rs_toygeom_scenarioC

Geometric transformations, affine model, with statistical analysis of nesting by output dimension

Scenario C for rs_toygeom_sim: illustration of geometric models, focusing on nesting by output dimension

Geometric model simulated: affine (4 dimensions) Shows that: * With low noise (subject 1), modeling all 4 output dimensions is an improvement over modeling dimensions 1-3 * With medium noise (subject 2), adding output dimension 4 is not a statistically significant improvement * With high noise (subject 3), adding output dimension 3 is not a statistically significant improvement

clear; %parameters are read from the workspace, so start from a clean one
scenario_name='scenario C';

transform selection

transform_names={'affine'};
affine_mag=0.3;

paradigm customizations

paradigm_names={'Axes'};

subject customizations

nsubjs=3;
subjs_disp=[1:3];
ncoords=4;
ncoords_noise=1;
noise_add_subj=[0 0.2 2.0]; %a range of noise levels

geometric model selection

model_list={'affine_offset'};

opts_geof=struct;
opts_geof.if_stats=1;

create the stimuli and datasets, and fit the models (rs_toygeom_sim)

Included script: rs_toygeom_sim

rs_toygeom_sim: geometric modeling with toy simulated datasets; used by rs_toygeom_scenario*

This illustrates fitting of geometric models via creation of simulated coordinate set structures. It is intended that several of the parameters, especially paradigm_names, transform_names, are edited, or that rs_toygeom_scenario*.m is run first; running as is will be very time-consuming.

Any of the parameters may be set in the workspace before running, in which case the default is not applied; that is what rs_toygeom_scenario*.m does. So run this from a clear workspace unless you intend to supply parameters yourself.

Constructs several transformations, designated by the strings in transform_names: null: the identity procrustes: rotation, no offset, inversion forced affine: affine (linear) transformation, no offset projective: projective transformation pwaffine: piecewise affine transformation with one cutplane

Creates several stimulus sets in a space of ncoords dimensions, designated by the strings in paradigm_names Axes: points in positive and negative directions along each of the axes Rings_C12, Rings_C13, Rings_C23: concentric rings in planes (1,2), (1,3), and (2,3) Random: random points that avoid the axes RandomAndAxisEnds: random points and the endponts of Axes These coordinate set structures imported into sims.(paradigm_name).stimspace, using rs_import_coordsets, and are displayed with rs_disp_coordsets or rs_disp_enh_coordsets. Note that coordinate values are always integers, in range [-9:9]

Sets up nsubjs subjects. For each subject, the parameters of the above transformations are jittered and applied to stimspace using rs_xform_apply, noise is added, and, optionally, extra dimensions (ncoords_noise) with just noise are added. These are in sims.(paradigm_name).dataspace.(transform_name). and are displayed with rs_disp_coordsets or rs_disp_enh_coordsets.

Optionally knits together the stimuli across pardigms, creating the coordinate set structure sims.knitted.stimspace, and, for each subject, sims.knitted.dataspace.(transform_name)

The transformation from stimspace to dataspace are then fitted with geometric models with rs_geofit. The model list defaults to {procrustes_noscale_offset','procrustes_scale_offset','affine_offset','projective'}) but can modified by changing model_list

Results of the fitting are displayed with rs_toygeom_disp, which invokes rs_disp_geofit.

This script has no documentation page of its own: each rs_toygeom_scenario* page shows it where the scenario runs it.

