brainets / frites Goto Github PK

@brovelli as you pointed out, the current FIT input parameter for computing directed FIT is a boolean value called directed. As you said, it's not really explicit since it's always directed and it would be better to have a "uni/bidirectional" input parameter.

We should discuss about it. Usually I prefer using short input parameters therefore I would exclude the use of the complete words "unidirectional" or "bidirectional".

I propose the two following solutions instead of having directed=True/False :
1 - direction="uni"/"bi"
2 - unidirectional=True/False
3 - net=True/False

My favorite order is 3 / 2 / 1. Do you have other propositions?

When exporting conjunction analysis of get_params, the first one contains a subject dimension while the second is subjects. This can be problematic when saving datasets. Need to be fixed.

Hi, after that I used frites a lot, I ran a sanity check on me and I found out that my cholesterol level is too high.
I suggest to change the oil more often.
Thanks in advance.

@nvoges @brovelli

The current version of the code failed on Nicoles' simulated data, because of the cholesky decomposition when the data are redundant.

@brovelli suggested np.clip(mi, 0, mi). This avoid negative MI values (I think it should also works on negative infinite). @nvoges do you get infinite values or NaN values? Because if NaN, the clip is not going to make a difference. In addition, I've seen negative MI because of the bias correction so I'm not sure it's a good idea to put a hard threshold at 0, especially because we also compute permutations and I'm afraid of having many permutations to zero because of that.

What do u think?

Hi,

I spoke with the MNE core dev team earlier and we saw that you built a package on top of MNE-Python. Exciting!

We recently moved all the connectivity submodule into a new package called mne-connectivity (https://github.com/mne-tools/mne-connectivity). We've made some abstractions for data structures concerning connectivity and also have a CI/unit-testing pipeline in place with MNE. In favor of trying to pool resources together and have a 1-stop shop for MNE connectivity type analyses, would you be interested in joining forces with mne-connectivity?

Lmk if I missed anything.

PS. I did a year at Aix-Marseille actually 2017-18 under a fellowship during my PhD :p

cc: @drammock

Hi,

Would it be desirable for a PR to add automated templates for GH issues and PRs? Similar to what we have here https://github.com/mne-tools/mne-connectivity/tree/main/.github?

Github action for :

1. Test different versions of Python
2. Build the doc
3. Flake
4. PR

Follow up the Brainhack Marseille :

• Integration of the contributors
• Integration of the core functions
• Integration in connectivity functions (agnostic)
• Integration into the documentation

Here is the code:

import numpy as np

from mne import EpochsArray, create_info
from frites.dataset import DatasetEphy

import matplotlib.pyplot as plt
n_subjects = 5
n_epochs = 10
n_channels = 5
n_times = 100
sf = 512

x, ch = [], []
for k in range(n_subjects):
    # generate single subject data
    x_suj = np.random.rand(n_epochs, n_channels, n_times)
    # generate some random channel names
    ch_suj = np.array([f"ch_{r}" for r in range(n_channels)])
    # concatenate in a list
    x.append(x_suj)
    ch.append(ch_suj)
# finally lets create a time vector
times = np.arange(n_times) / sf

x_mne = []
for k in range(n_subjects):
    # create some informations
    info = create_info(ch[k].tolist(), sf)
    # create the Epoch of this subject
    epoch = EpochsArray(x[k], info, tmin=times[0], verbose=False)
    # finally, replace it in the original list
    x_mne.append(epoch)

dt = DatasetEphy(x_mne)

mi_type = 'cc'

# define the workflow
wf = WfMi(mi_type=mi_type)
# compute the mutual information
mi, _ = wf.fit(dt, mcp=None, n_jobs=1)

Using the development version of Frites

When printing a DatasetEphy, a xarray.Dataset is created. However, two issues emerge with this definition :

1. It creates a duplicates of the data
2. To be able to align the data across subjects, xarray fill the missing cases with NaN. This can potentially lead to misleading interpretations

Need to be fixed (see here)

Hello @EtienneCmb ,
when 4d arrays are used in DatasetEphy as x (analysis on multiples frequency bands), the reshape in 3d form does not take in consideration the single subject's number of epochs.
Here the code I'm referring at:
in frites/dataset/ds_ephy.py in line 142

n_e, n_r, n_f, n_t = x[0].shape
for k in range(len(x)):
    x[k] = x[k].reshape(n_e, n_r, n_f * n_t)

Thank you so much BG! 😘

Salut Etienne,

Petit truc: j'ai vue que la permutation de y pour generer y_p
https://github.com/brainets/frites/blob/master/frites/workflow/wf_mi.py#L127
est repeté dans chaque for loop sur les ROIs.
Je crois que chaque permutation peut etre identique pour les differents ROIs. Donc on pourrait sortir la permutation de y de la boucle dur les ROIs, non ?

https://github.com/brainets/frites/blob/master/frites/workflow/wf_mi.py#L125