These are chat archives for epnev/ca_source_extraction

20th
Jul 2017
eftychios pnevmatikakis
@epnev
Jul 20 2017 13:31
@shijiegu glad to hear this!
Hi @Jan10e sorry for the late reply due to traveling. The demo_memmap.m does not have a file since it is meant to work on large files and we don't have any files like that available. However, you can see some other demos that use memory mapping here especially this one that works on the publicly available 02.00 neurofinder dataset. For the the error you provide I'll need more context to understand where it occurs, what file you're using etc.
eftychios pnevmatikakis
@epnev
Jul 20 2017 13:41
Hi @sppatankar we have more or less abandoned the usage of FISTA in our pipeline. Our initialization occurs through the function initialize_components.m If you can modify the FISTA approach to give the outputs in that format then it should be easy to integrate with our framework. I'll be happy to incorporate these changes in our framework.
eftychios pnevmatikakis
@epnev
Jul 20 2017 14:00
Hi @jkung1995 If you just have one neuron all you need is to separate the activity of the neuron from any possible neuropil contamination. Most rank-2 factorization approaches (ICA, NMF) should work well.
sppatankar
@sppatankar
Jul 20 2017 15:37
Thanks for your reply @epnev. I had another quick question, could you point me to the file/functions that deal with the case when two neurons overlap on top of each other. Or is this special case sort of dealt with by the initialization methods?
eftychios pnevmatikakis
@epnev
Jul 20 2017 20:03
@sppatankar update_spatial_components deals with finding possibly overlapping components but in essence it's the whole idea of factorization and not a specific function.
Jan10e
@Jan10e
Jul 20 2017 20:22

@epnev Thanks for your response. I am going to try out the large datasets.
About the error message in run_pipeline.m, I get this as soon as I run the script:

>> run_pipeline
Starting parallel pool (parpool) using the 'local' profile ... connected to 6 workers.
The file was loaded successfully. Elapsed time : 0.172 s.
Warning: File /media/jantine/Data/02_CaAnalysis/Epnev/ca_source_extraction/demoMovie_nr.h5already exists. Saving motion
corrected file as/media/jantine/Data/02_CaAnalysis/Epnev/ca_source_extraction/demoMovie_nr_20170720T161950.h5 
> In normcorre_batch (line 208)
  In run_pipeline (line 28) 
Template initialization complete. 
Error using shift_reconstruct (line 42)
Matrix dimensions must agree.

Error in normcorre_batch (line 260)
        parfor ii = 1:lY

Error in run_pipeline (line 28)
            [M,shifts,template] = normcorre_batch(fullname,options_nonrigid,template);

Does this help? Or would you like to have more information?
Cheers!

Jan10e
@Jan10e
Jul 20 2017 21:08

@epnev I am trying out the demo_CSHL2017_neurofinder.02.00.m and need the function read_neurofinder:

Undefined function or variable 'read_neurofinder'.

Error in demo_CSHL2017_neurofinder_02_00 (line 17)
    Ycon = read_neurofinder(foldername,fullfile(foldername,'neurofinder0200.tif'));

Where can I find the function? I've been looking in the repo of ca_source_extraction' and 'neurofinder but couldn't find it.

eftychios pnevmatikakis
@epnev
Jul 20 2017 21:43
@Jan10e it's in the repo
eftychios pnevmatikakis
@epnev
Jul 20 2017 21:49
@Jan10e I made some changes in the motion correction code a few weeks ago. If you downloaded the code before that you'll need to update it. The error might be due to that
shijiegu
@shijiegu
Jul 20 2017 22:10
@epnev In motion correction, there is an iteration, I wonder what is changed after each iteration.
eftychios pnevmatikakis
@epnev
Jul 20 2017 22:17
@shijiegu In short datasets, it might be desirable to go over the dataset more than once just to have more time to get a better template. But I rarely use it in practice.
shijiegu
@shijiegu
Jul 20 2017 22:18
but each one is independent?