See also: rs_import_coordsets, rs_disp_coordsets, rs_disp_geofit rs_disp_enh_coordsets, rs_xform_apply, RS_CONCAT_DATASETS, rs_extract_coordsets, rs_toygeom_disp.

these are the main simulation parameters that may be edited, or have values set before running

if ~exist('transform_names') transform_names={'null','procrustes','affine','projective','pwaffine'}; end %some may be deleted
if ~exist('affine_mag') affine_mag=0.5; end %magnitude of distortion in affine transforms
if ~exist('projective_mag') projective_mag=0.1; end %controls amount of distortion in projective transform
if ~exist('pwaffine_mag') pwaffine_mag=0.25; end %controls difference in linear transforms of piecewise affine

if ~exist('paradigm_names') paradigm_names={'Axes','Rings_C12','Rings_C13','Rings_C23','Random','RandomAndAxisEnds'}; end %some may be deleted
if ~exist('axis_samples') axis_samples=[2 4 6 8]; end %sample points in each direction along each axis
if ~exist('ring_radii') ring_radii=[4 6 8]; end %radii for the rings
if ~exist('ring_angles') ring_angles=8; end %number of sample points in a ring
if ~exist('nrandom') nrandom=16; end %number of random stimuli

if ~exist('nsubjs') nsubjs=1; end % number of subjects to simulate; at least 1; subjects have progressively more noise
if ~exist('subjs_disp') subjs_disp=unique([1,nsubjs]); end %which subjects to show in plots of response space

if ~exist('ncoords') ncoords=3; end %number of coordinates in stimulus set, should be at least 3
if ~exist('ncoords_noise') ncoords_noise=2; end %simulations can have noise on additional coordinates
if ~exist('noise_transform_mag') noise_transform_mag=0; end %set to nonzero (e..g, 0.2) to allow each subject's transformation to differ
if ~exist('noise_transform_subj') noise_transform_subj=[1:nsubjs]; end % multiplies noise_transform_mag to vary transform noise for each subject
if ~exist('noise_add_mag') noise_add_mag=1; end %additive Gaussian noise
if ~exist('noise_add_subj') noise_add_subj=[1:nsubjs]; end % mutliplies noise_add_mag to vary noise for each subject

if ~exist('if_disp_coordsets') if_disp_coordsets=1; end %set to 0 to suppress plots of coordinate sets for stimuli and subjects
if ~exist('if_frozen') if_frozen=1; end %set to 0 for random numbers each time, negative integer for fixed alternative seeds

if ~exist('if_knit') if_knit=0; end

if ~exist('model_list') model_list={'procrustes_noscale_offset','procrustes_scale_offset','affine_offset','projective'}; end

if (if_frozen~=0) 
    rng('default');
    if (if_frozen<0)
        rand(1,abs(if_frozen));
    end
else
    rng('shuffle');
end

define the coordinates

ncoords=max(3,ncoords);
coord_labels=cell(1,ncoords);
for ic=1:ncoords
    coord_labels{ic}=char('a'+ic-1);
end
ncoords_tot=ncoords+ncoords_noise;

define the paradigm (stimulus choices within each paradigm)

paradigm_type='toygeom';

random_max=9; %maximum random value; plotted range will be [-1,1]*(random_max+1)

define the simulations: several transformations of the stimulus space, of dimension ncoords

transform_names_avail={'null','procrustes','affine','projective','pwaffine'};
ntransforms=length(transform_names);
transform_class_table=struct;
transform_class_table.null='affine';
transform_class_table.procrustes='affine';
transform_class_table.affine='affine';
transform_class_table.projective='projective';
transform_class_table.pwaffine='pwaffine';
for it=1:ntransforms
    transform_classes{it}=transform_class_table.(transform_names{it});
end

null transformation

transforms.null.T=eye(ncoords);
transforms.null.b=1;
transforms.null.c=zeros(1,ncoords);

transforms.procrustes.T=eye(ncoords);
transforms.procrustes.T(1,1)=-1; %invert
for ic=1:ncoords-1
    for jc=ic+1:ncoords
        ang=2*pi*rand(1);
        rot=zeros(ncoords);
        rot([ic jc],[ic jc])=[cos(ang) sin(ang);-sin(ang) cos(ang)];
        roteye=setdiff(1:ncoords,[ic jc]);
        rot(roteye,roteye)=eye(ncoords-2);
        transforms.procrustes.T=transforms.procrustes.T*rot;
    end
end
transforms.procrustes.b=1;
transforms.procrustes.c=zeros(1,ncoords);

