Thanks for the information Chris, that's a smart idea. I'll have a look into this, though as you say it would be a bit inefficient, as I'd need to run this multiple times if I had a large dataset.
________________________________________
From: PRoNTo users <[log in to unmask]> on behalf of Christophe Phillips <[log in to unmask]>
Sent: 28 October 2013 11:49
To: [log in to unmask]
Subject: Re: Testing new data on trained model
Hi James,
We had a meeting last week and we are planning a new version for the new
year. Hopefully before May 2014 and the next course... No guarantee there!
For your case, you coudl of course hack thorough the code and the PRT
structure to do what you want. The idea of inputting a manually defined
'flexible CV' matrix stems from this.
One way to do it with the current version and GUI, albeit not very
efficient, would be to input your novel test data with a "random" score
as the last (or first) element of the data set into PRoNTo. This way,
when doing a leave-one-subject-out CV, at some point your novel data IS
the test data while the rest is the training date set... You could then
look at the predicted value. The rest of the CV would be useless of course.
Hope this helps,
Chris
Le 25/10/2013 18:29, Cole, James a écrit :
> OK, thanks for the speedy response Joćo. Any ideas when the next version is due?
> Do you think it's possible to take the output of Pronto and hack it about myself in matlab to achieve the aim of testing on novel data? Presumably the details of the trained model are stored in the PRT.mat somewhere?
> James
> ________________________________________
> From: De Matos Monteiro, Joao
> Sent: 25 October 2013 17:04
> To: PRoNTo users
> Cc: Cole, James
> Subject: Re: Testing new data on trained model
>
> Dear James,
>
> Unfortunately, it is not possible to do that in the current version of
> PRoNTo.
> However, we will include it on the next version.
>
> Regards,
> Joćo
>
>
> On 25/10/13 14:38, James Cole wrote:
>> Dear Pronto developers,
>> I'm trying to use Pronto to predict age using grey matter segmentations and get a reasonably good prediction accuracy (r=0.87). However, I'm stuck at the next stage. How do I use the trained model to get an image predict on a new subject? I basically want to take another set of images that weren't involved in the original training and testing and get out their predicted ages, so I can see how well the model generalises.
>> Thanks very much for the help,
>> James
>>
> --
> Joćo Matos Monteiro
> PhD Student
> Computer Science Department,
> University College London,
> Gower Street, WC1E 6BT, London
>
>
|
