I created this Python function to generate a sigmoid function where I can modify position and width:

import numpy as np
import matplotlib.pyplot as plt

def sigmoid(x,a,b):
# sigmoid function with parameters a = center; b = width
return 1/(1+np.exp(-(x-a)/b))

For example changing the parameter b I can make it wider or narrower:

# testing changing sigmoid width (and slope at the same time) - parameter b
 x = np.linspace(0,10,256)
 y = sigmoid(x,5,1) # default
 ymax = sigmoid(x,5,1.75)
 ymin = sigmoid(x,5,0.25)
 # Create the plot

The above is fairly straightforward.

But what if I wanted a rotated sigmoid function? I simulated my intended result "graphically" in the figure below:

enter image description here

which I plotted by just creating a new independent variable x ('test' variable below) and swapping x and y:


Is there a way to achieve this shape with a new sigmoid-like function, or to rotate the original?

It has occurred to me that I could use something like -np.sinh(x) but what I'd really like is to have a similar exponential expression with parameters a and b to control the curve's shape.

This question is also posted on Stack Overflow https://stackoverflow.com/questions/25851287/how-to-generate-a-rotated-by-90-degrees-logistic-sigmoid-function-in-python

  • $\begingroup$ What about applying a transformation (rotation in this case) to your $x-y$ data? $\endgroup$ – nicoguaro Sep 16 '14 at 18:42
  • $\begingroup$ Hi nicoguaro, that's one of the things I was thinking about but I don't have any prior experience with it. HOw would you do it? $\endgroup$ – MyCarta Sep 17 '14 at 2:09
  • $\begingroup$ You can rotate a vector with coordinates $(x,y)$ by an angle $\theta$ using a rotation matrix, like $$\begin{pmatrix} x' \\ y' \end{pmatrix} = \begin{pmatrix} \cos\theta &-\sin\theta\\ \sin\theta & \cos\theta \end{pmatrix} \begin{pmatrix} x \\ y \end{pmatrix}$$ Just as described in Wikipedia. $\endgroup$ – nicoguaro Sep 17 '14 at 2:17
  • $\begingroup$ OK, I've use that before in other subject matters. I will give it a try, have to figure out how to do it in Python. Thanks a lot! $\endgroup$ – MyCarta Sep 17 '14 at 2:29
  • 1
    $\begingroup$ You can use numpy, like import numpy as np R = np.array([[np.cos(t),-np.sin(t)],[np.sin(t),np.cos(t)]]) X_new = np.dot(R,X_old) where X_old is a $2\times n$ array, $n$ the number of data you have. $\endgroup$ – nicoguaro Sep 17 '14 at 2:33

Are you sure you don't want the logit function?

| cite | improve this answer | |

I am turning my comments into an answer as a petition of Mycarta. One easy option that I would use is to rotate the $(x,y)$. To rotate a vector with coordinates $(x,y)$ by an angle $\theta$ you can use a rotation matrix $$\begin{pmatrix} x' \\ y' \end{pmatrix} = \begin{pmatrix} \cos\theta &-\sin\theta\\ \sin\theta & \cos\theta \end{pmatrix} \begin{pmatrix} x \\ y \end{pmatrix}$$

In Python this can be easily done as

import numpy as np
R = np.array([[np.cos(t),-np.sin(t)],[np.sin(t),np.cos(t)]])
X_new = np.dot(R,X_old)

where X_old is a $2\times n$ array, being $n$ the number of data you have.

| cite | improve this answer | |
  • $\begingroup$ Hi nicoguaro, I choose k20's answer as it is the best out of two valid solutions to my problem. As soon as my reputation reaches 15 I will come back and upvote your answer. And thanks for the explanations. $\endgroup$ – MyCarta Sep 19 '14 at 0:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.