lcford2 / cdhmm

An easy to use python library consisting implementation of Continuous Density Hidden Markov Models.

cdhmm

An easy to use python library consisting implementation of Continuous Density Hidden Markov Models.

Continuous Density Hidden Markov Models(CD-HMM) are a type of HMM which consists of Emission probabilities in the form of a distribution like gaussian or uniform distribution. Despite its use in Speech processing, very less codes are available on the internet regarding CD-HMM. This library has implementation of all HMM algorithms applied on continuous density observations.

After studying Hidden Markov Models(HMM) for a while now, I have came across many python libraries which implements HMM algorithms like forward, backward, Viterbi and Baum-Welch. However, most of these libraries work on discrete observations. cdhmm library is developed for can be used to do HMM operations on Continuous observations.

Supported Algorithms

1. Forward Algorithm

2. Backward Algorithm

3. Occupation Likelihood

4. Baum Welch Algorithm

Prerequisites

Install numpy, scipy and seaborn by executing following commands:

pip install numpy
pip install scipy
pip install seaborn

Installation

After installing prerequisites, install cdhmm library by executing following command:

pip install cdhmm-ashishmokalkar==0.0.2

Project Page

For Official project page, visit following website:

https://pypi.org/project/cdhmm-ashishmokalkar/0.0.2/

Tutorial

Import cdhmm library using following code:

import cdhmm as cd

Consider a Continuous density HMM with 2 states and 7 continuous observations.

import numpy as np

observations = [1.2, 2.4, 2.9, 3.7, 4.1, 4.8, 6.2]
transition_probs = np.array([[0.68, 0.29], [0.18, 0.91]])
pi = np.array([0.83, 0.24])  #initial probability 
means = [3.00, 5.00]
variances = [1.44, 0.64]
exit_p = [0.07, 0.06]

Now, building HMM model using these parameters is as simple as executing the following code

hmms = cd.simple_cdhmm(obs, means, variances, pi, transition_probs, exit_p)

Get probability density values at observation points :

Em = hmms.get_probability_densities()
print (Em)

Forward Algorithm:

alpha = hmms.forward()
print("Alphas :: ")
print(alpha)

Forward Likelihood

like_f = hmms.likelihood()
print("Forward Likelihood :: ")
print (like_f)

Backward Algorithm

beta = hmms.backward()
print("Beta :: ")
print (beta)

Backward Likelihood

like_b = hmms.likelihood_b()
print("Backward Likelihood :: ")
print (like_b)

occupation likelihood

occ = hmms.get_occupation_likelihoods()
print("Occupation likelihoods :: ")
print (occ)

Re-estimate Means

new_means = hmms.getNewMean()
print("New means :: ")
print (new_means)

Re-estimate Variances

new_vars = hmms.getNewVariance()
print("New Variances :: ")
print (new_vars)

For any queries regarding the library or open-source contribution, email me: [e-mail]