Skip to content
Tutorials Center
My learning

Topics

Lesson 7 of 36 ยท Fuzzy Sets

Properties of Fuzzy Sets

Learn the standard properties of a fuzzy set: support, core, height, normality, crossover points, cardinality and convexity, and compute each of them with Python.

To compare fuzzy sets, describe them, or check that your design is sensible, you need a small vocabulary of properties. Mathematicians use the same words in textbooks and in software documentation, so learning them pays off every time you read about a fuzzy toolbox. This lesson defines each one, shows it on a picture, and computes it with a few lines of NumPy.

We assume a universe sampled on a grid, which is how fuzzy sets are handled on a computer.

The key regions

For a fuzzy set A with membership function mu on a universe X:

  • Support: all x with mu(x) > 0. It is "where the set is at all true".
  • Core: all x with mu(x) = 1. It is "where the set is completely true".
  • Boundary (or "fringe"): points with 0 < mu(x) < 1, the partly-true zone between the core and the outside.
  • Crossover points: points with mu(x) = 0.5, where the concept is exactly "half true".
  • Height: the largest membership value, max mu(x).
  • Normal / subnormal: a set is normal if its height is 1 (it has a non-empty core) and subnormal otherwise.
flowchart LR
    S["Support<br/>mu > 0"] --> B["Boundary<br/>0 < mu < 1"]
    S --> C["Core<br/>mu = 1"]
    B --> X["Crossover<br/>mu = 0.5"]

The core sits inside the support; everything else in the support is the boundary.

Python Support, core, height and crossover

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The support runs from 2 to 9, the core from 4 to 6, and there are two crossover points: one on the rising side at 3 and one on the falling side at 7.5. A trapezoid always has exactly two crossover points; a triangle also has two.

Normal versus subnormal

Most fuzzy sets used in systems are normal: at least one value is "perfectly" a member. A subnormal set never reaches 1. This happens, for instance, after clipping a set at a rule's firing strength (a step in Mamdani Inference). A subnormal set can be normalised by dividing by its height, which stretches it so that the peak becomes 1.

Python Normalising a subnormal set

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Cardinality: how big is a fuzzy set?

A crisp set has a size: the number of elements. For a fuzzy set on a discrete universe the scalar cardinality (also called sigma-count) adds up the membership degrees:

TEXT
|A| = sum of mu(x) over all x

A person who is "tall to degree 0.5" counts as half a tall person. The relative cardinality divides by the number of elements in the universe and gives the "fraction" of the universe that the set covers.

Python Counting fuzzy members

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The sigma count (3.5) is a smoother answer than the crisp count (3): it does not make a person suddenly appear or vanish when a height crosses the threshold.

Convex fuzzy sets

A fuzzy set is convex if, for any two points, every point between them has a membership at least as high as the smaller of the two. Informally, the graph has no dips: it goes up, maybe stays flat, and then goes down. Triangles, trapezoids and bells are convex. A set with two separate peaks is not.

Convexity is a sensible property for meanings such as "warm" or "medium", which should describe one contiguous range of values. A set like "extreme temperature" (very cold or very hot) is legitimately non-convex, but then it is better to build it as a combination of two convex sets (cold OR hot).

Python Testing convexity

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The function above finds the highest point and checks that the values never fall before it and never rise after it. That is equivalent to convexity for sets with a single plateau.

Other useful words

  • Fuzzy number: a normal, convex fuzzy set on the real line whose core is a single point, such as "about 5". See Alpha-Cuts and the Extension Principle.
  • Singleton: a fuzzy set with support of a single point.
  • Symmetric set: a set whose graph is a mirror image around its peak.
  • Fuzzy partition: a family of sets whose memberships add up to 1 at every point.

Check your understanding

Quick check ยท 5 questions

of 5 answered You scored / 5 ยท perfect!

  1. 1. What is the core of a fuzzy set?

    The core is the set of points that belong completely: mu(x) = 1. Points with mu > 0 form the support.
  2. 2. A set has maximum membership 0.8. It is called

    A set is normal only when its height is 1. A height of 0.8 means it is subnormal.
  3. 3. What is the scalar cardinality of the fuzzy set 0.5/a + 1/b + 0.25/c?

    Sum the degrees: 0.5 + 1 + 0.25 = 1.75.
  4. 4. Which set is NOT convex?

    A dip between two higher values breaks convexity.
  5. 5. How do you normalise a subnormal set?

    Dividing by the maximum membership scales the peak up to 1 and keeps the shape.

Summary

  • Support is where mu > 0, core is where mu = 1, crossover is where mu = 0.5, and height is the maximum degree.
  • A set is normal if its height is 1; clipping produces subnormal sets, which can be normalised.
  • Scalar (sigma-count) cardinality adds all degrees; relative cardinality divides by the universe size.
  • A convex set has no dips; meanings like "warm" are usually convex.
  • Fuzzy numbers, singletons and partitions are named special cases.

Next: Alpha-Cuts and the Extension Principle