Sebastian Raschka
last modified: 03/31/2014
$ p(x | \omega_j) = \frac{1}{\pi b} \; \dot \; \frac{1}{1 \; + \; \bigg(\frac{x - a_i}{b} \bigg)^2} \quad i = 1, 2$
$ P(\omega_1) = P(\omega_2) = 0.5 $
where
$ \lambda(\alpha_i|\omega_j) = \lambda_{ij}$,
the loss occured if $action_i$ is taken if the actual true class is $ \omega_j$ (assuming that $action_i$ classifies sample as $\omega_i$)
\begin{equation} \lambda = \begin{pmatrix} \lambda_{11} \quad \lambda_{12} \\ \lambda_{21} \quad \lambda_{22} \end{pmatrix} = \begin{pmatrix} 0 \quad 1 \\ 1 \quad 0 \\ \end{pmatrix}\\ \end{equation}$ a_1 = 0, \quad a_2 = 2 ,\quad b=0.5$
$ \Rightarrow \lambda_{21}P(\omega_1|x) + \lambda_{22}P(\omega_2|x) > \lambda_{11}P(\omega_1|x) + \lambda_{12}P(\omega_2|x) $
$ \Rightarrow (\lambda_{21} - \lambda_{11}) \; P(\omega_1|x) > (\lambda_{12} - \lambda_{22}) \; P(\omega_2|x) $
$ \Rightarrow \frac{P(\omega_1|x)}{P(\omega_2|x)} > \frac{(\lambda_{12} - \lambda_{22})}{(\lambda_{21} - \lambda_{11})} $
$ \Rightarrow \frac{p(x|\omega_1) \; P{(\omega_1)}}{p(x|\omega_2) \; P{(\omega_2)}} > \frac{(\lambda_{12} - \lambda_{22})}{(\lambda_{21} - \lambda_{11})} $
$ \Rightarrow \frac{p(x|\omega_1)}{p(x|\omega_2)} > \frac{(\lambda_{12} - \lambda_{22}) \; P{(\omega_2)}}{(\lambda_{21} - \lambda_{11}) \; P{(\omega_1)}} $
$ \Rightarrow \frac{p(x|\omega_1)}{P(x|\omega_2)} > \frac{(1 - 0) \; (0.5)}{(1 - 0) \; (0.5)} $
$ \Rightarrow \frac{p(x|\omega_1)}{P(x|\omega_2)} > 1 $
$\Rightarrow \frac{1}{\pi b} \; \dot \; \frac{1}{1 \; + \; \bigg(\frac{x - a_1}{b} \bigg)^2} > \frac{1}{\pi b} \; \dot \; \frac{1}{1 \; + \; \bigg(\frac{x - a_2}{b} \bigg)^2}$
$\Rightarrow \frac{1}{1 \; + \; \bigg(\frac{x - a_1}{b} \bigg)^2} > \frac{1}{1 \; + \; \bigg(\frac{x - a_2}{b} \bigg)^2}$
$\Rightarrow \bigg(\frac{x - a_1}{b} \bigg)^2 > \bigg(\frac{x - a_2}{b} \bigg)^2$
$\Rightarrow x-a_1 > x-a_2 $
$ \Rightarrow x < \frac{a_1+a_2}{2} $
Decide $ \omega_1 $ if $ x < \frac{a_1+a_2}{2}$ else decide $ \omega_2 $.
In [12]:
%pylab inline
import numpy as np
from matplotlib import pyplot as plt
def cauchy(x, a, b):
"""
Calculates the Cauchy distribution's probability density
function.
"""
term1 = 1 / (np.pi * b)
term2 = 1 / (1 + ((x - a)/b)**2)
return term1 * term2
def decision_boundary(a1, a2):
return (a1 + a2) / 2
def posterior(likelihood, prior):
"""
Calculates the posterior probability (after Bayes Rule) without
the scale factor p(x) (=evidence).
