def observe(self, state):
"""Generate a particulat observation for the current state"""
# Compile a distribution over observations in current state
obs_dist = {obs: self.pe(state, obs)
for obs in self.get_observations()}
# Sample from the distribution
return weighted_random_choice(obs_dist)
def selection(chromosomes, fitnesses, method):
"""
Implements selection amongst generations
:param chromosomes: population
:param fitnesses: fitness values
:param method: Choice between different selection strategies
:return: new population
"""
if method is None:
method = 'fitness_proportionate'
dict_fitnesses = fitnesses.to_dict()
n_population = chromosomes.shape[0]
selected_individuals = pd.DataFrame(data=None, columns=chromosomes.columns)
# Or use df.sample(weights='a')
if method == 'fitness_proportionate':
for _ in range(n_population):
loc = utils.weighted_random_choice(dict_fitnesses)
selected_individuals = selected_individuals.append(chromosomes.iloc[loc])
else:
selected_individuals = chromosomes.copy()
return selected_individuals.reset_index(drop=True)
def step(self, state):
"""Make a time step: generate a particular next state for the current state"""
# Compile a distribution over next states
next_dist = {next_state: self.pt(state, next_state)
for next_state in self.get_targets(state)}
# Sample from the distribution
return weighted_random_choice(next_dist)
def next_move_dir(self):
"""Return the direction of next move"""
return weighted_random_choice(self.move_probs)
def read(self):
"""Return a single sensor reading depending on robot position"""
p = self.get_value_probabilities()
return weighted_random_choice(p)