credibility

credible_interval(positive, negative, credibility=0.5, prior=(1, 1))

What is the shortest interval that contains probability(positive) with `credibility`% probability?

Parameters:

• positive (int) – number of times the first possible outcome has been seen

• negative (int) – number of times the second possible outcome has been seen

• credibility (float) – The probability that the true p(positive) is contained within the reported interval

• prior (tuple) – psueodcount for positives and negatives

Returns:

(lower bound, upper bound)

prob_below(positive, negative, cutoff, prior=(1, 1))

What is the probability P(positive) is unacceptably low?

Parameters:

• positive (int) – number of times the positive outcome has been seen

• negative (int) – number of times the negative outcome has been seen

• cutoff (float) – lowest acceptable value of P(positive)

• prior (tuple) – psueodcount for positives and negatives

Returns:

Probability that P(positive) < cutoff

prob_greater_cmp(positive1, negative1, positive2, negative2, prior1=(1, 1), prior2=(1, 1), err=1e-05)

Probability the first set comes from a distribution with a greater proportion of positive than the other.

Parameters:

• positive1 (int) – number of positive instances in the first dataset

• negative1 (int) – number of negative instances in the first dataset

• positive1 – number of positive instances in the second dataset

• negative1 – number of negative instances in the second dataset

• prior1 (tuple) – psueodcount for positives and negatives

• prior2 (tuple) – psueodcount for positives and negatives

• err (float) – upper bound of frequentist sample std from monte carlo simulation.

roc_auc_preprocess(positives, negatives, roc_auc)

ROC AUC analysis must be preprocessed using the number of positive and negative instances in the entire dataset and the AUC itself.

Parameters:

• positives (int) – number of positive instances in the dataset

• negatives (int) – number of negative instances in the dataset

• roc_auc (float) – ROC AUC

Returns:

(positive, negative) tuple that can be used for prob_below and

credible_interval