T=(1-affine_mag)*eye(ncoords)+affine_mag*randn(ncoords,ncoords); %mix the identity with a random matrix
transforms.affine.T=T/sqrt(max(eig(T'*T))); %limit the max dilation to keep in range
transforms.affine.b=1;
transforms.affine.c=affine_mag*randn(1,ncoords);

transforms.projective=transforms.affine;
transforms.projective.p=projective_mag*randn(ncoords,1);

transforms.pwaffine=transforms.affine;

construct by adding a linear transformation to a rectification params needed to ensure continuity across boundary

pw_vcut=randn(1,ncoords);
pw_vcut=pw_vcut./sqrt(pw_vcut*pw_vcut'); %random direction of unit normal to cutplane
pw_acut=randn(1); %random offset
pw_h=randn(1,ncoords); %the output of the rectification

the cutplane is x*pw_cut'=pw_acut.

Tcut=pw_vcut'*pw_h; %if x*pw_vcut'=pw_acut, then x*Tcut=a*pw_h,so x*Tcut-pw_acut*pw_h is zero on the boundary
transforms.pwaffine.T=repmat((1-pwaffine_mag)*transforms.affine.T,[1 1 2])+pwaffine_mag*cat(3,Tcut,-Tcut);
transforms.pwaffine.c_off=randn(1,ncoords);
transforms.pwaffine.c=pwaffine_mag*pw_acut*[-pw_h;pw_h]+repmat(transforms.pwaffine.c_off,2,1);
transforms.pwaffine.b=1;
transforms.pwaffine.acut=pw_acut;
transforms.pwaffine.vcut=pw_vcut;
transforms.pwaffine.h=pw_h;

disp(sprintf(' %2.0f transforms set up, on %3.0f coordinates.',ntransforms,ncoords));

Output:

  1 transforms set up, on   4 coordinates.

define the number of subjects and levels of noise for each

noise_transform_base=noise_transform_mag*noise_transform_subj; %subjects have varying amounts of noise in transform
noise_transform=noise_transform_base(1+mod([0:nsubjs-1],length(noise_transform_base))); %in case noise_transform_subj is too short
noise_add_base=noise_add_mag*noise_add_subj; %subjects have varying amounts of additive noise
noise_add=noise_add_base(1+mod([0:nsubjs-1],length(noise_add_base))); %in case noise_add_subj is too short

create the transformations for each subject by corrupting the parameters of the basic transforms by noise note that all transforms are treatd, even if they won't be used, since some transforms are dependent on others

transforms_noisy=cell(1,nsubjs);
for it=1:length(transform_names_avail)
    transform_name=transform_names_avail{it};
    transform=transforms.(transform_name);
    fns=fieldnames(transform);
    for is=1:nsubjs
         for ifn=1:length(fns)
            if strcmp(fns{ifn},'p')
                noise_param_mult=projective_mag;
            else
                noise_param_mult=1;
            end
            transforms_noisy{is}.(transform_name).(fns{ifn})=transform.(fns{ifn})+noise_param_mult*noise_transform(is)*randn(size(transform.(fns{ifn})));
        end %fn
        transforms_noisy{is}.(transform_name).b=1; %scale factor unchanged; redundant with T and it will disrupt pwaffine continuity

adjustment for procrustes to ensure that the transform is isotropic

        if strcmp(transform_name,'procrustes')
            [Torth,Torthonormal]=grmscmdt(transforms_noisy{is}.procrustes.T);
            transforms_noisy{is}.procrustes.T=Torthonormal;
        end

adjustments for pwaffine to ensure continuity at boundary

        if strcmp(transform_name,'pwaffine')
            pw_vcut=transforms_noisy{is}.pwaffine.vcut;
            pw_acut=transforms_noisy{is}.pwaffine.acut;
            pw_h=transforms_noisy{is}.pwaffine.h;

