psigmf
For a given domain x and parameters params (or [a1 c1 a2 c2]), return the corresponding y values for the product of two sigmoidal membership functions.
The argument x must be a real number or a non-empty vector of strictly increasing real numbers, and a1, c1, a2, and c2 must be real numbers. This membership function satisfies the equation:
f(x) = (1/(1 + exp(-a1*(x - c1)))) * (1/(1 + exp(-a2*(x - c2))))
The function is bounded above by 1 and below by 0.
If a1 is positive, a2 is negative, and c1 and c2 are far enough apart with c1 < c2, then:
(a1)/4 ~ the rising slope at c1 c1 ~ the left inflection point (a2)/4 ~ the falling slope at c2 c2 ~ the right inflection point
and at each inflection point, the value of the function is about 0.5:
f(c1) ~ f(c2) ~ 0.5.
(Here, the symbol ~ means "approximately equal".)
To run the demonstration code, type "demo psigmf" (without the quotation marks) at the Octave prompt.
See also: dsigmf, gauss2mf, gaussmf, gbellmf, pimf, sigmf, smf, trapmf, trimf, zmf
x = 0:100; params = [0.5 20 -0.3 60]; y1 = psigmf(x, params); params = [0.3 20 -0.2 60]; y2 = psigmf(x, params); params = [0.2 20 -0.1 60]; y3 = psigmf(x, params); figure('NumberTitle', 'off', 'Name', 'psigmf demo'); plot(x, y1, 'r;params = [0.5 20 -0.3 60];', 'LineWidth', 2) hold on; plot(x, y2, 'b;params = [0.3 20 -0.2 60];', 'LineWidth', 2) hold on; plot(x, y3, 'g;params = [0.2 20 -0.1 60];', 'LineWidth', 2) ylim([-0.1 1.1]); xlabel('Crisp Input Value', 'FontWeight', 'bold'); ylabel('Degree of Membership', 'FontWeight', 'bold'); grid; |