Try to uniformize the function of generating the noise

[lionelkusch]

By merging the 2 functions of the noise, I get a strange result for plot_2D.
@bthirion Do you have an explanation for it?

I made a snippet of code here: https://github.com/lionelkusch/hidimstat/blob/PR_spatial_noise/spatial_noise/plot_2D_simulation_example.py

[lionelkusch]

For the parameters, I use:

• n_samples = 100
• shape = (40, 40)
• n_features = shape[1] * shape[0]
• roi_size = 4 # size of the edge of the four predictive regions
• sigma = 2.0 # noise standard deviation
• rho = 0.5 # level of "spatial correlation"
• snr = 50.0

I get this result:

I don't know if it's interesting or not.
This previous result will replace the following result:

[bthirion]

My impression is that the smoothing is applied only horizontally, which leads to this very anisotropic clusters.

[lionelkusch]

My impression is that the smoothing is applied only horizontally, which leads to this very anisotropic clusters.
Yes, I think it's the case.

Do you think that it can be interesting to improve the function multivariate_simulation by proposing different types of correlation/smoothing? Or Do we stay with 2 different functions and I improve multivariate_simulation_spatial for having a similar signature that multivariate_simulation?

[bthirion]

My impression is that the smoothing is applied only horizontally, which leads to this very anisotropic clusters.
Yes, I think it's the case.

Do you think that it can be interesting to improve the function multivariate_simulation by proposing different types of correlation/smoothing? Or Do we stay with 2 different functions and I improve multivariate_simulation_spatial for having a similar signature that multivariate_simulation?

No there's no need for that. We should only keep isotropic smoothing