def get_final_paretos(self): # pull all non-expanded models in queue out and push into final # could also use spea2_list here, they should have same contents final = self.final + self.nsga2_list # generate final pareto fronts self.fitness_calc(final) final_list = selection.sortLogNondominated(final, len(final)) return final_list
def heap_select(self, heap_list, pop_count, fitness_calc):
fitness_calc(heap_list)
if self.use_nsga2:
selected = selection.selNSGA2(heap_list, pop_count, nd='log')
else:
selected = selection.sortLogNondominated(heap_list, pop_count, True)
selected = [ s for slist in selected for s in slist]
# print("selected", selected)
return selected
def heap_pop(self, heap_list, pop_count, fitness_calc):
fitness_calc(heap_list)
if self.use_nsga2:
popped = selection.selNSGA2(heap_list, pop_count, nd='log')
else:
popped = selection.sortLogNondominated(heap_list, pop_count, True)
popped = [ p for plist in popped for p in plist]
# print("popped", popped)
heap_list = [m for m in heap_list if not m in popped]
if self.max_heap_size > 0 and len(heap_list) > self.max_heap_size:
heap_list = heap_list[:self.max_heap_size]
return popped, heap_list
def print_best(self, count, file=sys.stdout):
print ("Best " + str(count) + " of " + str(len(self.final)), file=file)
# self.fitness_calc(self.final)
models = self.heap_select(self.final, count, self.fitness_calc)
best = selection.sortLogNondominated(models, count)
print (best[0][0].print_long_columns(), file=file)
print ("-----------------------------------------------------------------------------------", file=file)
cnt = 0
self.best_size = 0
self.best_err = 10000000000
self.best_r2 = 0
self.best_vari = 0
self.ave_size = 0
self.ave_err = 0
self.ave_r2 = 0
self.ave_vari = 0
for front in best:
for m in front:
if cnt >= count:
break
cnt += 1
print (" ", m.print_long(), file=file)
if m.score < self.best_err:
self.best_size = m.size()
self.best_err = m.score
self.best_r2 = m.r2
self.best_vari = m.evar
self.ave_size += m.size()
self.ave_err += m.score
self.ave_r2 += m.r2
self.ave_vari += m.evar
print ("", file=file)
if cnt >= count:
break
self.ave_size /= float(count)
self.ave_err /= float(count)
self.ave_r2 /= float(count)
self.ave_vari /= float(count)
print ("-----------------------------------------------------------------------------------", file=file)
def finalize(self, nfronts=4):
# finalization
print ("\n\nFinalizing\n\n")
# pull all non-expanded models in queue out and push into final
# could also use spea2_list here, they should have same contents
final = self.final + self.nsga2_list
# generate final pareto fronts
self.fitness_calc(final)
final_list = selection.sortLogNondominated(final, len(final))
self.print_final_models(final_list, 128, self.logs["final"])
# print first 4 pareto fronts
print ("Final Results")
print (final_list[0][0].print_columns())
print ("-----------------------------------------------------------------------------------")
for front in final_list[:nfronts]:
for m in front:
print (" ", m)
print ("")
print ("\n")
print ("num peekd models: ", self.peekd_models)
print ("num evald models: ", self.evald_models)
print ("num peek evals: ", self.peek_nfev, self.peek_nfev * self.peek_npts)
print ("num eval evals: ", self.eval_nfev, self.eval_nfev * self.eval_npts)
print ("num total evals: ", self.peek_nfev * self.peek_npts + self.eval_nfev * self.eval_npts)
# print ("\n\n", nx.info(self.GRAPH), "\n\n")
# handle issue with extra stray node with parent_id == -2 (at end of nodes list)
# del_n = []
# for n in nx.nodes_iter(self.GRAPH):
# modl = self.memoizer.get_by_id(n)
# if modl.score is None:
# del_n.append(n)
# for n in del_n:
# self.GRAPH.remove_node(n)
if self.workers > 0:
print ("\n\nstopping workers")
for proc in self.peek_procs:
self.peek_in_queue.put(None)
for proc in self.eval_procs:
self.eval_in_queue.put(None)
for proc in self.alge_procs:
self.alge_in_queue.put(None)
for proc in self.expd_procs:
self.expd_in_queue.put(None)
for proc in self.peek_procs:
proc.join()
for proc in self.eval_procs:
proc.join()
for proc in self.alge_procs:
proc.join()
for proc in self.expd_procs:
proc.join()
if self.remote_eval == True:
self.ws.close()
print ("\n\ndone\n\n")
if self.log_details:
for name, file_obj in self.logs.items():
if file_obj is not sys.stdout:
file_obj.close()