            pw_vcut=pw_vcut./sqrt(pw_vcut*pw_vcut'); %random direction of unit normal to cutplane
            Tcut=pw_vcut'*pw_h; %if x*pw_vcut'=pw_acut, then x*Tcut=a*pw_h
            transforms_noisy{is}.pwaffine.T=repmat((1-pwaffine_mag)*transforms_noisy{is}.affine.T,[1 1 2])+pwaffine_mag*cat(3,Tcut,-Tcut);
            transforms_noisy{is}.pwaffine.c=pwaffine_mag*pw_acut*[-pw_h;pw_h]+repmat(transforms_noisy{is}.pwaffine.c_off,2,1);
            transforms_noisy{is}.pwaffine.vcut=pw_vcut;
        end %pwaffine
    end %is
end %it
disp(sprintf('jittered transformations created for %2.0f subjects',nsubjs));

Output:

jittered transformations created for  3 subjects

define the coordinates of the stimulus sets

nparadigms=length(paradigm_names);
sims=struct;
sign_chars={'m','z','p'}; %tokens for negative, zero, or positive
for ip=1:length(paradigm_names)
    paradigm_name=paradigm_names{ip};
    sim=struct;
    switch paradigm_name
        case 'Axes'
            sim.nstims=2*ncoords*length(axis_samples)+1; %bidirectional samples on each axis, and the origin
            sim.type_coords=zeros(sim.nstims,ncoords);
            coord_ptr=0;
            for ic=1:ncoords
                for isamp=1:length(axis_samples)
                    for isign=-1:2:1
                        coord_ptr=coord_ptr+1;
                        sim.type_coords(coord_ptr,ic)=isign*axis_samples(isamp);
                    end
                end
            end
            sim.if_findrays=1;
            sim.if_rings=0;
        case {'Rings_C12','Rings_C13','Rings_C23'}
            sim.nstims=length(ring_radii)*ring_angles+1; %rings in the plane two coords, and the origin
            sim.type_coords=zeros(sim.nstims,ncoords);
            coord_ptr=0;
            if contains(paradigm_name,'C12')
                ring_plane=[1 2];
            elseif contains(paradigm_name,'C13')
                ring_plane=[1 3];
            else
                ring_plane=[2 3];
            end
            angs=2*pi*[0:ring_angles-1]'/ring_angles;
            for ir=1:length(ring_radii)
                ring_locs=ring_radii(ir)*[cos(angs) sin(angs)];
                sim.type_coords(coord_ptr+[1:ring_angles],ring_plane)=ring_locs;
                coord_ptr=coord_ptr+ring_angles;
            end
            sim.if_findrays=1;
            sim.if_rings=1;
        case {'Random','RandomAndAxisEnds'}
            if strcmp(paradigm_name,'RandomAndAxisEnds')
                if_axend=1;
            else
                if_axend=0;
            end
            sim.nstims=nrandom+if_axend*(2*ncoords+nrandom+1); %random samples and possibly (bidirectional samples at the ends of the axes and the origin)
            sim.type_coords=zeros(sim.nstims,ncoords);
            random_vals=[[-random_max:-1] [1:random_max]]; %random values avoid the axes;
            rand_samps=ceil(2*random_max*rand(nrandom,ncoords));
            sim.type_coords(1:nrandom,:)=random_vals(rand_samps);
            if if_axend

make the samples at the ends of the axes

                coord_ptr=nrandom;
                for ic=1:ncoords
                    for isign=-1:2:1
                        coord_ptr=coord_ptr+1;
                        sim.type_coords(coord_ptr,ic)=isign*axis_samples(end);
                    end
                end
            end
            sim.if_findrays=0;
            sim.if_rings=0;
        otherwise
            rs_warning('unrecognized paradigm name',1);
    end
    sims.(paradigm_name)=sim;
    disp(sprintf('coordinate sets created for paradigm %s',paradigm_name));
end

