[Eeglablist] running AMICA with multiple models for subject data

Hauge,Theresa C theresa.hauge at ufl.edu
Wed Jun 19 19:00:44 PDT 2024


Hello Seyed,

Yes, I am happy to provide the results of my run of AMICA within our HPC! When operating within a Matlab session on HPC resources, I changed the input so that a batch script was submitted to the HPC. (%% NOTE: check the runamica17_nsg.m function to update filepaths to your organizations resources).

EEG = runamica17_nsg(EEG, 'outdir', outdir, 'do_reject', 1, 'numrej', 5, 'rejsig', 4, 'num_models', 3,'batch',1);
Batch_script.sh
Nodes=1
Ntasks=32
Ntasks-per-node=32
Cpus-per-task=4
Mem=249208M
Time=04:00:00

AMICA run results, HPC session (batch job submission)

  1.  Num_chans = 111
  2.  Data_points = 2084354 (~70 minutes of data, EEG.srate=500Hz)
  3.  Number of models = 3
  4.  Step time = For each iteration (max_iterations=2500), average step time= ~1.5s
  5.  Overall time = 1 hour

I will update to MATLAB 2022b and run the postAmicaUtility functions within that version, and can post any issues I may encounter to the GitHub. Thank you for all of your help!


Best,
            Theresa

Theresa C. Hauge, M.A.
She/Her/Hers
PhD Candidate, Applied Physiology & Kinesiology
Neuromotor Behavior Laboratory
University of Florida
theresa.hauge at ufl.edu
@TheresaCHauge


From: Seyed Yahya Shirazi <shirazi at ieee.org>
Sent: Wednesday, June 19, 2024 16:40
To: Hauge,Theresa C <theresa.hauge at ufl.edu>
Cc: EEGLAB List <eeglablist at sccn.ucsd.edu>
Subject: Re: [Eeglablist] running AMICA with multiple models for subject data

[External Email]
Hi Theresa,

I am very glad you could run AMICA 1.7 on your HPC. If you don't mind sharing the results of your run, including (1) number of channels, (2) number of data points, (3) number of models, (4) step time (one showing at each iteration), and (5) overall time. I think it would be inspiring for others to have a data point on AMICA's performance on HPC.

Regarding postAmicaUtilty, I have had similar issues, and those were mostly related to the platform and Matlab version. Do you mind trying loading the results on your local PC with Matlab 2022b and let us know if you observe a similar issue? You can directly post the potential issues on GitHub: https://urldefense.com/v3/__https://github.com/sccn/postAmicaUtility/issues__;!!Mih3wA!AupcELh21I3fYZCcB48pLTS7Dh2wj1EXKkJ8uq9jhEWoxN1EeHhpa2dahk22m0yDiLo7P4bXf02BpD-sQoGVbeM7FLpzFg$ 

Best,
Seyed


More information about the eeglablist mailing list