tefs package

Submodules

tefs.core module

class tefs.core.TEFS(features: ndarray, target: ndarray, k: int, lag_features: list[int], lag_target: list[int], direction: str, verbose: int = 1, var_names: list[str] = None, n_jobs: int = 1)

Bases: object

fit() → None

Run the feature selection algorithm and return the results.

get_result() → List[Dict[str, Dict[int, float] | float | int]]

Return the result of the feature selection algorithm.

map_output_to_var_names()

A decorator that modifies the output of the decorated method. Assumes the method returns a list of indices and uses these indices to return the corresponding values from self.var_names.

plot_te_results(ax: Axes) → None

Plot the results of the TE estimation for each iteration.

Parameters:

  • scores_iterations (list[IterationResult]) – A list of results of the TE scores, one per iteration.

  • var_names (list[str]) – A list of variable names.

  • ax (matplotlib.axes.Axes) – The axis to plot the results on.

select_features(*args, **kwargs)
select_n_features(*args, **kwargs)
tefs.core.score_features(features: ndarray, target: ndarray, conditioning: ndarray, k: int, lag_features: list[int], lag_target: list[int], direction: str, n_jobs=1) → ndarray

Computes the transfer entropy score for each feature \(X_i\) in \(X\), to the target \(Y\), given the conditioning set \(X_A\).

Parameters:

  • features (np.ndarray of shape (n_samples, n_features)) – Sample of a (multivariate) random variable representing the input

  • target (np.ndarray of shape (n_samples, n_targets)) – Sample of a (multivariate) random variable representing the target

  • conditioning (np.ndarray of shape (n_samples, n_conditioning)) – Sample of a (multivariate) random variable representing the conditioning

  • k (int) – number of nearest neighbors for the CMI estimation

  • lag_features (list[int]) – the lags applied on features and conditioning

  • lag_target (list[int]) – the lags applied on the target

  • direction (str) – the direction of the transfer entropy, either “forward” or “backward”

  • n_jobs (int) – number of parallel jobs to run

Returns:

a dictionary with key = feature index and value = transfer entropy score

tefs.core.score_features_parallel(args)

Helper function to compute the score of a single feature in parallel.

tefs.estimation module

tefs.estimation.estimate_cmi(X: ndarray, Y: ndarray, Z: ndarray, k: int = 5, estimation_method: str = 'digamma') → float

Estimate the Conditional Mutual Information (CMI) between \(X\) and \(Y\) given \(Z\), i.e. \(I(X;Y \mid Z)\), using the equivalance

\[I(X;Y \mid Z) = I(X,Z;Y) - I(Z;Y)\]

Note that \(I(X;Y \mid Z) = I(Y;X \mid Z)\).

Parameters:

  • X (numpy.ndarray) – The input variable X.

  • Y (numpy.ndarray) – The input variable Y.

  • Z (numpy.ndarray) – The input variable Z.

  • k (int) – The number of nearest neighbors for k-nearest neighbor estimation (default is 5).

  • estimation_method (str) – The estimation method to use (default is “digamma”).

Returns:

The estimated CMI between X and Y given Z.

Return type:

float

tefs.estimation.estimate_conditional_transfer_entropy(X: ndarray, Y: ndarray, Z: ndarray, k: int, lag_features: list[int] = [1], lag_target: list[int] = [1], lag_conditioning: list[int] = None) → float

Computes the conditional transfer entropy from X to Y given Z, using the specified lags.

Parameters:

  • X (np.ndarray of shape (n_samples, n_features)) – Sample of a (multivariate) random variable representing the input

  • Y (np.ndarray of shape (n_samples, n_targets)) – Sample of a (multivariate) random variable representing the target

  • Z (np.ndarray of shape (n_samples, n_conditioning)) – Sample of a (multivariate) random variable representing the conditioning

  • lag_features (List[int]) – the lag applied on X

  • lag_target (List[int]) – the lag applied on Y

  • lag_conditioning (List[int]) – the lag applied on Z, if None it is set to lag_features

Returns:

a scalar of the value of the transfer entropy

tefs.estimation.estimate_mi(X: ndarray, Y: ndarray, k: int = 5, estimation_method: str = 'digamma') → float

Estimate the Mutual Information (MI) between \(X\) and \(Y\), i.e. \(I(X;Y)\), based on Mixed Random Variable Mutual Information Estimator - Gao et al..

Parameters:

  • X (numpy.ndarray) – The first input array.

  • Y (numpy.ndarray) – The second input array.

  • k (int, optional) – The number of nearest neighbors to consider, defaults to 5.

  • estimation_method (str, optional) – The estimation method to use, can be either ‘digamma’ or ‘log’, defaults to ‘digamma’.

Returns:

The estimated mutual information.

Return type:

float

tefs.plotting module

tefs.utils module

Module contents