Output:

coordinate sets created for paradigm Axes

create the type names, e.g., conceptual coordinate [-3 0 4] -> 'am3 cp4' and create a simulated dataset in which data coordinates are equal to conceptual coordinates

ray_opts=struct;
ray_opts.ray_minpts=length(ring_radii); %a point on each ring qualifies as a ray

for ip=1:length(paradigm_names)
    paradigm_name=paradigm_names{ip};
    sim=sims.(paradigm_name);
    typenames=cell(sim.nstims,1);
    for istim=1:sim.nstims
        tn=' ';
        for ic=1:ncoords
            cval=sim.type_coords(istim,ic);
            if abs(cval)>0.5 %omit zeros from stimulus names to shorten them
                tn=cat(2,tn,coord_labels{ic},sign_chars{2+sign(cval)},sprintf('%1.0f',round(abs(cval))),' ');
            end
        end
        typenames{istim}=strtrim(tn);
    end
    sims.(paradigm_name).typenames=typenames;

    aux_stimspace=struct;
    aux_stimspace.opts_import.nstims=sim.nstims;
    aux_stimspace.opts_import.typenames=typenames;
    aux_stimspace.opts_import.type_coords=sim.type_coords;
    aux_stimspace.opts_import.paradigm_type=paradigm_type;
    aux_stimspace.opts_import.paradigm_name=paradigm_name;
    aux_stimspace.opts_import.subj_id='stimspace';
    aux_stimspace.opts_import.subj_id_short='stims';
    aux_stimspace.opts_import.label_long='conceptual coordinates';
    import_coords=cell(1,ncoords);
    for ic=1:ncoords
        import_coords{ic}=sim.type_coords(:,1:ic);
    end
    [sim.stimspace,sim.stimspace_import_auxout]=rs_import_coordsets(import_coords,aux_stimspace);

    if sim.if_findrays
        [sim.stimspace_rays,wmsg,sim.stimspace_findrays_auxout]=rs_findrays(sim.stimspace.sas{1},[],ray_opts);
        disp(sprintf('ray structure created for paradigm %s',paradigm_name));
    else
        sim.stimspace_rays=struct();
        disp(sprintf('ray structure skipped for paradigm %s',paradigm_name));
    end
    sims.(paradigm_name)=sim;
end

Output:

ray structure created for paradigm Axes

set up a page for each paradigm, with space for multiple subplots, and plot the conceptual coordinates

fig_rows=length(subjs_disp)+1; %show 
fig_cols=ntransforms;
aux_stimdisp=struct;
for ip=1:length(paradigm_names)
    paradigm_name=paradigm_names{ip};
    sim=sims.(paradigm_name);

    if if_disp_coordsets %set up a page for each paradigm name
        figure;
        set(gcf,'Position',[100 100 1400 800]);
        set(gcf,'NumberTitle','off');
        set(gcf,'Name',paradigm_name);

        aux_stimdisp.opts_disp.fig_handle=gcf;
        aux_stimdisp.opts_disp.fig_name=paradigm_name;
        aux_stimdisp.opts_disp.axis_range='list';
        aux_stimdisp.opts_disp.axis_range_list=[-1 1]*(1+random_max);
        aux_stimdisp.opts_disp.callout_amount=0.3;
        aux_stimdisp.opts_disp.axis_labels=coord_labels;
        aux_stimdisp.opts_disp.legend_location='North';
        aux_stimdisp.opts_disp.set_labels='stims';
        aux_stimdisp.opts_disp.axis_handles={subplot(fig_rows,1,1)}; %show centered in column

        if sim.if_findrays
            aux_stimdisp.opts_disp_enh.if_rings=sim.if_rings;
            aux_stimdisp.opts_disp_enh.if_points=1; %so legends are set labels
            aux_stimdisp.opts_disp.data_show_method='last'; %last point is random; plotting a point allows rs_disp_enh_coordsets to make simple legend
            sim.stimspace_disp_auxout=rs_disp_enh_coordsets(sim.stimspace,aux_stimdisp,sim.stimspace_rays);
        else
            aux_stimdisp.opts_disp.data_show_method='all'; 
            aux_stimdisp.opts_disp.data_label_method='list';
            aux_stimdisp.opts_disp.data_label_list=find(~contains(sim.stimspace.sas{1}.typenames,' ')); %label only the on-axis points (off-axis point names all have blanks)
            sim.stimspace_disp_auxout=rs_disp_coordsets(sim.stimspace,aux_stimdisp);
        end
       drawnow;
    end 
    sims.(paradigm_name)=sim;
end

