pyepilepsy.metrics.compute_sensitivity_chance_pred

pyepilepsy.metrics.compute_sensitivity_chance_pred(preictal_duration: float, tau: float, rho: float) → float

Compute the sensitivity of a random poisson based chance predictor given the time in warning ratio of the predictor.

Parameters:

preictal_durationfloat

The preictal duration in minutes

taufloat

The firing duration in minutes

rhofloat

The time in warning ratio of the predictor to compare with chance

Returns:

float

Sensitivity of a random poisson based chance predictor

Notes

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

Note that I am not sure about the validity of the formula without preictal_duration=tau, even if it looks logical.

Examples

Let’s say we have built a classifier with a postprocessing stage that gives tiw=0.2 for a preictal duration of 60min. To compare with alarm systems, we use tau=preictal_duration, so that any preictal detection leads to a warning during the seizure. We then have:

>>> from pyepilepsy.metrics import compute_sensitivity_chance_pred
>>> compute_sensitivity_chance_pred(preictal_duration=60, tau=60, rho=0.2)
0.19999999999999996

We find, as in the paper, that \(Sen_{chance-predictor} \approx rho\)