findchangepts( obj.MuLat1{i},'Statistic',obj.stati,'MaxNumChanges',obj.MaxNumChanges,'color',clrs(i,:));
tmp=findchangepts( obj.MuLat1{1},'Statistic',obj.stati,'MaxNumChanges',obj.MaxNumChanges);
Utilities.boundedline(1:length(obj.timeWindow),obj.MuLat1{i},mean(abs(CI'),2),'alpha','cmap',clrs(i,:),'linewidth',2);
% Utilities.boundedline(timeWindow,MuLat1{1},abs(CI'),'alpha','cmap',clrs(1,:),'linewidth',2);
Pfit = lsqcurvefit(model,P0,1:length( obj.MuLat1{i}), obj.MuLat1{i},lb,ub,options);
modelpred = model(Pfit,sort(1:length( obj.MuLat1{i})));
plot(sort(1:length( obj.MuLat1{i})),modelpred,'color',clrs(i,:),'linewidth',2)
plt.vline([Pfit(4) Pfit(4)+Pfit(5)],{'color',clrs(i,:)},'linewidth',2)
end
%%
set(gca,'XTick',1:25:length(obj.timeWindow))
obj.timeWindow( cellfun(@(x)str2num(x),get(gca,'XTickLabel')))
set(gca,'XTickLabel',obj.timeWindow( cellfun(@(x)str2num(x),get(gca,'XTickLabel'))))
% if ~isempty(basename)
% SaveID=sprintf('%s_FaceScratch', basename);
% Basedir=fullfile(env.get('result'),'FaceScratch','ImagExp','Figs');
% colormap(jet)
% plt.SaveFigure(1,Basedir,SaveID,'PDF')
% end
end
end
end
Computing file changes ...
