PSILOGIT.PSILOGIT

class PSILOGIT(regularization=1, truetheta=None, X=None, n=None, p=None, yobs=None, M=None, SM=None, sampling_algorithm=None, seed=1)

Class allowing to conduct post-selection inference procedures in the logistic model with l1 penalty.

__init__(regularization=1, truetheta=None, X=None, n=None, p=None, yobs=None, M=None, SM=None, sampling_algorithm=None, seed=1)

Methods

SEI_SLR([temperature, delta, ...])	SEI-SLR (Selection Event Identification for the Sparse Logistic Regression) algorithm.
SEI_by_sampling(sig[, nb_ite])	Computes states belonging to the selection event.
__init__([regularization, truetheta, X, n, ...])
change_sampling_algorithm(sampling_algorithm)
compute_power(lists_pvalues, names[, alpha, ...])
compute_selection_event([...])	Finds all the vectors belonging to the selection event.
compute_theta_bar(barpi[, grad_descent])	Computes \(\overline \theta(\theta^*) \in \mathbb R^s\) which is the unique vector satisfying \(\mathbf X_M^{\top}\sigma(\mathbf X_M \overline \theta (\theta^*))=\mathbf X_M^{\top} \overline \pi^{\theta^*}\) where \(\overline \pi^{\theta^*}\) is the input parameter 'barpi'.
ellipse_testing(states, barpi[, signull, ...])	For a selected support of size 2, this method show the proportion of stats following in the ellipse characterizing the rejection rejection of the hypothesis test with SIGLE in the selected model.
histo_time_in_selection_event(states, ...[, ...])	Histogram showing the time spent in the selection event using the SEI-SLR algorithm.
last_visited_states(states[, ...])	Shows that time spent in the selection event using the SEI-SLR algorithm.
low_up_taylor(theta_obs, SM, gamma)	Computes the truncation bounds for the approximate gaussian distribution of the statistic used in the post-selection inference method from Taylor & Tibshirani '18.
params_saturated(bern, states)	Computes the probability of the vector of bits 'z' when the expected value of the response vector is given by 'bern'
plot_cdf_pvalues(lists_pvalues, names[, ...])	Shows the cumulative distribution function of the p-values.
pval_SIGLE(states, barpi[, net, ...])	Computes the P-values using the post-selection inference method SIGLE (both in the saturated and the selected model).
pval_taylor(states[, mode, gamma, ...])	Computes the P-values using the post-selection inference method from Taylor & Tibshirani '18.
pval_weak_learner(statesnull, statesalt, barpi)	Computes the P-values obtained from the weak-learner which is a two-sided test based on the statistic \(\sum_{i=1}^n |\overline \pi^{\pi^0}_i-y_i|\) where \(\overline \pi^{\pi^0}\) is the expectation of the vector of observations under the null conditional to the selection event.
time_in_selection_event(states, ...[, fig_name])	Shows that time spent in the selection event using the SEI-SLR algorithm for several different excursions.
upper_bound_condition_CCLT(states, barpi, ...)	Computes quantities arising in the assumption of our conditional Central Limit Theorem.