def run(self, ngen=10, startfile=None, checkpoint=None):
"""Run the genetic algorithm, updating the population over ngen number of generations.
Keywork arguments:
ngen -- number of generations to run the genetic algorithm.
startfile -- File containing existing population (default None)
checkpoint -- File containing existing checkpoint (default None)
Output:
population -- Resulting population after ngen generations.
"""
if startfile:
population = self.readpop(startfile)
else:
population = self.newpopulation()
if checkpoint:
self.writepop(population, filename=f"0_{checkpoint}")
for cur_g in range(1, ngen + 1):
print(f"generation {cur_g} of population size {len(population)}")
population = self.nextgen(population)
seg = Segmentors.algoFromParams(self.hof[0])
mask = seg.evaluate(self.img)
fitness = Segmentors.FitnessFunction(self.mask, mask)
print(f"#BEST - {fitness[0]} - {self.hof[0]}")
if checkpoint:
print(f"Writing Checkpoint file - {checkpoint}")
self.writepop(population, filename=f"{checkpoint}_{cur_g}")
for cur_p in range(len(population)):
logging.getLogger().info(population[cur_p])
return population
def func(img=img, mask=gmask, **kwargs):
"""Find mask and fitness for current algorithm. Show masked image."""
print(seg.params["algorithm"])
for k in kwargs:
seg.params[k] = kwargs[k]
mask = seg.evaluate(img)
fit = Segmentors.FitnessFunction(mask, gmask)
fig = showtwo(img, mask)
def run(self, ngen=10, population=None,startfile=None, checkpoint=None, cp_freq=1):
"""Run the genetic algorithm, updating the population over ngen number of generations.
Keywork arguments:
ngen -- number of generations to run the genetic algorithm.
startfile -- File containing existing population (default None)
checkpoint -- File containing existing checkpoint (default None)
Output:
population -- Resulting population after ngen generations.
"""
if startfile:
try:
print(f"Reading in {startfile}")
population = self.readpop(startfile)
except FileNotFoundError:
print("WARNING: Start file not found")
except:
raise
if not population:
print(f"Inicializing new randome population")
population = self.newpopulation()
if checkpoint:
self.writepop(population, filename=f"{checkpoint}")
for cur_g in range(0, ngen+1):
print(f"generation {cur_g} of population size {len(population)}")
_, population = self.popfitness(population)
bestsofar = self.hof[0]
seg = Segmentors.algoFromParams(bestsofar)
mask = seg.evaluate(self.img)
fitness = Segmentors.FitnessFunction(mask, self.mask)
print(f"#BEST - {fitness} - {bestsofar}")
if checkpoint and cur_g%cp_freq == 0:
print(f"Writing Checkpoint file - {checkpoint}")
copyfile(f"{checkpoint}", f"{checkpoint}.prev")
self.writepop(population, filename=f"{checkpoint}")
for cur_p in range(len(population)):
logging.getLogger().info(population[cur_p])
if cur_g < ngen+1:
population = self.mutate(population)
if checkpoint:
print(f"Writing Checkpoint file - {checkpoint}")
copyfile(f"{checkpoint}", f"{checkpoint}.prev")
self.writepop(population, filename=f"{checkpoint}")
for cur_p in range(len(population)):
logging.getLogger().info(population[cur_p])
return population
def continuous_search(input_file,
input_mask,
startfile='',
checkpoint='checkpoint.txt',
best_mask_file="temp_mask.png",
pop_size=10):
img = imageio.imread(input_file)
gmask = imageio.imread(input_mask)
fid_out = open(f"{input_file}.txt", "w+")
#TODO: Read this file in and set population first
#Run the search
my_evolver = GeneticSearch.Evolver(img, gmask, pop_size=pop_size)
best_fitness = 2.0
iteration = 0
while (best_fitness > 0.0):
print(f"running {iteration} iteration")
if (startfile):
population = my_evolver.run(ngen=1, startfile=None)
startfile = None
else:
population = my_evolver.run(ngen=1)
#Get the best algorithm so far
params = my_evolver.hof[0]
#Generate mask of best so far.
seg = Segmentors.algoFromParams(params)
mask = seg.evaluate(img)
fitness = Segmentors.FitnessFunction(mask, gmask)[0]
if (fitness < best_fitness):
best_fitness = fitness
print(
f"\n\n\n\nIteration {iteration} Finess Improved to {fitness}")
my_evolver.writepop(population, filename="checkpoint.pop")
#imageio.imwrite(best_mask_file,mask);
fid_out.write(f"[{iteration}, {fitness}, {params}]\n")
fid_out.flush()
###TODO Output [fitness, seg]
iteration += 1
def conduct_genetic_search(img, gmask, num_gen, pop_size):
"""
Note: this task could be sped up by
rewriting it to send the evaluation and fitness function
calls to other workers as tasks.
