CODEWITHSUNDEEP
pythonplotellipseeigenvector
Remz1337
Remz1337

Reputation: 175

Plotting confidence ellipse with eigen vectors in 2D

I am trying to replicate the plot demonstrated in this guide, where a confidence ellipse is shown on top of a scatter plot, along with the eigen vectors. After scavenging the internet for a Python example without success.

Upvotes: 1

Views: 1570

Answers (1)

Remz1337
Remz1337

Reputation: 175

I decided to port the Matlab code provided in the article to Python.

I am sharing my work as I could not find a similar example. Here is the resulting plot:

enter image description here

Here is my Python code:

import traceback
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.patches import Ellipse
from numpy import linalg as LA
import matplotlib.transforms as transforms

def plot_ellipse():

    # Create some random data
    s = np.array([2,2])
    x = np.random.randn(334);
    y1 = np.random.normal(s[0] * x, 1)
    y2 = np.random.normal(s[1] * x, 1)
    data = np.array([y1,y2])

    # Calculate the eigenvectors and eigenvalues
    covariance = np.cov(data)
    eigenval, eigenvec = LA.eig(covariance)

    # Get the largest eigenvalue
    largest_eigenval = max(eigenval)

    # Get the index of the largest eigenvector
    largest_eigenvec_ind_c = np.argwhere(eigenval == max(eigenval))[0][0]
    largest_eigenvec = eigenvec[:,largest_eigenvec_ind_c]


    # Get the smallest eigenvector and eigenvalue
    smallest_eigenval = min(eigenval)
    if largest_eigenvec_ind_c == 0:
        smallest_eigenvec = eigenvec[:,1]
    else:
        smallest_eigenvec = eigenvec[:,0]

    # Calculate the angle between the x-axis and the largest eigenvector
    angle = np.arctan2(largest_eigenvec[1], largest_eigenvec[0]);

    # This angle is between -pi and pi.
    # Let's shift it such that the angle is between 0 and 2pi
    if (angle < 0):
        angle = angle + 2*np.pi;

    # Get the coordinates of the data mean
    avg_0 = np.mean(data[0]);
    avg_1 = np.mean(data[1]);

    #% Get the 95% confidence interval error ellipse
    chisquare_val = 2.4477;
    X0=avg_0;
    Y0=avg_1;

    ax=plt.gca()

    pearson = covariance[0, 1]/np.sqrt(covariance[0, 0] * covariance[1, 1])

    ell_radius_x = np.sqrt(1 + pearson)
    ell_radius_y = np.sqrt(1 - pearson)
    ellipse = Ellipse((0, 0), width=ell_radius_x * 2, height=ell_radius_y * 2, facecolor='none', edgecolor=(1, 0, 0.2))


    scale_x = np.sqrt(covariance[0, 0] * chisquare_val)
    mean_x = X0
    scale_y = np.sqrt(covariance[1, 1] * chisquare_val)
    mean_y = Y0

    transf = transforms.Affine2D().rotate_deg(45).scale(scale_x, scale_y).translate(mean_x, mean_y)
    ellipse.set_transform(transf + ax.transData)
    ax.add_patch(ellipse)


    # Plot the original data
    plt.plot(data[0],data[1], '.');
    plt.xlim([min(data[0])-3, max(data[0])+3]);
    plt.ylim([min(data[1])-3, max(data[1])+3]);

    # Plot the eigenvectors
    quiveropts = dict(headaxislength=0, color='red', headlength=0, units='xy',angles='xy',scale=1)

    plt.quiver(X0, Y0, largest_eigenvec[0]*np.sqrt(largest_eigenval), largest_eigenvec[1]*np.sqrt(largest_eigenval), **quiveropts)
    plt.quiver(X0, Y0, smallest_eigenvec[0]*np.sqrt(smallest_eigenval), smallest_eigenvec[1]*np.sqrt(smallest_eigenval), **quiveropts)

    plt.show()

    return

if __name__ == '__main__':
    try:
        plot_ellipse()
    except Exception as e:
        print("Caught exception:" + str(e))
        print(traceback.format_exc())

Upvotes: 3

Related Questions