refactor continues

RidgeGLM.m

@@ -1,22 +1,19 @@
-% CovGLM sub-class for regularised ridge fits.
-% gl = RidgeGLM(X,data,k,[unscale],[covariatedegree]);
-classdef RidgeGLM < CovGLM
+% GLM sub-class for regularised ridge fits.
+% gl = RidgeGLM(X,data,k,[unscale])
+classdef RidgeGLM < GLM
     properties
         k % regularisation parameter (gets scaled by k*nsamples in ridgevec)
         unscale % return data to original scale (default 0)
     end
 
     methods
-        function gl = RidgeGLM(X,data,k,unscale,covariatedegree)
+        function gl = RidgeGLM(X,data,k,unscale)
             if nargin == 0
                 % support no input case
                 [X,data,k] = deal([]);
             end
             % initialise with super-class constructor
-            if ieNotDefined('covariatedegree')
-                covariatedegree = [];
-            end
-            gl = gl@CovGLM(X,data,covariatedegree);
+            gl = gl@GLM(X,data);
             gl.k = k;
             if ieNotDefined('unscale')
                 unscale = 0;

containers/BaseVolume.m

@@ -98,12 +98,12 @@
                 metafeatures);
         end
 
-        function processchunks(self,fname,varargin)
+        function filterbychunk(self,fname,varargin)
         % general method for fevaling some fname which takes sample data
         % from each chunk as a first input and anything else in the
         % following.
         %
-        % processchunks(self,fname,[varargin])
+        % filterbychunk(self,fname,[varargin])
             for c = 1:self.desc.samples.nunique.chunks
                 chunkind = self.meta.samples.chunks == ...
                     self.desc.samples.unique.chunks(c);
@@ -112,28 +112,15 @@ function processchunks(self,fname,varargin)
             end
         end
 
-        function medianfilter(self,n)
-        % medianfilter(n)
-        % filter the data in place along the sample dimension (time) with
-        % filter size n. Operates separately on each chunk.
-            processchunks(self,'medianfilter',n);
-        end
-
         function sgdetrend(self,k,f)
             % use a Savitzky-Golay filter to detrend the data across time.
             % Operates separately on each chunk
-            processchunks(self,@(data,k,f)data-sgolayfilt(data,k,f,[],...
+            filterbychunk(self,@(data,k,f)data-sgolayfilt(data,k,f,[],...
                 1),k,f);
         end
 
         function sgfilter(self,k,f)
-            processchunks(self,'sgolayfilt',k,f,[],1);
-        end
-
-        function zscore(self)
-            % Zscore the data across time, operating separately on each
-            % chunk.
-            processchunks(self,'zscore',[],1);
+            filterbychunk(self,'sgolayfilt',k,f,[],1);
         end
 
         function checkmeta(self)

decoding/roidata_lindisc.m

@@ -4,12 +4,6 @@
 %
 % Named inputs:
 %
-% (for vol2glm_batch)
-% sgolayK: degree of optional Savitsky-Golay detrend 
-% sgolayF: filter size of optional Savitsky-Golay detrend
-% covariatedeg: polynomial detrend degree (or 'adaptive')
-% targetlabels: labels to explicitly include (default all)
-% ignorelabels: labels to explicitly exclude (default none)
 % glmclass: char defining CovGLM sub-class (e.g. 'CovGLM')
 % glmvarargs: any additional arguments for GLM (e.g. k for RidgeGLM)
 %
@@ -21,10 +15,9 @@
 % [res,nulldist,bootdist] = roidata_lindisc(rois,designvol,epivol,contrasts,varargin)
 function [res,nulldist,bootdist] = roidata_lindisc(rois,designvol,epivol,contrasts,varargin)
 
-ts = varargs2structfields(varargin,struct('sgolayK',[],'sgolayF',[],...
-    'split',[],'covariatedeg','adaptive','targetlabels',[],...
-    'ignorelabels','responses','glmclass','CovGLM','glmvarargs',[],...
-    'nperm',1,'nboot',0,'usegpu',false));
+ts = varargs2structfields(varargin,struct(...
+    'split',[],'glmclass','GLM','glmvarargs',[],...
+    'nperm',1,'nboot',0));
 
 if ~iscell(ts.split)
     % so we can easily deal to cvgroup field later
@@ -72,13 +65,9 @@
     % (we will need to strip out 0 cons eventually for speed - maybe move
     % this inside contrast loop and apply a skipcon that gets applied just
     % before volume stage?)
-    model = vol2glm_batch(designvol,epivol(:,validvox),...
-        'sgolayK',ts.sgolayK,'sgolayF',ts.sgolayF,'split',[],...
-        'covariatedeg',ts.covariatedeg,'targetlabels',ts.targetlabels,...
-        'ignorelabels',ts.ignorelabels,'glmclass',ts.glmclass,...
-        'glmvarargs',ts.glmvarargs,'usegpu',ts.usegpu);
+    model = vol2glm(designvol,epivol(:,validvox),ts.glmclass,...
+        ts.glmvarargs{:});
     % nb model can still have multiple array entries
-    model = model{1};
     [model.cvgroup] = ts.split{:};
     res.nfeatures(r) = model.nfeatures;
 

rsa/roidata2rdmvol_lindisc.m

@@ -11,7 +11,7 @@
 %
 % named varargs:
 %
-% glmclass: char defining CovGLM sub-class (e.g. 'CovGLM')
+% glmclass: char defining GLM class (default GLM)
 % glmvarargs: any additional arguments for GLM (e.g. k for RidgeGLM)
 % split: indices to define GLM cvgroups (if crossvalidate) - NB NOT used in
 %   vol2glm_batch
@@ -29,7 +29,7 @@
 function disvol = roidata2rdmvol_lindisc(rois,designvol,epivol,varargin)
 
 ts = varargs2structfields(varargin,struct(...
-    'split',[],'covariatedeg',[],...
+    'split',[],...
     'glmclass','GLM','glmvarargs',{},'sterrunits',false,'crossvalidate',...
     false,'minvoxeln',0,'batchsize',5000));