def print_best_genome(ga_engine):
best = ga_engine.bestIndividual()
gen = ga_engine.getCurrentGeneration()
raw_fitn = best.getRawScore()
print "Best genome at generation %d had fitness %f (raw %f), error %f." % \
(gen, best.getFitnessScore(), raw_fitn, _fitness_to_error(raw_fitn))
print "Best genome at generation %d:" % gen, pu.raw_genes(best, True)
sys.stdout.flush()
def print_best_genome(ga_engine):
best = ga_engine.bestIndividual()
gen = ga_engine.getCurrentGeneration()
raw_fitn = best.getRawScore()
print "Best genome at generation %d had fitness %f (raw %f), error %f." % \
(gen, best.getFitnessScore(), raw_fitn, _fitness_to_error(raw_fitn))
print "Best genome at generation %d:" % gen, pu.raw_genes(best, True)
sys.stdout.flush()
def _set_indiv(self, indiv):
self.indiv = indiv
self.model = indiv.getParam('model')
self.day = self.model.day
self.loci = self.model.loci
self.genome = pu.raw_genes(indiv, True)
train_iter = self.model.dataset.train_data_iterator()
train_data = [t for t in train_iter()]
(self.data_in, self.data_out) = random.choice(train_data)
def _set_indiv(self, indiv):
self.indiv = indiv
self.model = indiv.getParam('model')
self.day = self.model.day
self.loci = self.model.loci
self.genome = pu.raw_genes(indiv, True)
train_iter = self.model.dataset.train_data_iterator()
train_data = [t for t in train_iter()]
(self.data_in, self.data_out) = random.choice(train_data)
def _fitness(indiv):
"""Generic fitness function, calls preprocessors, transformer and
postprocessors for each model, calculates error from error function."""
global _model
model = _model
num_trials = indiv.getParam('num_trials')
genome = pu.raw_genes(indiv, True)
loci = model.loci
def test_one(data_in, data_out):
key = _make_cache_key(genome, data_in, data_out)
_cache_mutex.acquire()
try:
error = _cache[key]
except KeyError:
_cache_mutex.release()
model_out = process_data(data_in, genome, model, loci)
assert (np.all(model_out.index == data_out.index))
error = model.error_func(model_out.values, data_out.values)
_cache_mutex.acquire()
_cache[key] = error
_cache_mutex.release()
return error
def test_loop():
error = 0
#start = time.time()
train_iter = model.dataset.train_data_iterator()
for (data_in, data_out) in train_iter():
error += test_one(data_in, data_out)
#print "Testing completed in ", time.time() - start
#sys.stdout.flush()
return error
global _catch_exceptions_during_fitness_evaluation
if _catch_exceptions_during_fitness_evaluation:
try:
error = test_loop()
except Exception, e:
print >> sys.stderr, "Caught exception during fitness evaluation!"
print >> sys.stderr, " Offending genome was:", genome[:]
tb = " " + traceback.format_exc(limit=50)[:-1]
print >> sys.stderr, tb.replace("\n", "\n ")
print >> sys.stderr, " Setting fitness to 0 to avoid selection."
return 0
def _fitness(indiv):
"""Generic fitness function, calls preprocessors, transformer and
postprocessors for each model, calculates error from error function."""
global _model
model = _model
num_trials = indiv.getParam('num_trials')
genome = pu.raw_genes(indiv, True)
loci = model.loci
def test_one(data_in, data_out):
key = _make_cache_key(genome, data_in, data_out)
_cache_mutex.acquire()
try:
error = _cache[key]
except KeyError:
_cache_mutex.release()
model_out = process_data(data_in, genome, model, loci)
assert(np.all(model_out.index == data_out.index))
error = model.error_func(model_out.values, data_out.values)
_cache_mutex.acquire()
_cache[key] = error
_cache_mutex.release()
return error
def test_loop():
error = 0
#start = time.time()
train_iter = model.dataset.train_data_iterator()
for (data_in, data_out) in train_iter():
error += test_one(data_in, data_out)
#print "Testing completed in ", time.time() - start
#sys.stdout.flush()
return error
global _catch_exceptions_during_fitness_evaluation
if _catch_exceptions_during_fitness_evaluation:
try:
error = test_loop()
except Exception, e:
print >>sys.stderr, "Caught exception during fitness evaluation!"
print >>sys.stderr, " Offending genome was:", genome[:]
tb = " " + traceback.format_exc(limit=50)[:-1]
print >>sys.stderr, tb.replace("\n", "\n ")
print >>sys.stderr, " Setting fitness to 0 to avoid selection."
return 0
def _run_models(models, dataset):
data_desc = dataset.desc
for model in models:
if not is_mpi_slave(options):
print "Optimizing parameters for model", model.name, "."
run_GA(model, options)
if is_mpi_slave(options):
continue
raw_genes = pu.raw_genes(model.genome, strip=True)
print "Best genes found during evolution: ", raw_genes
target, predictions = test_genome(raw_genes, model)
_save_test_prediction(target, predictions, data_desc)
error = concat_and_calc_error(predictions, target, model.error_func)
print "Error on test phase for best genome found, " \
"%s, %i days: %5.4f" % (data_desc, len(predictions), error)
if not options.no_plot:
plot_test_prediction(target, predictions, model.name, data_desc,
model.error_func)
def _run_models(models, dataset):
data_desc = dataset.desc
for model in models:
if not is_mpi_slave(options):
print "Optimizing parameters for model", model.name, "."
run_GA(model, options)
if is_mpi_slave(options):
continue
raw_genes = pu.raw_genes(model.genome, strip=True)
print "Best genes found during evolution: ", raw_genes
target, predictions = test_genome(raw_genes, model)
_save_test_prediction(target, predictions, data_desc)
error = concat_and_calc_error(predictions, target, model.error_func)
print "Error on test phase for best genome found, " \
"%s, %i days: %5.4f" % (data_desc, len(predictions), error)
if not options.no_plot:
plot_test_prediction(target, predictions,
model.name, data_desc, model.error_func)
def print_population(ga_engine):
for indiv in ga_engine.getPopulation():
print "%9e, %9e" % (indiv.score, indiv.fitness), pu.raw_genes(
indiv, True)
def print_population(ga_engine):
for indiv in ga_engine.getPopulation():
print "%9e, %9e" % (indiv.score, indiv.fitness), pu.raw_genes(indiv, True)