apply each subject's transform to the stimuli in each stimulus set, and display

for ip=1:length(paradigm_names)
    paradigm_name=paradigm_names{ip};
    sim=sims.(paradigm_name);

    for it=1:ntransforms
        transform_name=transform_names{it};
        xform_subj=cell(1,nsubjs);
        for is=1:nsubjs
            xforms=struct;
            xforms.ts{1}{ncoords}=transforms_noisy{is}.(transform_name);
            opts_xform=struct;
            opts_xform.class=transform_classes{it};
            xform_subj{is}=rs_xform_apply(sim.stimspace,xforms,setfield(struct,'opts_xform',opts_xform));

fill in lower and higher dimensions and add noise

            for ic=1:ncoords_tot
                xform_subj{is}.ds{1}{ic}=noise_add(is)*randn(sim.nstims,ic)+...
                    [xform_subj{is}.ds{1}{ncoords}(:,1:min(ic,ncoords)),zeros(sim.nstims,max(0,ic-ncoords))];
            end

adjust metadata

            sets=xform_subj{is}.sets{1};
            sets.dim_list=[1:ncoords_tot]; %since we added lower and higher dimensions
            sets.subj_id=sprintf('%s%s','subject ',zpad(is,2));
            sets.subj_id_short=sprintf('%s%s','s',zpad(is,2));
            sets.label_long=sprintf('simulation: %s',transform_names{it});
            sets.label=transform_names{it};
            xform_subj{is}.sets{1}=sets;
        end

concatenate datasets aross subjects

        dataspace=xform_subj{1};
        for is=2:nsubjs
            dataspace=rs_concat_coordsets(dataspace,xform_subj{is});
        end
        sim.dataspace.(transform_name)=dataspace;

plot

        aux_datadisp=struct;
        if if_disp_coordsets
            for is_ptr=1:length(subjs_disp)
                is=subjs_disp(is_ptr);
                aux_datadisp.opts_disp=sim.stimspace_disp_auxout.opts_disp; %starting point for plot is how the transforms were plotted

aux_datadisp.opts_disp.axis_handles={subplot(fig_rows,fig_cols,it)}; %show each transform in a separate column

                aux_datadisp.opts_disp=rmfield(aux_datadisp.opts_disp,'axis_labels'); %use default labels
                aux_datadisp.opts_disp.set_labels=sprintf('subj %1.0f %s',is,transform_name);
                aux_datadisp.opts_disp.set_select=is; %just show this subject
                aux_datadisp.opts_disp.set_colors={'k'}; %black

                figure(aux_datadisp.opts_disp.fig_handle); %activate the figure for this paradigm type
                aux_datadisp.opts_disp.axis_handles={subplot(fig_rows,fig_cols,it+is_ptr*fig_cols)}; %show each transform in a separate column

                if sim.if_findrays %setups for which points to label, etc, are inherited from sim.xformspace_disp_auxout
                    aux_datadisp.opts_disp_enh=sim.stimspace_disp_auxout.opts_disp_enh; %had if_rings info
                    sim.dataspace_disp_auxout.(transform_name){is}=rs_disp_enh_coordsets(dataspace,aux_datadisp,sim.stimspace_rays);
                else
                    sim.dataspace_disp_auxout.(transform_name){is}=rs_disp_coordsets(dataspace,aux_datadisp);
                end
                drawnow;
             end %is
         end
         disp(sprintf('dataspace created for paradigm %20s and transform %s',paradigm_name,transform_name));
    end %it
    sims.(paradigm_name)=sim;
end

