Skip to main content

Membership functions

For the math behind each shape, see Membership functions in Concepts.

ff.mf factories​

The usual way to create a membership function. Each returns a Type1MF (or an IT2MF for it2) that can be assigned to a variable term.

FactoryParametersConstraint (else ValueError)Analytic center()
ff.mf.triangle(a, b, c)feet a, c; peak ba ≤ b ≤ cb
ff.mf.trapezoid(a, b, c, d)feet a, d; plateau b..ca ≤ b ≤ c ≤ d(b + c) / 2
ff.mf.gaussian(center, sigma)peak center, spread sigmasigma > 0center
ff.mf.bell(center, width, slope)1/(1+∣(x−c)/w∣2s)1 / (1 + \lvert (x-c)/w \rvert^{2s})width > 0, slope > 0center
ff.mf.sigmoid(center, slope)1/(1+e−s(x−c))1 / (1 + e^{-s(x-c)}); slope > 0 rises, < 0 fallsnonecenter
ff.mf.singleton(x)1 exactly at x, 0 elsewherenonex
ff.mf.it2(upper, lower, lower_height=1.0)interval type‑2 set, see belowlower ≤ upper everywhere; 0 < lower_height ≤ 1midpoint of the two centers

Invalid parameters raise ff.FuzzyError (a ValueError).

Shoulders: triangle(a, a, c) is a left shoulder (1 at a, falling to 0 at c) and triangle(a, c, c) a right shoulder. Same for trapezoid(a, a, c, d) and trapezoid(a, b, d, d).

import ferrifuzz as ff

hot = ff.mf.triangle(20, 40, 40) # right shoulder
print(hot.membership(30.0)) # 0.5
print(hot.shape(), hot.center()) # Triangular 40.0

ff.mf.gaussian(50, 10).membership(60.0) # 0.6065... (one sigma away)
ff.mf.bell(50, 15, 2).membership(65.0) # 0.5 (|x - c| = width)
ff.mf.sigmoid(50, 0.3).membership(50.0) # 0.5

Type1MF​

Returned by every ff.mf factory except it2.

MethodReturnsDescription
membership(x) / mf(x)float or arraydegree at x; pass a list/array to evaluate many points at once
center()floatanalytic center (table above); the z used by Sugeno inference
shape()str"Triangular", "Trapezoidal", "Gaussian", "Singleton", "Bell" or "Sigmoid"
key_points()list[float]the breakpoints/centers that define the shape, e.g. [0.0, 5.0, 10.0] for triangle(0, 5, 10)
repr(mf)stre.g. triangle(0, 5, 10)
mf == otherboolsame shape and parameters

Vectorized evaluation​

membership also accepts a list or 1‑D NumPy array and returns a NumPy array of the same length (a list if NumPy isn't installed). This is much faster than a Python loop and is the easiest way to plot a term:

import numpy as np

xs = np.linspace(0, 40, 9)
print(hot.membership(xs))
# [0. 0. 0. 0. 0. 0.25 0.5 0.75 1. ]

IT2MF​

An interval type‑2 set bounded by an upper MF (the outer edge of the footprint of uncertainty) and a lower MF (the inner edge). Its membership is an interval (lower, upper).

upper = ff.mf.triangle(0, 5, 10)
lower = ff.mf.triangle(2, 5, 8)
blurry = ff.mf.it2(upper, lower)

print(blurry.membership(3.0)) # (0.3333..., 0.6) -> (lower, upper)
print(blurry.membership(5.0)) # (1.0, 1.0)
print(blurry.shapes()) # ('Triangular', 'Triangular') -> (upper, lower)
print(blurry.upper, blurry.lower)
MemberDescription
membership(x) / it2mf(x)(lower_mu, upper_mu) at x; for an array x, a pair of arrays (lower, upper)
upper, lowerthe two Type1MFs (properties)
lower_heightscale applied to the lower MF (property, default 1.0)
center()midpoint of the upper and lower centers
shapes()(upper_shape, lower_shape)

ff.IT2MF(upper, lower, lower_height=1.0) is the same as ff.mf.it2(...).

Argument order is it2(upper, lower). Ferrifuzz checks that the lower MF never rises above the upper one and raises ff.FuzzyError ("are they swapped?") otherwise.

lower_height scales the lower MF down, which gives the textbook IT2 triangle whose inner triangle is shorter than the outer one:

blurry = ff.mf.it2(ff.mf.triangle(0, 5, 10), ff.mf.triangle(2, 5, 8), lower_height=0.6)
print(blurry.membership(5.0)) # (0.6, 1.0)

Mixed shapes are fine, e.g. a Gaussian upper with a narrower Gaussian lower (uncertain standard deviation), the most common IT2 design:

uncertain_sigma = ff.mf.it2(ff.mf.gaussian(0, 0.3), ff.mf.gaussian(0, 0.15))

See Interval type‑2 for how these sets are used in inference.

Stand‑alone shape classes​

ff.TriangularMF(a, b, c), ff.TrapezoidalMF(a, b, c, d), ff.GaussianMF(center, sigma), ff.BellMF(center, width, slope), ff.SigmoidMF(center, slope) and ff.SingletonMF(x) construct a single shape directly, with the same parameters and checks as the factories. They only have membership(x). They are accepted anywhere a Type1MF is (for example as the arguments of ff.mf.it2), but the factories are the recommended way to build terms.

ff.BellMF(center=50.0, width=15.0, slope=2.0).membership(50.0)   # 1.0