nispace.stats.effectsize.prc

nispace.stats.effectsize.prc(a, b)

Percent change (a - b) / a * 100, element-wise.

Parameters:

• a (array_like, shape (n_obs, n_features)) – Paired values (a is the reference/denominator); must have identical shape.

• b (array_like, shape (n_obs, n_features)) – Paired values (a is the reference/denominator); must have identical shape.

Returns:

p – NaN where a == 0 (avoids division by zero).

Return type:

np.ndarray, shape (n_obs, n_features)

Raises:

ValueError – If a.shape != b.shape.

Notes

NaN in a/b (other than the a == 0 guard) is not explicitly masked – it propagates through the arithmetic like any other float NaN; there is no separate prc_nan variant. Not called anywhere in NiSpace’s internal pipeline: Y_transform=”prc(a,b)” dispatches to the numba-jitted prc_fast (core/transform_y.py), not this function.