Output:

dataspace created for paradigm                 Axes and transform affine

optinally align and knit the stimuli across paradigms

if if_knit
    check_nowarn=setfield(struct,'opts_check',setfield(struct,'if_warn',0));
    sims.knitted=struct;
    aux_align=struct;
    aux_align.opts_align.if_log=0;
    aux_knit=struct;
    aux_knit.opts_knit.if_log=0;

    disp(' ');
    for ip=1:length(paradigm_names)
        paradigm_name=paradigm_names{ip};
        stim_data=sims.(paradigm_name).stimspace;
        if ip==1
            concat=stim_data;
        else
            concat=rs_concat_coordsets(concat,stim_data,check_nowarn);
        end
    end
    disp('aligning stimuli across paradigms');
    aligned=rs_align_coordsets(concat,aux_align);
    disp('knitting data across paradigms');
    knitted=rs_knit_coordsets(aligned,aux_knit);
    sims.knitted.stimspace=knitted;

for each transformation and subject, align and knit the data across paradigms

    sims.knitted.dataspace=struct();
    for it=1:ntransforms
        transform_name=transform_names{it};
        concat_knitted=struct;
        for is=1:nsubjs
            concat=struct;
            for ip=1:length(paradigm_names)
                paradigm_name=paradigm_names{ip};
                subj_data=rs_extract_coordsets(sims.(paradigm_name).dataspace.(transform_name),is); %extract this subject's data
                if ip==1
                    concat=subj_data;
                else
                    concat=rs_concat_coordsets(concat,subj_data,check_nowarn);
                end
            end
            disp(' ');
            disp(sprintf('aligning data from subject %1.0f, transform %s, across paradigms',is,transform_name))
            aligned=rs_align_coordsets(concat,aux_align);
            disp(sprintf('knitting data from subject %1.0f, transform %s, into a single dataset',is,transform_name))
            knitted=rs_knit_coordsets(aligned,aux_knit);
            if is==1
                concat_knitted=knitted;
            else
                concat_knitted=rs_concat_coordsets(concat_knitted,knitted);
            end
        end
        sims.knitted.dataspace.(transform_name)=concat_knitted;
    end
end %if_knit

fit models to transformation between stimulus space and data space

paradigms_all=fieldnames(sims); %includes original paradigms and knitted
if ~exist('opts_geof') opts_geof=struct; end
opts_geof=filldefault(opts_geof,'model_list',model_list);
opts_geof=filldefault(opts_geof,'dimpairs_method','all');
opts_geof=filldefault(opts_geof,'if_stats',1);
opts_geof=filldefault(opts_geof,'nshuffs',20);
opts_geof=filldefault(opts_geof,'if_nestbymodel',-1);
opts_geof=filldefault(opts_geof,'if_nestbydim',-1);
opts_geof=filldefault(opts_geof,'if_log',0);
opts_geof=filldefault(opts_geof,'if_fit_summary',0);
aux_geof=struct;
aux_geof.opts_geof=opts_geof;

gfs=cell(ntransforms,length(paradigms_all));
xs=cell(ntransforms,length(paradigms_all));
aux_geof_out=cell(ntransforms,length(paradigms_all));
for it=1:ntransforms
    transform_name=transform_names{it};
    disp(' ');
    for ip=1:length(paradigms_all)
        paradigm_name=paradigms_all{ip};
        disp(sprintf('modeling transform %20s from stimulus space to subject space with paradigm %20s',transform_name,paradigm_name));
        data_in=sims.(paradigm_name).stimspace;
        data_out=sims.(paradigm_name).dataspace.(transform_name);
        [gfs{it,ip},xs{it,ip},aux_geof_out{it,ip}]=rs_geofit(data_in,data_out,aux_geof);
    end
end
sims.transform_names=transform_names;
sims.paradigm_names=paradigm_names;
sims.paradigms_all=paradigms_all;
sims.nsubjs=nsubjs;
sims.gfs=gfs;
sims.xs=xs;
sims.aux_geof_out=aux_geof_out;