"""
# Only needed on some images
# Convert the RGB 3-channel image into a 1-channel image
# gmask = (np.sum(gmask, axis=2) > 0)
download_and_store_image(img)
download_and_store_image(gmask)
img = imageio.imread(get_path(img))
gmask = imageio.imread(get_path(gmask))
my_evolver = GeneticSearch.Evolver(img, gmask, pop_size=pop_size)
# Conduct the genetic search
population = None
for _ in range(num_gen):
# if population is uninitialized
if population is None:
population = my_evolver.run(ngen=1)
else:
# Simulate a generation and store population in population variable
population = my_evolver.run(ngen=1, population=population)
# Take the best segmentor from the hof and use it to segment the rgb image
seg = Segmentors.algoFromParams(my_evolver.hof[0])
mask = seg.evaluate(img)
# Calculate and print the fitness value of the segmentor
fitness = Segmentors.FitnessFunction(mask, gmask)[0]
params = my_evolver.hof[0]
# Combine data into a single object
data = {}
data["fitness"] = fitness
data["params"] = params
return data
def test_FitnessFunction():
"""Unit test for FitnessFunction function. Checks fitness value is as
expected for a variety of extreme cases."""
# create test image
ground_truth = np.zeros((20, 20))
ground_truth[4:10, 4:10] = 1
inferred = np.zeros((20, 20))
inferred[4:10, 4:6] = 1
inferred[4:10, 6:10] = 2
assert Segmentors.FitnessFunction(inferred, ground_truth) == [
2**np.log(3),
]
inferred = np.zeros((20, 20))
inferred[4:10, 3:6] = 1
inferred[4:10, 6:10] = 2
assert Segmentors.FitnessFunction(inferred, ground_truth) == [
8**np.log(3),
]
inferred = np.zeros((20, 20))
inferred[4:10, 3:6] = 1
inferred[4:10, 6:10] = 2
inferred[3:5, 3:6] = 3
assert Segmentors.FitnessFunction(inferred, ground_truth) == [
9**np.log(4),
]
inferred = np.zeros((20, 20))
assert Segmentors.FitnessFunction(inferred, ground_truth) == [
sys.maxsize,
]
inferred = np.arange(400).reshape(ground_truth.shape)
assert Segmentors.FitnessFunction(inferred, ground_truth) == [
2**np.log(400),
]
inferred = np.zeros((20, 20))
inferred[1:19, 1:19] = 1
assert Segmentors.FitnessFunction(inferred, ground_truth) == [
sys.maxsize,
]
args = parser.parse_args()
print(args)
random.seed(args.seed)
logging.basicConfig(stream=sys.stdout, level=logging.ERROR)
#logging.basicConfig(stream=sys.stdout, level=logging.INFO)
img = imageio.imread(args.image)
gmask = imageio.imread(args.mask)
if len(gmask.shape) > 2:
gmask = color.rgb2gray(gmask)
ee = GeneticSearch.Evolver(img, gmask, pop_size=args.pop)
ee.run(args.generations, checkpoint=args.checkpointfile)
seg = Segmentors.algoFromParams(ee.hof[0])
mask = seg.evaluate(img)
imageio.imwrite(args.outputfolder+"file.jpg", mask)
fitness,_ = Segmentors.FitnessFunction(mask,gmask)
print(f"{fitness} {ee.hof[0]}")
def run(self, ngen=10, population=None,startfile=None, checkpoint=None, cp_freq=1):
"""Run the genetic algorithm, updating the population over ngen number of generations.
Keywork arguments:
ngen -- number of generations to run the genetic algorithm.
startfile -- File containing existing population (default None)
checkpoint -- File containing existing checkpoint (default None)
Output:
population -- Resulting population after ngen generations.
"""
if startfile:
try:
print(f"Reading in {startfile}")
population = self.readpop(startfile)
except FileNotFoundError:
print("WARNING: Start file not found")
except:
raise
if not population:
print(f"Initializing a new random population")
population = self.newpopulation()
if checkpoint:
self.writepop(population, filename=f"{checkpoint}")
for cur_g in range(0, ngen+1):
print(f"Generation {cur_g} of population size {len(population)}")
histogram = Segmentors.popCounts(population)
print(f"#HIST - {histogram}")
_, population = self.popfitness(population)
bestsofar = self.hof[0]
seg = Segmentors.algoFromParams(bestsofar)
mask = seg.evaluate(self.img)
fitness = Segmentors.FitnessFunction(mask, self.mask)
print(f"#BEST - {fitness} - {bestsofar}")
if checkpoint and cur_g%cp_freq == 0:
print(f"Writing Checkpoint file - {checkpoint}")
copyfile(f"{checkpoint}", f"{checkpoint}.prev")
self.writepop(population, filename=f"{checkpoint}")
for cur_p in range(len(population)):
logging.getLogger().info(population[cur_p])
if cur_g < ngen+1:
if bestsofar.fitness.values[0] >= 0.95:
population = self.newpopulation()
# if the best fitness value is at or above the
# threshold of 0.95, discard the entire current
# population and randomly select a new population
# for the next generation
# note: setting keep_prob = 0 and mutate_prob = 1
# as mutate arguments
# should have same result as self.new_population()
else:
population = self.mutate(population)
# if the best fitness value is below this threshold,
# proceed as normal, mutating the current population
# to get the next generation
if checkpoint:
print(f"Writing Checkpoint file - {checkpoint}")
copyfile(f"{checkpoint}", f"{checkpoint}.prev")
self.writepop(population, filename=f"{checkpoint}")
for cur_p in range(len(population)):
logging.getLogger().info(population[cur_p])
return population
