Dear Eva,
I can't answer your first question without playing with the data
unfortunately. It seems that the kernel is not well defined, which is
not a straightforward issue.
Re the LOSO/ 5-folds difference in correlation, again it is hard to say
if there is something wrong with your analysis without seeing exactly
what you did. Have you by any chance organized your data in one way or
another? For example by sorting according to the target? If so I would
suggest shuffling that order. Illustration: you sort your subjects by
age and predict age. In fold 1, the youngest subjects are left out, and
only the oldest are used to build the model. Your model will perform
worse than if it had a mix of old and new subjects to train on. However,
it could also be linked to the number of subjects you have in your data
set and could reflect that your model simply does not generalize well
(you should trust more the results of 5-folds than of LOSO).
Let us know if you'd like us to investigate your data (we need PRT.mat,
feature sets and kernels).
HTH,
Best,
Jessica
On 7/3/2017 12:50 PM, Eva Tolmeijer wrote:
> Dear PRoNTo users,
>
> For one of my analyses, I am able to run the GPR and KRR model but not the RVR model. I received the following error code:
>
> Model 'modelRVR' not found in PRT.mat. Creating...
> Updating PRT.mat.......>>
> Model 'modelRVR' found in PRT.mat.
> Model 'modelRVR' found in PRT.mat.
> Loading data files.....>>
> > running CV fold: 1 of 20 ...
> > running CV fold: 2 of 20 ...
> > running CV fold: 3 of 20 ...
> > running CV fold: 4 of 20 ...
> > running CV fold: 5 of 20 ...
> prt_machine: machine did not run sucessfully.
> SOLUTION: Please read the message below and attempt to correct the problem, or ask the developpers for assistance by copy-pasting all messages and explaining the exact steps that led to the problem.
> These kinds of issues are typically caused by Matlab path problems.
> 12 : function [@(hObject,eventdata)prt_ui_model('buildbutt_Callback',hObject,eventdata,guidata(hObject))] in file [] at line [0]
> 11 : function [prt_ui_model] in file [/Users/evatolmeijer/Documents/Pronto 2.01/prt_ui_model.m] at line [50]
> 10 : function [gui_mainfcn] in file [/Applications/MATLAB_R2015a.app/toolbox/matlab/guide/gui_mainfcn.m] at line [95]
> 9 : function [buildbutt_Callback] in file [/Users/evatolmeijer/Documents/Pronto 2.01/prt_ui_model.m] at line [1170]
> 8 : function [prt_cv_model] in file [/Users/evatolmeijer/Documents/Pronto 2.01/prt_cv_model.m] at line [101]
> 7 : function [prt_cv_fold] in file [/Users/evatolmeijer/Documents/Pronto 2.01/prt_cv_fold.m] at line [75]
> 6 : function [prt_machine] in file [/Users/evatolmeijer/Documents/Pronto 2.01/machines/prt_machine.m] at line [227]
> 5 : function [prt_machine_rvr] in file [/Users/evatolmeijer/Documents/Pronto 2.01/machines/prt_machine_rvr.m] at line [49]
> 4 : function [prt_rvr] in file [/Users/evatolmeijer/Documents/Pronto 2.01/machines/prt_rvr.m] at line [65]
> 3 : function [regression0] in file [/Users/evatolmeijer/Documents/Pronto 2.01/machines/prt_rvr.m] at line [88]
> 2 : function [rvr1a] in file [/Users/evatolmeijer/Documents/Pronto 2.01/machines/prt_rvr.m] at line [115]
> 1 : function [logdet] in file [/Users/evatolmeijer/Documents/Pronto 2.01/machines/prt_rvr.m] at line [533]
> Warning: Prediction method did not return [Error running machine prt_machine_rvr: MATLAB:posdef Error using
> chol
> Matrix must be positive definite.]
>> In prt_cv_fold (line 77)
> In prt_cv_model (line 101)
> In prt_ui_model>buildbutt_Callback (line 1170)
> In gui_mainfcn (line 95)
> In prt_ui_model (line 50)
> In @(hObject,eventdata)prt_ui_model('buildbutt_Callback',hObject,eventdata,guidata(hObject))
> Warning: model.type not specified, defaulting to classifier
>> In prt_stats (line 35)
> In prt_cv_model (line 114)
> In prt_ui_model>buildbutt_Callback (line 1170)
> In gui_mainfcn (line 95)
> In prt_ui_model (line 50)
> In @(hObject,eventdata)prt_ui_model('buildbutt_Callback',hObject,eventdata,guidata(hObject))
> Attempted to access stats.con_mat(0,2); index must be a positive integer or logical.
>
> Error in prt_stats>compute_stats_classifier (line 68)
> stats.con_mat(pred_lb,true_lb) = stats.con_mat(pred_lb,true_lb) + 1;
>
> Error in prt_stats (line 43)
> stats = compute_stats_classifier(model, tte, nk);
>
> Error in prt_cv_model (line 114)
> stats = prt_stats(model, targets.test, nc); %targets.train
>
> Error in prt_ui_model>buildbutt_Callback (line 1170)
> prt_cv_model(PRT, in);
>
> Error in gui_mainfcn (line 95)
> feval(varargin{:});
>
> Error in prt_ui_model (line 50)
> gui_mainfcn(gui_State, varargin{:});
>
> Error in @(hObject,eventdata)prt_ui_model('buildbutt_Callback',hObject,eventdata,guidata(hObject))
>
>
> Error while evaluating UIControl Callback
>
> I tried to restart my laptop, restart matlab, and performed the analysis multiple times but I keep getting the same error code. Does anyone have an idea what might be the problem?
>
> Besides this, I am only getting quite different results per model for one type of ROI analysis. For the GPR model using a LOSO-CV I get a correlation of 0.29 whereas a 0.02 for the same model using a 5-fold CV. Could this happen by chance or does this mean something is not running properly?
>
> Thank you for any help!
>
> Best wishes,
> Eva
|