def random_population(genome_alphabet, genome_size, num_organisms,
fitness_calculator):
"""Generate a population of individuals with randomly set genomes.
Arguments:
o genome_alphabet -- An Alphabet object describing all of the
possible letters that could potentially be in the genome of an
organism.
o genome_size -- The size of each organisms genome.
o num_organism -- The number of organisms we want in the population.
o fitness_calculator -- A function that will calculate the fitness
of the organism when given the organisms genome.
"""
all_orgs = []
# a random number generator to get letters for the genome
letter_rand = random.Random()
# figure out what type of characters are in the alphabet
if isinstance(genome_alphabet.letters[0], str):
if sys.version_info[0] == 3:
alphabet_type = "u" # Use unicode string on Python 3
else:
alphabet_type = "c" # Use byte string on Python 2
elif isinstance(genome_alphabet.letters[0], int):
alphabet_type = "i"
elif isinstance(genome_alphabet.letters[0], float):
alphabet_type = "d"
else:
raise ValueError(
"Alphabet type is unsupported: %s" % genome_alphabet.letters)
for org_num in range(num_organisms):
new_genome = MutableSeq(array.array(alphabet_type), genome_alphabet)
# generate the genome randomly
for gene_num in range(genome_size):
new_gene = letter_rand.choice(genome_alphabet.letters)
new_genome.append(new_gene)
# add the new organism with this genome
all_orgs.append(Organism(new_genome, fitness_calculator))
return all_orgs
def random_population(genome_alphabet, genome_size, num_organisms,
fitness_calculator):
"""Generate a population of individuals with randomly set genomes.
Arguments:
o genome_alphabet -- An Alphabet object describing all of the
possible letters that could potentially be in the genome of an
organism.
o genome_size -- The size of each organisms genome.
o num_organism -- The number of organisms we want in the population.
o fitness_calculator -- A function that will calculate the fitness
of the organism when given the organisms genome.
"""
all_orgs = []
# a random number generator to get letters for the genome
letter_rand = random.Random()
# figure out what type of characters are in the alphabet
if isinstance(genome_alphabet.letters[0], str):
if sys.version_info[0] == 3:
alphabet_type = "u" # Use unicode string on Python 3
else:
alphabet_type = "c" # Use byte string on Python 2
elif isinstance(genome_alphabet.letters[0], int):
alphabet_type = "i"
elif isinstance(genome_alphabet.letters[0], float):
alphabet_type = "d"
else:
raise ValueError("Alphabet type is unsupported: %s" %
genome_alphabet.letters)
for org_num in range(num_organisms):
new_genome = MutableSeq(array.array(alphabet_type), genome_alphabet)
# generate the genome randomly
for gene_num in range(genome_size):
new_gene = letter_rand.choice(genome_alphabet.letters)
new_genome.append(new_gene)
# add the new organism with this genome
all_orgs.append(Organism(new_genome, fitness_calculator))
return all_orgs
def random_motif(self):
"""Create a random motif within the given parameters.
This returns a single motif string with letters from the given
alphabet. The size of the motif will be randomly chosen between
max_size and min_size.
"""
motif_size = random.randrange(self._min_size, self._max_size)
motif = ""
for letter_num in range(motif_size):
cur_letter = random.choice(self._alphabet.letters)
motif += cur_letter
return MutableSeq(motif, self._alphabet)
def viterbi(self, sequence, state_alphabet):
"""Calculate the most probable state path using the Viterbi algorithm.
This implements the Viterbi algorithm (see pgs 55-57 in Durbin et
al for a full explanation -- this is where I took my implementation
ideas from), to allow decoding of the state path, given a sequence
of emissions.
Arguments:
o sequence -- A Seq object with the emission sequence that we
want to decode.
o state_alphabet -- The alphabet of the possible state sequences
that can be generated.
"""
# calculate logarithms of the initial, transition, and emission probs
log_initial = self._log_transform(self.initial_prob)
log_trans = self._log_transform(self.transition_prob)
log_emission = self._log_transform(self.emission_prob)
viterbi_probs = {}
pred_state_seq = {}
state_letters = state_alphabet.letters
# --- recursion
# loop over the training squence (i = 1 .. L)
# NOTE: My index numbers are one less than what is given in Durbin
# et al, since we are indexing the sequence going from 0 to
# (Length - 1) not 1 to Length, like in Durbin et al.
for i in range(0, len(sequence)):
# loop over all of the possible i-th states in the state path
for cur_state in state_letters:
# e_{l}(x_{i})
emission_part = log_emission[(cur_state, sequence[i])]
max_prob = 0
if i == 0:
# for the first state, use the initial probability rather
# than looking back to previous states
max_prob = log_initial[cur_state]
else:
# loop over all possible (i-1)-th previous states
possible_state_probs = {}
for prev_state in self.transitions_to(cur_state):
# a_{kl}
trans_part = log_trans[(prev_state, cur_state)]
# v_{k}(i - 1)
viterbi_part = viterbi_probs[(prev_state, i - 1)]
cur_prob = viterbi_part + trans_part
possible_state_probs[prev_state] = cur_prob
# calculate the viterbi probability using the max
max_prob = max(possible_state_probs.values())
# v_{k}(i)
viterbi_probs[(cur_state, i)] = (emission_part + max_prob)
if i > 0:
# get the most likely prev_state leading to cur_state
for state in possible_state_probs:
if possible_state_probs[state] == max_prob:
pred_state_seq[(i - 1, cur_state)] = state
break
# --- termination
# calculate the probability of the state path
# loop over all states
all_probs = {}
for state in state_letters:
# v_{k}(L)
all_probs[state] = viterbi_probs[(state, len(sequence) - 1)]
state_path_prob = max(all_probs.values())
# find the last pointer we need to trace back from
last_state = ''
for state in all_probs:
if all_probs[state] == state_path_prob:
last_state = state
assert last_state != '', "Didn't find the last state to trace from!"
# --- traceback
traceback_seq = MutableSeq('', state_alphabet)
loop_seq = list(range(1, len(sequence)))
loop_seq.reverse()
# last_state is the last state in the most probable state sequence.
# Compute that sequence by walking backwards in time. From the i-th
# state in the sequence, find the (i-1)-th state as the most
# probable state preceding the i-th state.
state = last_state
traceback_seq.append(state)
for i in loop_seq:
state = pred_state_seq[(i - 1, state)]
traceback_seq.append(state)
# put the traceback sequence in the proper orientation
traceback_seq.reverse()
return traceback_seq.toseq(), state_path_prob
