Back to PLS HelpVarious errors when trying to run PLS analysis
Posted on 06/28/12 14:11:56
Number of posts: 15
Hi there,
I've been trying to run a mean centered PLS analysis but the permutation progress bar either:
1) Sticks at 20% on "Computing Permutations" when run from the gui with these errors:
??? Error using ==> reshape
Size arguments must be real integers.
Error in ==> rri_perm_order at 57
tmp = reshape([first:last],num_subj_lst(g),num_cond);
Error in ==> fmri_deviation_perm_test at 103
perm_order = rri_perm_order(subj_group,num_conditions,num_perm);
Error in ==> fmri_pls_analysis at 323
[brainlv,s,designlv,b_scores,d_scores,perm_result,lv_evt_list] = ...
Error in ==> fmri_analysis_ui>ExecutePLS at 1928
fmri_pls_analysis(PLSoptions.profiles, [], ...
Error in ==> fmri_analysis_ui at 93
ExecutePLS;
??? Error using ==> waitfor
Error while evaluating uicontrol Callback
2) Sticks at 20% on "Working with PLS" when run via command line with these errors:
??? Subscript indices must either be real positive integers or logicals.
Error in ==> fmri_deviation_perm_test>gen_grp_dev_data at 210
g_evt_list(g_range) = evt_list(g_range);
Error in ==> fmri_deviation_perm_test at 59
dev_data = gen_grp_dev_data(st_datamat,num_conditions,evt_list,subj_group);
Error in ==> fmri_pls_analysis at 323
[brainlv,s,designlv,b_scores,d_scores,perm_result,lv_evt_list] = ...
Error in ==> batch_pls_analysis at 363
fmri_pls_analysis(PLSoptions.profiles, PLSoptions.ContrastFile, ...
Error in ==> batch_plsgui at 36
batch_pls_analysis(fid);
Can you recommend a course of action for me?
Thanks,
Niron Khan
Replies:
Untitled Post
I'm Online
Posted on 06/28/12 14:19:29
Number of posts: 291
This means the problem in your data. If you could upload all session / datamat files to somewhere I can access, and post the link here, I could take a look into your problem. Also tell me which session files belong to which group. Please DO NOT attach anything to the email.
Untitled Post
Posted on 07/03/12 13:51:30
Number of posts: 15
Hi Jimmy,
I have compressed our datamat/session profiles into a zip archive and uploaded them to 4shared.com: http://www.4shared.com/zip/kA6deabQ/Archive.html?
We have 18 session profile/datamat pairs since subject 5 was dropped.
Untitled Post
Posted on 07/03/12 13:53:48
Number of posts: 15
Hi Jimmy,
I have compressed our datamat/session profiles into a zip archive and uploaded them to 4shared.com: http://www.4shared.com/zip/kA6deabQ/Archive.html?
We have 18 session profile/datamat pairs since subject 5 was dropped.
They are all in one group as we are comparing conditions.
Thanks a lot :)
Untitled Post
I'm Online
Posted on 07/03/12 14:39:15
Number of posts: 291
Seems that some datamats were not created properly, and it could be a bug in PLS gui program. Would you please send me 3 more files below:
Subject09run2TT.nii
Subject09run3TT.nii
Subject09run4TT.nii
I need to use them to identify the problem. Thanks.
Untitled Post
Posted on 07/04/12 14:05:34
Number of posts: 15
Hi Jimmy,
The .nii files are here: http://www.4shared.com/zip/ayGzvydU/ArchiveSubject9nii.html?
Thanks so much for the help :)
Niron
Untitled Post
Posted on 07/04/12 14:17:36
Number of posts: 15
We also ran the analysis with subjects 1,2,3,4,6,7 and 8 just now and it was completed without errors. However, we also tried to run the analysis with all participants minus subject 9 and got the same errors as before. Thought this might be useful information.
