Experiments from the paper: Multivariate Convolutional Sparse Coding for Electromagnetic Brain Signals

[dinalzein]

I'm trying to replicate some experiments from the paper: "Multivariate Convolutional Sparse Coding for Electromagnetic Brain Signals " (mainly figure 3) but I'm not sure how is GreedyCDL supported for univariate signals (is't only by forcing rank1=False and uv_constraint='auto' )?
Also, for simulating the data, are you using load_data in simulate.py or is there another function that simulates data for multiple channels?

[agramfort]

with only 1 channel there is no Rank 1 constraint required. See our examples with univariate simulations eg https://alphacsc.github.io/auto_examples/noise_alpha/plot_simulate_alphacsc.html#sphx-glr-auto-examples-noise-alpha-plot-simulate-alphacsc-py now Fig 3 in the paper your mention varies P ie the number of channels so it's not univariate in this figure.

…

Message ID: ***@***.***>

[dinalzein]

Yes, but I wanted to compare the vanilla CSC with GreedyCSC algorithm on univariate signal. I tried this by having rank1=False, uv_constraint='auto' , and solver_d = 'fista'; it worked.

In the paper, there is no comparison with and without the rank1 constraint on the multivariate signal on the mne dataset. I want to do this experiment but I'm getting the following error when having rank1=False, uv_constraint='auto', andsolver_d = 'fista' on the mne dataset.

The error seems to come from line 39 @numba.jit((numba.float64[:, :, :],), nopython=True, cache=True) in compute_constants.py

[agramfort]

please share a self-contained code snippet to replicate the problem

…

Message ID: ***@***.***>