roc-utils

This Python package provides tools to compute and visualize ROC curves, which are used to graphically assess the diagnostic ability of binary classifiers.

Use roc_utils to perform ROC analyses, including the calculation of the ROC-AUC (the area under the ROC curve) and the identification of optimal classification thresholds for different objective functions. In addition, it is possible to compute mean, tolerance interval (TI) and confidence interval (CI) for a set of (related) ROC curves. Finally, error bounds can be estimated and visualized by means of boostrap sampling.

Installation:

pip install roc-utils

Use the following commands for a quick verification of the installation.

python -c "import roc_utils; print(roc_utils.__version__)"
python -c "import roc_utils; roc_utils.demo_bootstrap()"

Usage:

See examples/tutorial.ipynb for step-by-step introduction.

import numpy as np
import matplotlib.pyplot as plt
import roc_utils as ru

# Construct a binary classification problem
x, y = ru.demo_sample_data(n1=300, mu1=0.0, std1=0.5,
                           n2=300, mu2=1.0, std2=0.7)

# Compute the ROC curve...
pos_label = True
roc = ru.compute_roc(X=x, y=y, pos_label=pos_label)

# ...and visualize it
ru.plot_roc(roc, label="Sample data", color="red")
plt.show()

# To perform a ROC analysis using bootstrapping
n_samples = 20
ru.plot_roc_bootstrap(X=x, y=y, pos_label=pos_label,
                      n_bootstrap=n_samples,
                      title="Bootstrap demo");
plt.show()

Build from source:

To fetch the project and run the tests or examples:

git clone https://github.com/hirsch-lab/roc-utils.git
cd roc-utils
python tests/test_all.py
python examples/examples.py

To create distribution packages (a source archive and a wheel):

python setup.py sdist bdist_wheel

To install the newly created Python package from the source archive:

pip uninstall roc-utils
pip cache remove roc_utils
pip install dist/roc_utils*.tar.gz

# Verify installation
python -c "import roc_utils; print(roc_utils.__version__)"
python -c "import roc_utils; roc_utils.demo_bootstrap()"