statkit 1.0.0

Confidence intervals and p-values for sci-kit learn.

Statkit

Quickstart | Reference docs

Supplement your sci-kit learn models with 95 % confidence intervals, p-values, and decision curves.

Contents

• Quickstart
• Installation

Quickstart

• Estimate 95 % confidence intervals for your test scores.

For example, to compute a 95 % confidence interval of the area under the receiver operating characteristic curve (ROC AUC):

from sklearn.metrics import roc_auc_score
from statkit.non_parametric import bootstrap_score

y_prob = model.predict_proba(X_test)[:, 1]
auc_95ci = bootstrap_score(y_test, y_prob, metric=roc_auc_score)
print('Area under the ROC curve:', auc_95ci)

• Compute p-value to test if one model is significantly better than another.

For example, to test if the area under the receiver operating characteristic curve (ROC AUC) of model 1 is significantly larger than model 2:

from sklearn.metrics import roc_auc_score
from statkit.non_parametric import paired_permutation_test

y_pred_1 = model_1.predict_proba(X_test)[:, 1]
y_pred_2 = model_2.predict_proba(X_test)[:, 1]
p_value = paired_permutation_test(y_test, y_pred_1, y_pred_2, metric=roc_auc_score)

• Perform decision curve analysis by making net benefit plots of your sci-kit learn models. Compare the utility of different models and with decision policies to always or never take an action/intervention.

from matplotlib import pyplot as plt
from sklearn.datasets import make_blobs
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.linear_model import LogisticRegression
from statkit.decision import NetBenefitDisplay

centers = [[0, 0], [1, 1]]
X_train, y_train = make_blobs(
    centers=centers, cluster_std=1, n_samples=20, random_state=5
)
X_test, y_test = make_blobs(
    centers=centers, cluster_std=1, n_samples=20, random_state=1005
)

baseline_model = LogisticRegression(random_state=5).fit(X_train, y_train)
y_pred_base = baseline_model.predict_proba(X_test)[:, 1]

tree_model = GradientBoostingClassifier(random_state=5).fit(X_train, y_train)
y_pred_tree = tree_model.predict_proba(X_test)[:, 1]

NetBenefitDisplay.from_predictions(y_test, y_pred_base, name='Baseline model')
NetBenefitDisplay.from_predictions(y_test, y_pred_tree, name='Gradient boosted trees', show_references=False, ax=plt.gca())

Detailed documentation can be on the Statkit API documentation pages.

Installation

pip3 install statkit

Support

You can open a ticket in the Issue tracker.

Contributing

We are open for contributions. If you open a pull request, make sure that your code is:

• Well documented,
• Code formatted with black,
• And contains an accompanying unit test.

Authors and acknowledgment

Hylke C. Donker

License

This code is licensed under the MIT license.