"""
return likelihood * prior
# generating some sample data for plotting the posterior probabilities
x = np.arange(-10, 10, 0.05)
# probability density functions
posterior1 = posterior(cauchy(x, a=0, b=0.5), prior=0.5)
posterior2 = posterior(cauchy(x, a=2, b=0.5), prior=0.5)
# plotting probability density functions
plt.plot(x, posterior1)
plt.plot(x, posterior2)
plt.title('Posterior Probabilities w. Decision Boundary for a Cauchy Distribution')
plt.ylabel('P(w)')
plt.xlabel('random variable x')
plt.legend(['P(w_1|x)', 'P(w_2|x)'], loc='upper right')
plt.ylim([0,0.4])
plt.xlim([-4,6])
# plotting the decision boundary
bound = decision_boundary(a1=0, a2=2)
plt.vlines(bound, -10, 10, color='r', alpha=0.8, linestyle=':', linewidth=2)
plt.annotate('R1', xy=(0.0, 0.35), xytext=(0.0, 0.35))
plt.annotate('R2', xy=(2.0, 0.35), xytext=(2.0, 0.35))
plt.show()
In [23]:
import math
import random as rnd
def cauchy(location, scale):
"""
Generates random cauchy distributed data
reference: http://www.johndcook.com/python_cauchy_rng.html
"""
p = 0.0
while p == 0.0:
p = rnd.random()
return location + scale*math.tan(math.pi*(p - 0.5))
# Parameters
a_1 = 0
a_2 = 2
b = 0.5
# Generating 20 Cauchy distributed random samples for class 1 & 2
x1_samples = [cauchy(a_1, b) for i in range(20)]
x2_samples = [cauchy(a_2, b) for i in range(20)]
y = [0.1 for i in range(20)]
plt.scatter(x1_samples, y, marker='o', color='green', s=40, alpha=0.5)
plt.scatter(x2_samples, y, marker='^', color='blue', s=40, alpha=0.5)
plt.ylim([0,0.4])
plt.xlim([-4,6])
# Plotting the decision boundary
bound = decision_boundary(a1=0, a2=2)
plt.vlines(bound, -10, 10, color='r', alpha=0.8, linestyle=':', linewidth=2)
plt.annotate('R1', xy=(0.0, 0.2), xytext=(0.0, 0.2))
plt.annotate('R2', xy=(2.0, 0.2), xytext=(2.0, 0.2))
# Plot annotation
plt.title('Classifying random example data from 2 classes')
plt.ylabel('')
plt.xlabel('random variable x')
plt.legend(['w_1', 'w_2'], loc='upper right')
plt.show()
In [21]:
w1_as_w2, w2_as_w1 = 0, 0
for x1,x2 in zip(x1_samples, x2_samples):
if x1 > decision_boundary(a1=0, a2=2):
w1_as_w2 += 1
if x2 <= decision_boundary(a1=0, a2=2):
w2_as_w1 += 1
emp_err = (w1_as_w2 + w2_as_w1) / float(len(x1_samples) + len(x2_samples))
print('Empirical Error: {}%'.format(emp_err * 100))
$ P(error) = \int_{x=-\infty}^{\infty} P(error/x) \; \cdot \; p(x) \; dx $
$ P(error|x) = \left\{ \begin{array}{l l} P(\omega_1 | x) & \quad \text{if $x > \frac{a_1 + a_2}{2}$}\\ P(\omega_2 | x) & \quad \text{if $x < \frac{a_1 + a_2}{2}$} \end{array} \right.$
$ \Rightarrow \int_{x=-\infty}^{\frac{a_1+a_2}{2}} P(\omega_2|x) \; \cdot \; p(x) \; dx + \int_{x=\frac{a_1+a_2}{2}}^{\infty} P(\omega_1|x) \; \cdot \; p(x) \; dx$
$ \Rightarrow \int_{x=-\infty}^{\frac{a_1+a_2}{2}} p(x|\omega_2) \; \cdot \; P(\omega_2) \; dx + \int_{x=\frac{a_1+a_2}{2}}^{\infty} p(x|\omega_1) \; \cdot \; P(\omega_1) \; dx$
$ \Rightarrow \int_{x=-\infty}^{\frac{a_1+a_2}{2}} \frac{1}{\pi b} \; \dot \; \frac{1}{1 \; + \; \bigg(\frac{x - a_2}{b} \bigg)^2} \; \frac{1}{2} \; dx + \int_{x=\frac{a_1+a_2}{2}}^{\infty} \frac{1}{\pi b} \; \dot \; \frac{1}{1 \; + \; \bigg(\frac{x - a_1}{b} \bigg)^2} \; \frac{1}{2} \; dx$
$ \Rightarrow \frac{1}{2} - \frac{1}{\pi} \; tan^{-1} \bigg| \frac{a_2 - a_1}{2b} \bigg|$
In [11]:
def bayes_error(a1, a2, b):
""" Calculates the Bayes Error for 2 Cauchy densities ."""
term1 = 1/2.0 - 1/np.pi
term2 = np.arctan(np.fabs((a2-a1)/(2*b)))
return term1 * term2
print('Bayes Error: {:.2f}%'.format(100* bayes_error(0, 2, 0.5)))
In [ ]: