kuffmode/msa

Hopefully, a compact and general-purpose Python package for Multiperturbation Shapley value Analysis (MSA).

I study computational neuroscience and compose music. That's it for now!

msa's Issues

Documentation Refactoring

Hi @luisa-hoerauf

The documentation is old and the current developers are "overfitted" to it. New eyes are appreciated.

Tasks:

• Go over all the documentation and suggest changes.

All communication can happen on this issue by commenting on your progress and doubts. Once you're done, you can open a pull request which will close this issue.

Regards,
Shrey Dixit

Causal influences and Pandas 1.5

Causal influence matrix gets shifted in Pandas 1.5 but all tests pass. We have to dig into this and see if it's the case for Pandas 2.

Update requirements

So some of the packages we are using can be updated. I realized seaborn v 0.11 has a bug that is fixed in 0.13 but we're still using 0.11

ValueError: Objective function returned values of different data types

def lesion_esn_behavior(complements, network, training_data, ground_truth):
    """
    This function lesions nodes and tracks their contributions to the output signal. This is for the classic MSA part.
    """
    if len(complements) == len(network.W):
        return 0 # every node in the network is lesioned so toss a zero to your MSA oh valley of plenty.

    lesioned_network = deepcopy(network)
    for target in complements: # same as above
        lesioned_network.W[target] = 0.00001
        lesioned_network.W_fb[target] = 0.00001
        lesioned_network.W_in[target] = 0.00001

    lesioned_network.fit(X=None, y=training_data)
    y_pred = lesioned_network.predict() # contrary to the previous function, here we want the output signal.

    return float(np.linalg.norm(y_pred,2))

I want to use float number to be ovjective function, but it is not work. If i using int number, it is work. I don't know how to solve it.