Output:

modeling transform               affine from stimulus space to subject space with paradigm                 Axes

sims.ncoords=ncoords;
sims.ncoords_noise=ncoords_noise;
sims.noise_transform_mag=noise_transform_mag;
sims.noise_transform_subj=noise_transform_subj;
sims.noise_transform=noise_transform;
sims.noise_add_mag=noise_add_mag;
sims.noise_add_subj=noise_add_subj;
sims.noise_add=noise_add;
sims.model_list=model_list;

disp('consider saving the structure ''sims'', and using rs_toygeom_disp to display results')

Output:

consider saving the structure 'sims', and using rs_toygeom_disp to display results

End of included script rs_toygeom_sim.

geometric model fit display customizations

paradigms_fit_show={'Axes'};
subjs_fit_show=[1:nsubjs];
opts_dgeo=struct;
opts_dgeo.view=[-15 45];
opts_dgeo.if_nestbydim_in_show=0;

display model-fitting results (rs_toygeom_disp)

Included script: rs_toygeom_disp

rs_toygeom_disp: display the results of rs_toygeom_sim with `rs_disp_geofit`

sims: a structure containing simulations, geometric model fits, and key parameters

'sims' must already be in the workspace, created by rs_toygeom_sim, which the rs_toygeom_scenario* scripts run first. Display options (opts_dgeo, transforms_fit_show, paradigms_fit_show, subjs_fit_show) are read from the workspace when they are defined.

This script has no documentation page of its own: each rs_toygeom_scenario* page shows it where the scenario runs it.

See also: rs_toygeom_sim, rs_disp_geofit

if ~exist('sims')
    error('no sims structure found: run one of the rs_toygeom_scenario* scripts, or rs_toygeom_sim, first');
end
if ~exist('opts_dgeo') opts_dgeo=struct; end
aux_dgeo=struct;
aux_dgeo.opts_dgeo=opts_dgeo;
if ~exist('transforms_fit_show') transforms_fit_show=sims.transform_names; end
if ~exist('paradigms_fit_show') paradigms_fit_show=sims.paradigms_all; end
if ~exist('subjs_fit_show') subjs_fit_show=[1:sims.nsubjs]; end

for it=1:length(sims.transform_names)
    transform_name=sims.transform_names{it};
    for ip=1:length(sims.paradigms_all)
        paradigm_name=sims.paradigms_all{ip};
        for is=1:sims.nsubjs
            if ~isempty(strmatch(transform_name,transforms_fit_show,'exact'))  & ~isempty(strmatch(paradigm_name,paradigms_fit_show,'exact')) & ismember(is,subjs_fit_show)
                aux_dgeo_out=rs_disp_geofit(sims.gfs{it,ip}{is}.gf,aux_dgeo);
                fig_handles=aux_dgeo_out.opts_dgeo.fig_handles;
                fig_names=aux_dgeo_out.opts_dgeo.fig_names;
                for ifig=1:length(fig_handles)
                    figure(fig_handles{ifig});
                    set(gcf,'Name',cat(2,fig_names{ifig},sprintf(' subj %1.0f',is),' ',transform_name,' ',paradigm_name));
                    if exist('scenario_name')
                        axes('Position',[0.80,0.05,0.01,0.01]); %for text
                        text(0,0,scenario_name,'Interpreter','none');
                        axis off;
                    end
                    axes('Position',[0.50,0.05,0.01,0.01]); %for text
                    text(0,0,fig_names{ifig},'Interpreter','none');
                    axis off;
                    axes('Position',[0.50,0.03,0.01,0.01]); %for text
                    text(0,0,sprintf('transform: %s, paradigm %s',transform_name,paradigm_name),'Interpreter','none');
                    axis off;
                    axes('Position',[0.50,0.01,0.01,0.01]);
                    text(0,0,sprintf('subject %2.0f: transform noise %4.2f additive noise %4.2f',is,sims.noise_transform(is),sims.noise_add(is)));
                    axis off;
               end 
            end %select
        end %subject
    end %paradigm name
end %transform

End of included script rs_toygeom_disp.