DUNEuro WARNING: You are using AMG without SuperLU!

Discussions
1

Dear Francois,

When I run DUNEuro, there is a warning:
WARNING: You are using AMG without SuperLU! Please consider installing SuperLU, or set the usesuperlu flag to false to suppress this warning.
Do I need to install SuperLU, and whether SuperLU can improve the running speed?
My operating system is windows10. The following is the running log.

DUNEURO> System call: "C:\Users\ZZD.brainstorm\bst_duneuro\bin\bst_duneuro_meeg_win64.exe" "C:\Users\ZZD.brainstorm\tmp\duneuro_minifile.mini"
Reading 3d Gmsh grid...
version 2.2 Gmsh file detected
file contains 773970 nodes
file contains 4277455 elements
number of real vertices = 773970
number of boundary elements = 0
number of elements = 4277455
volume_conductor.time_reading_gmsh = 250.329
tensors = 5
time = 0
volume_conductor.time_reading_tensors = 0
volume_conductor.time_reordering_indices = 0.134641
volume_conductor.time = 250.464
degree = 1
element_type = tetrahedron
WARNING: You are using AMG without SuperLU! Please consider installing SuperLU, or set the usesuperlu flag to false to suppress this warning.
time = 0.0019932
time = 0.79987

This is a new model. Compute the transfer Matrix.
solver.electrode_1.time_rhs_assembly = 0
=== matrix setup (max) 5.7606 s

Thanks for your help.
Best,
Wu, Shuang

2

I used SimNIBS to generate FEM mesh from the subject's MRI. The DUNEuro running time of different people varies widely, ranging from 1 hour to dozens of hours. The CPU occupancy is very low, taking up only a single thread.

3

@tmedani @juangpc ?

4

Hello all,

superLU library can speed slightly the computation, but it's not really required.
If you are using Linux OS this message may not be displayed.
Improving the computation time with superLU and the multithreaded for Windows is under investigation.

If the subjects have a similar FEM mesh model (number of nodes), similar source space (number of dipoles), similar modality (EEG, MEG ...), and the same number of sensors, then the computation time should be in the same range.

Regarding the number of threads,
which OS are you using?
also, when did you download the Duneuro module for Brainstorm?

Best,

5

Thanks, temdani

My OS is Windows10.
I re-downloaded bst_duneuro on 2020-12-28. The modification date of the file is 2020-8-31.

Wu, Shuang

6

you're welcome,

If you are with windows, then you have the latest version of bst_duneuro.

Best,

7

Hi,
I am using ubuntu 20 from within a virtual box and I got the same message.

WARNING: You are using AMG without SuperLU! Please consider installing SuperLU, or set the usesuperlu flag to false to suppress this warning.

What should i do?

I generated FEM using brainstorm

Regards,
Saima Safdar

8

Hi,

This is just a warning message, the FEM computation will run even without the SuperLu library.
You can proceed and compute the forward model.
We will solve this issue and remove this message on the next release.

Thank you.

9

The program uses only single thread even with multiple cores. Its extremely slow even after 2 hours not getting the solution.

Mesh size is 334 mb

Code I am using is:
import duneuropy as dp
import simbiopy as sp
import numpy as np

read electrodes

electrodes = dp.read_field_vectors_3d('/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131/projCentroids.txt')

read dipoles

dipoles = dp.read_dipoles_3d('/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131/dipole.txt')

construct driver object

config = {
'type' : 'fitted',
'solver_type' : 'cg',
'element_type' : 'tetrahedron',
'volume_conductor' : {
'grid.filename' : '/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131/brain2mesh_mri.msh',
'tensors.filename' : '/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131/realistic_6c.cond'
}
}
driver = dp.MEEGDriver3d(config)
driver.setElectrodes(electrodes, {'type' : 'closest_subentity_center', 'codims' : [3] })

solution = driver.makeDomainFunction()
for dip in dipoles:
driver.solveEEGForward(dip, solution, {
'source_model.type' : 'partial_integration',
'solver.reduction' : 1e-8,
'post_process' : True,
'subtract_mean' : False})
lf = driver.evaluateAtElectrodes(solution)

lf -= np.mean(lf)

import os
folder_output = '/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131'

driver.write(solution,{
'format' : 'vtk',
'filename' :os.path.join(folder_output, 'case131_result_ubuntu'),
})

CPU usage for the solver part. I cannot run the program as a whole rather running it line by line

image
image897×890 143 KB

Any help?

Am I using it correctly?

Regards,
Saima Safdar

10

Hi,
I do not have channel files. is there a way to use duneuro only using the FEM mesh file.

I want to solve forward problem using a single dipole.

Regards,
Saima Safdar

11

Code I am using is:
import duneuropy as dp
import simbiopy as sp
import numpy as np

read electrodes

electrodes = dp.read_field_vectors_3d('/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131/projCentroids.txt')

read dipoles

dipoles = dp.read_dipoles_3d('/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131/dipole.txt')

construct driver object

config = {
'type' : 'fitted',
'solver_type' : 'cg',
'element_type' : 'tetrahedron',
'volume_conductor' : {
'grid.filename' : '/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131/brain2mesh_mri.msh',
'tensors.filename' : '/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131/realistic_6c.cond'
}
}
driver = dp.MEEGDriver3d(config)
driver.setElectrodes(electrodes, {'type' : 'closest_subentity_center', 'codims' : [3] })

solution = driver.makeDomainFunction()
for dip in dipoles:
driver.solveEEGForward(dip, solution, {
'source_model.type' : 'partial_integration',
'solver.reduction' : 1e-8,
'post_process' : True,
'subtract_mean' : False})
lf = driver.evaluateAtElectrodes(solution)

lf -= np.mean(lf)

import os
folder_output = '/media/sf_Ubuntu-sahre/S2_example_data_and_scripts/example_data/case131'

driver.write(solution,{
'format' : 'vtk',
'filename' :os.path.join(folder_output, 'case131_result_ubuntu'),
})

CPU usage for the solver part. I cannot run the program as a whole rather running it line by line

Hey,
This is not part of the Brainstorm, in Brainstorm we use only the compiled version with predefined flags.
In you case, you are using the the duneuro source code, where you need to handle all the compilation steps. In that case you need to change some flags in the compilation process to get some extra features (and libraries).
I may recommend you to post this issue to the DUNEuro GitLab Issues · duneuro · GitLab where you can get more help.

You can also refer to the supplementary material included within the DUNEuro paper: DUNEuro—A software toolbox for forward modeling in bioelectromagnetism

Thanks

1 Like
12

DUNEuro computes the FEM leadfield, you can't compute it without the sensors.

I think what you want to do is to compute the voltage on all the head volume and not in a specific sensors/channels?
In that case you can use all the FEM nodes as sensors, then you can get a value on each node. So you need to put the 3D locations of the nodes as sensors.

Expect that the computation time will be very long.
(do not compute the transfer matrix in that case, you need to review and check the associated script and set it to false)

You can do it by using only a single dipole with a specific orientation.

1 Like