Hi Benedetta,
1) In v2.1, the dimensionalities of the modalities combined should be
exactly the same, including second-level masks. That also means that
PRoNTo will ignore second-level masks if they were not entered for all
modalities. You should have a common mask for all your modalities (1st
and 2nd level) if you want to combine them.
2) The matrix on the left of the CV panel represents your different
samples to the model. When you use MKL, you treat different modalities
as different features, so the number of samples is unchanged. In the
implementation, we extract the sample matrix from the first modality and
do all computations (CV, ...) based on that matrix only.
HTH,
Best,
Jessica
On 03/08/2018 16:27, SUBSCRIBE PRONTO-USERS Benedetta Vai wrote:
> Dear PRoNTO developers,
>
> I am performing multimodal machine learning (MKL on 5 Modalities) in order to classify two groups by using beta version 2.1
>
> I encountered some issues, I tried to solve them looking in the previous posts and reading the manual but unsuccessfully.
>
> 1) I entered 2n level masks in the feature set for some modalities (4/5) (not flagging Build One Kernel Per region, and the mask was in the same space of all images), nevertheless the results are exactly the same compared to an identical feature set but no masked, looking in PRT.mat I was not able to find the mask. The mask is the skeleton mask for mean FA DTI maps thus should eliminate a lot of noise. When I look the weights for that modality seems that no masks have been applied and the kernels look identical. The models were MKL with opt. hp 0.1 1 10 100 1000, with k folds subjects per group 10 for both inner and outer cv. Data were centered and normalized. I am probably making a mistake but I don’t know where.
>
> 2) When I checked my model in review CV, despite I had selected MKL and had 5 modalities, the column for modality was all white, without indicating me different modalities. Is it right?
>
> 3) Concerning covariates, I did not found GLM option into ‘Specify Model’ as previously suggested in the post 'Covariates in Pronto, new update?'. Is it right to use ‘regress out covariates (subject level) for CV data? I use one hot encoding.
>
> Thanks for your attention, every help will be very appreciate.
>
> Best,
> Benedetta
>
> ########################################################################
>
> To unsubscribe from the PRONTO-USERS list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=PRONTO-USERS&A=1
########################################################################
To unsubscribe from the PRONTO-USERS list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=PRONTO-USERS&A=1
|