A stupid Markov Model :)

Markov model playground. Currently only handles discrete-time, discrete-space, first order Markov chains.

Requirements

Python 3

General markov model

The class MarkovModel represents a general first-order, space-discrete and time-discrete markov model.

It has to be initialized with its space (a set of all possible states) like:

model = MarkovModel({'A', 'B', 'C'}) // we create a model with 3 different states

It then has to be trained with a markov chain of those states:

model.train(['A', 'A', 'A', 'B', 'A', 'C', 'C', 'A', 'A', 'A', 'A', 'A']) // we train it against this ordered sequence of states

And it can then predict a the next state of a state:

model.next_state('A') // will likely return A, since most of the times, A are followed by A during the training phase

Or predict the probability of a sequence:

model.probability_of_chain(['A', 'C', 'C']) // the probability for this sequence to be generated

It can even create a sequence from a starting state:

model.generate_chain('C', 5) // generate a 5-states sequence beginning with 'C', using the probability of the dataset it has be trained with

Available Markov Models implementations

Text Markov Model

The time is the order of appearance of tokens (words) in a text, the space is made of the set of words.

Can generate a sentence after having been trained with a text. Code example: The sentences will look like english sentences, but will have no meaning.

from text import TextHMM

with open('text.txt', 'r') as file:
    text = file.read()
hmm = TextHMM(text)  # We create the HMM from the text
hmm.train()  # We train it
print(' '.join(hmm.generate_chain("the", 7)))  # We generate 7 words, starting with "the"

the very good sort of injustice towards Highbury

The example test text.txt comes from Shakespeare.

Improvements to be made

Handle not only first order Markov Model but n-th order chains
Live training (being able to use the model as we train it, that is being to update the probability matrix after having created it)

brendan-rius / markov-model