Source code for pyfli.cm_stats.chi_square
"""
Pearson and Neyman chi-square statistics and the degrees of freedom of a fit.
"""
from typing import Any
import numpy as np
[docs]
def degrees_of_freedom(n: Any, n_params: Any, minimum: Any = 1) -> Any:
"""``n - n_params``, raised to at least `minimum` (``None`` leaves it unbounded)."""
nu = n - n_params
if minimum is None:
return nu
if np.ndim(nu) == 0:
return max(nu, minimum)
return np.maximum(nu, minimum)
[docs]
def pearson_terms(model: Any, data: Any, floor: float = 1.0) -> Any:
"""Per-gate Pearson terms ``(d - mu)^2 / max(mu, floor)``."""
mu = np.asarray(model, dtype=float)
d = np.asarray(data, dtype=float)
return (d - mu) ** 2 / np.clip(mu, floor, None)
[docs]
def pearson_chi_square(
model: Any, data: Any, axis: int = -1, floor: float = 1.0
) -> Any:
"""
Pearson chi-square ``sum((d - mu)^2 / max(mu, floor))`` along `axis` -- the fit
statistic reported before the switch to the Poisson deviance. The default variance
floor of 1 count makes it read below ``n - p`` when many gates hold less than one
expected count.
"""
return np.sum(pearson_terms(model, data, floor), axis=axis)
[docs]
def neyman_chi_square(model: Any, data: Any, axis: int = -1, floor: float = 1.0) -> Any:
"""Neyman chi-square ``sum((d - mu)^2 / max(d, floor))`` along `axis` (variance from the data)."""
mu = np.asarray(model, dtype=float)
d = np.asarray(data, dtype=float)
return np.sum((d - mu) ** 2 / np.clip(d, floor, None), axis=axis)