simple_markov.py

class SimpleMCMC(object):
    def __init__(self, start_state, transition, goodness_measure):
        self.present_state = start_state
        self.transition = transition
        self.goodness_measure = goodness_measure

        self.present_goodness = self.goodness_measure(self.present_state)
        
    def take_step(self):
        # Use the transition function to find a candidate for the new state.
        # Compute the ratio of the candidate goodness over the present-state
        # goodness. If this ratio is >= 1: move to the candidate state. If it
        # is < 1: Then move to the candidate state with probability that ratio.
        
        candidate = self.transition(self.present_state)
        candidate_goodness = self.goodness_measure(candidate)
        if self.present_goodness == 0:
            if candidate_goodness == 0:
                relative_goodness = 0.5
            else:
                relative_goodness = 1.0
        else:
            relative_goodness = candidate_goodness/self.present_goodness
        
        if relative_goodness >= 1.0:
            self.present_state = candidate
            self.present_goodness = candidate_goodness
            
        else:
            chance = random.random()
            if chance <= relative_goodness:
                self.present_state = candidate
                self.present_goodness = candidate_goodness
        
    
    def run_chain(self, num_iterations):
        for k in range(num_iterations):
            self.take_step()            
        return self.present_state