Niron
Untitled Post
I'm Online
Posted on 07/04/12 15:25:24
Number of posts: 291
The problem is in the onsets value that you input in the session file.
I use subject 9 as example, the total number of scans is 269 volumes. i.e. the valid time points is ranging from 0 to 268. If you keep the default temporal window size as 8, the valid onsets values are ranging from 0 to 261.
In run1 of condition 1, 2, 4, you only input 1 value, which are 280, 282, 284. They are all exceed the range, so there is no value at all for that condition. In run 2 & run 3, the condition 4 also exceed the range. Therefore, even if you average across run, the condition 4 will never get any value, and this caused problem.
By the way, besides subject 9, you also have the same problem in subject 10, 11, 15, 18, 19. If you exclude all those subjects that I mentioned, you will not have problem when you run analysis. However, I still insist that you check all Onsets values, because it doesn't make sense to put time points 284 for 269 run.
I use subject 9 as example, the total number of scans is 269 volumes. i.e. the valid time points is ranging from 0 to 268. If you keep the default temporal window size as 8, the valid onsets values are ranging from 0 to 261.
In run1 of condition 1, 2, 4, you only input 1 value, which are 280, 282, 284. They are all exceed the range, so there is no value at all for that condition. In run 2 & run 3, the condition 4 also exceed the range. Therefore, even if you average across run, the condition 4 will never get any value, and this caused problem.
By the way, besides subject 9, you also have the same problem in subject 10, 11, 15, 18, 19. If you exclude all those subjects that I mentioned, you will not have problem when you run analysis. However, I still insist that you check all Onsets values, because it doesn't make sense to put time points 284 for 269 run.
Untitled Post
Posted on 07/04/12 16:19:30
Number of posts: 15
Ah, that explains it.
To be honest, those were intentional, as for some conditions we did not have any onsets in our behavioural data and PLS requires at least one onset per condition. So when we had no values for a condition we would put in a value such as 280 which we thought would have been dropped from the datamats since it says "excluding run due to out of bounds" when PLS creates the datamats.
Is there a way to leave a condition without any onsets since leaving it blank doesn't work?
Thanks a lot
Niron
Untitled Post
I'm Online
Posted on 07/04/12 16:31:40
Number of posts: 291
What you did is exactly correct, although I suggest people to use -1 just for simplicity. Detail has already been described in FAQ.txt in the package.Search for question "I have more than one runs in my MRI study. However, for some runs, I do not have all the conditions. How should I fill out the onset field?". In the answer, I emphasized that " ...... You can not choose "Within each run only", since you do not have all the conditions filled for every runs. In additon, for each condition, you must have at least one valid onset (not -1) in one of your runs." The condition 4 of your subject 9 does not meet this requirement, so it means that condition 4 should not be considered in the datamat and used in the analysis for comparison. Please check other problematic subjects, and make sure that at least one valid onset in one of your runs. Also, if one condition is removed from one subject (e.g. condition 4 in subject 9), it (condition4) has to be removed from all subjects, because we have to make condition exactly the same for one analysis.
Untitled Post
Posted on 07/10/12 13:36:56
Number of posts: 15
We deleted condition 4 from the datamats as you suggested and the analysis runs perfectly for all of our subjects now.
Thanks a lot for the help Jimmy!
Untitled Post
Posted on 07/17/12 14:52:31
Number of posts: 15
Hi there,
Just a quick question concerning the dataset. We ran the analysis using 100 bootstraps and a minimum different subject value of 50% (the default) and obtained good results (p values of .0140 and .0678 for our lv's of interest) but I noticed that the values become far more significant when using 70% (from .0140 to .0022 and .0678 to .0345). For a sample of 18 subjects what is the best value to use as I do not want to affect the validity of the study?
Niron
Untitled Post
I'm Online
Posted on 07/17/12 15:02:39
Number of posts: 291
Please use 50% only. This is a experimental feature, and is going to be removed in the new release of PLS.
Login to reply to this topic.
