ntqr.r2.evaluations

Evaluations for binary tests (R=2).

For any finite test there is a finite set of evaluations possible. The classes in this module compute them.

Classes:

SingleClassifierEvaluations: Class related to the evaluations for a single classifier consistent with its observed test responses.

Functions:

Misc variables:

Classes

SingleClassifierEvaluations	Single classifier evaluations in (Q_a, Q_b, R_{b_i, a}, R_{a_i,b}) space.
PosteriorSingleEvaluations	Evaluations logically consistent with observed responses.

Module Contents

class ntqr.r2.evaluations.SingleClassifierEvaluations(Q, single_axioms)

Bases: ntqr.evaluations.SingleClassifierEvaluations

Single classifier evaluations in (Q_a, Q_b, R_{b_i, a}, R_{a_i,b}) space.

Deprecated class.

number_apriori_evaluations()

Calculate all the possible evaluations for a binary response test.

Return type:

int

errors_at_qs(qs, responses)

Return all evaluations logically consistent with responses.

In binary classification we have Q_a + Q_b = Q. Thus, we really need to specify only two of the three (Q, Q_a, Q_b). Making a choice is arbitrary and breaks the symmetry in the algebra between the two labels. Instead, we specify Q_a, and Q_b and since we have Q from the instance initialization, we do a quality check (logic check) of the equality between the three quantities.

max_correct_at_qs(qs, responses)

Gives highest performing correct for each label.

Meant to save memory for alarm applications.

all_qs()

Return all possible question numbers.

_label_wrongs_(ql)

Return all possible incorrect given Q_l.

class ntqr.r2.evaluations.PosteriorSingleEvaluations(tvc: ntqr.r2.datasketches.VoteCounts)

Evaluations logically consistent with observed responses.

Q

tvc

all_possible_evaluations(classifier: int)

find_k_nearest_at_prevalence(classifier: int, qa: int, point, k: int)

find_k_nearest_at_prevalence_all_classifiers(qa: int, points, k: int)

distances_to_target(Qa, Raa, Rbb, point)

to_pspace(point)