Causal influences and Pandas 1.5

[kuffmode]

Causal influences and Pandas 1.5

[ShreyDixit]

Thanks for pointing it out. Can you provide a MCVE so that I can get started.

[kuffmode]

import numpy as np
from msapy import msa
import matplotlib.pyplot as plt
import networkx as nx
from copy import deepcopy
import seaborn as sns

We need to simulate a system so these functions:

def event_maker(
    n_units: int,
    timesteps: int,
    probability: float = 1,
    rng: np.random.Generator = np.random.default_rng(seed=2023),
) -> np.ndarray:

    if probability < 1:
        input = rng.choice(
            [0, 1], p=[1 - probability, probability], size=(n_units, timesteps)
        )
        input = input.astype(float)
    else:
        input = np.zeros((n_units, timesteps))
        for node in range(n_units):
            event_timepoints = rng.integers(0, timesteps, 1)
            input[node, event_timepoints] += 1

    return input

def identity(x):
    return x

def simulate_dynamical_system_parallel(adjacency_matrix:np.ndarray,
                       input_matrix:np.ndarray, 
                       coupling:float=1, 
                       dt:float=0.001, 
                       duration:int=10, 
                       timeconstant:float=0.01,
                       function = identity, 
                       )->np.ndarray:
    
    N = input_matrix.shape[0]
    T = np.arange(1, duration/dt + 1) # holds time steps
    X = np.zeros((N, len(T)+1)) # holds variable x

    dt = dt/timeconstant
    for timepoint in range(len(T)):
        for node in range(N):
            if timepoint == 0:
                X[node, timepoint] = input_matrix[node, timepoint]
            else:
                inputs = np.dot(coupling * adjacency_matrix[node, :], X[:, timepoint-1]) + input_matrix[node, timepoint]
                X[node, timepoint] = (1 - dt) * X[node, timepoint-1] + dt * function(inputs)

    return X

def lesion_simple_nodes(complements,
                        adjacency_matrix: np.ndarray,
                        index: int,
                        input: np.ndarray,
                        model = simulate_dynamical_system_parallel,
                        model_kwargs:dict={},
                        ) -> np.ndarray:

    lesioned_connectivity = deepcopy(adjacency_matrix)
    for target in complements:
        lesioned_connectivity[:, target] = 0.0
        lesioned_connectivity[target, :] = 0.0

    dynamics = model(adjacency_matrix = lesioned_connectivity,
                        input_matrix=input,
                        **model_kwargs)
    lesioned_signal = dynamics[index]
    return lesioned_signal

And then these parameters:

rng = np.random.default_rng(seed=2023)
chain = nx.to_numpy_array(nx.generators.path_graph(6,nx.DiGraph())).T
chain *= 0.9
timesteps = 100
example_input = event_maker(6,timesteps)

NOISE_STRENGTH = 0.05
DELTA = 0.01
TAU = 0.02
G = 0.74
DURATION = 1
model_params = {'dt':DELTA,'timeconstant':TAU,'coupling':G,'duration':DURATION}

After running the game you see the zeroth column went after 5 in pandas 1.5 and 2 but not 1.3.5.

lesion_params = {'adjacency_matrix': chain, 'input': example_input, 'model_kwargs': model_params}
ci_chain = msa.estimate_causal_influences(elements=list(range(len(chain))),
                                        objective_function=lesion_simple_nodes,
                                        objective_function_params=lesion_params,
                                        n_permutations=128, n_cores=1)