non linear running time when simulating raw signal

šŸ’¬ Support & Discussions
1

Dear mne community,

  • MNE version: e.g. 1.8.0
  • operating system: Kubuntu 24.04

Context:
Iā€™m running simulations of raw signal. After loading the head model, determining the source (that is the cortex patch that will be used as the source), and defining the souce time course, I add the source time course to the source SourceSimulator:


source_simulator = mne.simulation.SourceSimulator(src, tstep=tstep)
for t in np.arange(n_trials):
    source_simulator.add_data(vis_r, act_1[:,t], events)

Everything works fine!
I then use the SourceSimulator to simulate the raw signal


raw_vis = mne.simulation.simulate_raw(raw.info, source_simulator, forward=fwd)

No problem here neither.

The issue I have concerns the running time of this last line (simulate_raw) as a function of the number of trials I want to simulate (i.e. the number of event in events, which corresponds to the number of trials n_trials). I was expecting this running time to be linear, but it is far from being linear!

Here are some running time as a function of the number of events:

n_trials running time (s)
10 0.5
20 2.8
30 9
40 21
50 42

With such an exponential increase, it is unrealistic to run a ā€œlargeā€ number of trials (I was targeting ā€œonlyā€ 300ā€¦).

I quickly inspected the code of simulate_raw, but I could not identify what could cause the exponential increase.

Not sure this is relevant, but Iā€™m running the code in a jupyter notebookā€¦

Any idea of what could be the cause of the phenomenon ?

Thanks very much in advance,
B.

2

Dear community,

I guess I found out where the problem stems fromā€¦
It is related to misunderstanding of the SourceSimulator object, and its add_data method.
To add ā€˜n_time * n_trialsā€™ data, one should not do

for t in np.arange(n_trial):
   source_simulator.add_data(vis_r, act_1[:,t], events)

since, as far as I understand, we have ā€˜n_trialsā€™ events in ā€˜eventsā€™, the operation will be repeated ā€˜n_trials * n_trialsā€™ time; hence the exponential time.

One should instead simply run:

source_simulator.add_data(vis_r, act_1, events)

which will directly import the n_trials into source_simulator, which saves a huge amount of time.

Funny enough, the error of importing in a loop was still producing the correct number of trials in the raw object thoughā€¦

Hope this might be helpfull for otherā€¦
Best