Exponential approximation for vector input

I was double checking the behaviour of a sigmoid function used in my Simulink model and I noticed that I was getting incorrect approximations when I made the computation for a vector of values

vect =  [-5.0000   -5.0000   -5.0000    1.0000    0.9000    0.8000    0.7000   -5.0000   -5.0000];
y_vect = 1/(1+exp(-2*(vect'-1)));
% Value calculated using the vector
y_vect(4)

ans = 0

% Value calculated alone
y_val =  1/(1+exp(-2*(vect(4)-1)))

y_val = 0.5000

This approximation in my case causes great confussion due to the magnitude of the quantity expected.

Is there any way to solve this?

Answer 1

[Sulaymon Eshkabilov]

You have overlooked one dot. Here is the corrected commands:

vect =  [-5.0000   -5.0000   -5.0000    1.0000    0.9000    0.8000    0.7000   -5.0000   -5.0000];
y_vect = 1./(1+exp(-2*(vect-1)));
% Value calculated using the vector
y_vect(4)

ans = 0.5000

% Value calculated alone
y_val =  1/(1+exp(-2*(vect(4)-1)))

y_val = 0.5000

[Eduardo]

Oh nice to know!

I wrongly thought the broadcasting would be done automatically since we just had a scalar in the numerator

Answer 2

[Voss]

vect =  [-5.0000   -5.0000   -5.0000    1.0000    0.9000    0.8000    0.7000   -5.0000   -5.0000];

Using / (matrix right division), as you have it now:

y_vect = 1/(1+exp(-2*(vect'-1)));
disp(y_vect)

   1.0e-05 *

    0.6144         0         0         0         0         0         0         0         0

Using ./ (element-wise right division):

y_vect = 1./(1+exp(-2*(vect'-1)));
disp(y_vect)

    0.0000
    0.0000
    0.0000
    0.5000
    0.4502
    0.4013
    0.3543
    0.0000
    0.0000

https://www.mathworks.com/help/matlab/matlab_prog/matlab-operators-and-special-characters.html