HI James
A minimum of 10 is the rule of thumb. It is not really possible to give a better estimate because it will depend on the data. In my personal experience, much more than 10 is required. I would start with 10 and train FIX using the leave-one-out (LOO) option so that you can get an estimate of performance by looking at the LOO true-positive and true-negative rate. This will give you an idea if more training data is required.
I agree that you should try and make the training data as representative of the full data as you can by including a variety of subjects/sessions/tasks etc.
Cheers, Sean
> On 7 Mar 2019, at 13:57, James Livermore <[log in to unmask]> wrote:
>
> Hi all,
>
> We're planning on running FIX and are hand labelling some of our data now for a training dataset - we have in total 24 subjects x 2 sessions x 3 tasks plus resting states in each session. I'm guessing that a selection of each task and session is best, but is there a decent rule of thumb on how many hand-labelled sets we need to give it? It says minimum 10 subjects in the user guide but does that mean of each task, of each session etc?
>
> Thanks,
> James
>
> ########################################################################
>
> To unsubscribe from the FSL list, click the following link:
> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|