def __model_reload(sample):
# Get the estimator class
model_name = sample.__class__.__name__
model_name = model_name.replace('Model', '')
estimator_name = 'HowOldWebsite.estimators.estimator_{}.Estimator{}'.format(
model_name.lower(), model_name.capitalize())
estimator_obj = reflect_get_class(estimator_name)
# Reload
estimator_obj.estimator_load(force=True)
def __model_reload(sample):
# Get the estimator class
model_name = sample.__class__.__name__
model_name = model_name.replace('Model', '')
estimator_name = 'HowOldWebsite.estimators.estimator_{}.Estimator{}'.format(model_name.lower(),
model_name.capitalize())
estimator_obj = reflect_get_class(estimator_name)
# Reload
estimator_obj.estimator_load(force=True)
def __do_thread_train(cls, model_name, estimator, feature_jar, target):
print("{} Start".format(model_name.capitalize()))
try:
class_worker = 'HowOldWebsite.trainers.trainer_{}.Trainer{}'.format(
model_name, model_name.capitalize())
obj_worker = reflect_get_class(class_worker)
worker = obj_worker(estimator)
worker.train(feature_jar, target)
except Exception as e:
print(e)
pass
print("{} OK".format(model_name.capitalize()))
def model_delete(modeladmin, request, queryset):
# Get the estimator class
sample = queryset[0]
model_name = sample.__class__.__name__
model_name = model_name.replace('Model', '')
estimator_name = 'HowOldWebsite.estimators.estimator_{}.Estimator{}'.format(model_name.lower(),
model_name.capitalize())
estimator_obj = reflect_get_class(estimator_name)
# Delete from disk
for item in queryset:
estimator_obj.database_model_delete(item.id)
# Delete from database
queryset.delete()
# Reload
estimator_obj.estimator_load(force=True)
def model_delete(modeladmin, request, queryset):
# Get the estimator class
sample = queryset[0]
model_name = sample.__class__.__name__
model_name = model_name.replace('Model', '')
estimator_name = 'HowOldWebsite.estimators.estimator_{}.Estimator{}'.format(
model_name.lower(), model_name.capitalize())
estimator_obj = reflect_get_class(estimator_name)
# Delete from disk
for item in queryset:
estimator_obj.database_model_delete(item.id)
# Delete from database
queryset.delete()
# Reload
estimator_obj.estimator_load(force=True)
def get_benchmark(cls, model, model_name):
cls.load_std_data()
score = 0
try:
# Get class
class_name = 'HowOldWebsite.estimators.estimator_{}.Estimator{}'.format(model_name.lower(),
model_name.capitalize())
obj_class = reflect_get_class(class_name)
# Get score
features = obj_class.feature_combine(cls.feature_jar)
features = np.array(features)
score = model.score(features, cls.target_jar[model_name.lower()])
except Exception as e:
# print(e)
pass
return score
def __train_main(cls, model_names):
model_names = [m.lower() for m in model_names]
print("=" * 10 + " Train Start " + "=" * 10)
try:
faces = RecordFace.objects.filter(used_flag=1)
if not django.conf.settings.DEBUG:
if len(faces) < 100:
print("Error: The training set is too small.")
print("\t Skip the training!")
raise Exception()
image_jar = dict()
feature_jar = dict()
target_jar = dict()
estimator_jar = dict()
threads = list()
# Get estimator class
for m in model_names:
class_estimator = 'HowOldWebsite.estimators.estimator_{}.Estimator{}'.format(
m, m.capitalize())
estimator_jar[m] = reflect_get_class(class_estimator)
for face in faces:
face_id = face.id
# Get image
face_filename_color = os.path.join(
django.conf.settings.SAVE_DIR['FACE'],
str(face_id) + '.jpg')
# face_filename_gray = os.path.join(SAVE_DIR['FACE_GRAY'], str(face_id) + '.jpg')
cv_face_image = do_imread(face_filename_color)
cv_face_gray = do_rgb2gray(cv_face_image)
if 'rgb' not in image_jar.keys():
image_jar['rgb'] = list()
image_jar['rgb'].append(cv_face_image)
if 'gray' not in image_jar.keys():
image_jar['gray'] = list()
image_jar['gray'].append(cv_face_gray)
# Get target
if 'sex' not in target_jar.keys():
target_jar['sex'] = list()
target_jar['sex'].append(
(face.recordsex_set.first()).value_user)
if 'age' not in target_jar.keys():
target_jar['age'] = list()
target_jar['age'].append(
(face.recordage_set.first()).value_user)
if 'smile' not in target_jar.keys():
target_jar['smile'] = list()
target_jar['smile'].append(
(face.recordsmile_set.first()).value_user)
# Extract features
for m in model_names:
feature_jar = estimator_jar[m].feature_extract(
feature_jar, image_jar)
# Train
for m in model_names:
th = threading.Thread(target=cls.__do_thread_train,
args=(m, estimator_jar[m], feature_jar,
target_jar[m]))
threads.append(th)
th.start()
for item in threads:
item.join()
# Change the used flag
if not django.conf.settings.DEBUG:
faces.update(used_flag=2)
except Exception as e:
# print(e)
print("Error occurred while training")
pass
print("=" * 10 + " Train Finish " + "=" * 10)
# Set the busy flag
UtilTrainer.__busy = False
def extract_all(cls, feature_name, pic_jar, params=None):
object_class_name = cls.__make_full_class_name(feature_name=feature_name)
object_class = reflect_get_class(object_class_name)
features = object_class.extract_all(picture_jar=pic_jar, params=params)
return features