pyepilepsy.metrics.compute_ioc

pyepilepsy.metrics.compute_ioc(y_true: Series, y_pred: Series, threshold: float, epoch_size: float, overlap: float, preictal_duration: float, groups: Series, label_to_num: dict)

Compute the Improvement Over Chance.

The chance is a random process poisson based classifier that uses the same amount of time in warning. See notes.

Parameters:

y_truepd.Series

The true values.

y_predpd.Series

The predicted values, output of the classifier/postprocessing stage.

thresholdfloat

The threshold to use to consider a prediction as 0 or 1.

epoch_sizefloat

The epoch size in seconds.

overlapfloat

The overlap in seconds.

preictal_durationfloat

The preictal duration in minutes.

groupspd.Series

A pandas Series that has the same length as y_true and y_pred that contains the group number of each sample.

label_to_numdict

A dictionary containing the string labels as keys and corresponding integer values.

Returns:

float

Improvement Over Chance

Notes

See The Statistics of a Practical Seizure Warning System for more detail. (doi:10.1088/1741-2560/5/4/004)

Examples

>>> import pandas as pd
>>> import numpy as np
>>> from pyepilepsy.metrics import compute_ioc
>>> from datetime import datetime

>>> y_true_arr = [0, 0, 0, 0, 0, 1, 1, 7, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 7]
>>> y_pred_arr = [0.1, 0.12, 0.13, 0.56, 0.7, 0.8, 0.6, 0.9,  0.2, 0.1, 0.8, 0.1, 0.2, 0.1, 0.2, 0.7, 0.6, 0.7, 0.9]

>>> datetime_index = np.hstack([
...     pd.date_range("09:00", "12:30", freq="30min", tz="Europe/Paris").to_numpy(),
...     pd.date_range("14:00", "19:00", freq="30min", tz="Europe/Paris").to_numpy()
... ])
>>> y_true = pd.Series(
...     data=y_true_arr,
...     index=datetime_index
... )
>>> y_pred = pd.Series(
...     data=y_pred_arr,
...     index=datetime_index
... )
>>> groups = pd.Series(
...     data=[0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1],
...     index=datetime_index
...     )
>>> compute_ioc(
...     y_true=y_true,
...     y_pred=y_pred,
...     threshold=0.58,
...     epoch_size=60*30,
...     overlap=60*10,
...     preictal_duration=60,
...     groups=groups,
...     label_to_num= {'ictal': 7, 'preictal': 1, 'interictal': 0}
...     )
0.4736842105263158