def norm_stdev(self):
if self.color_mode == "rgb" or self.color_mode == "hsv":
stdev_chan1 = assign_fallback(
self.cp[self.SECTION].getfloat("normalize_stdev_chan1"),
self._normalize_stdev[0])
stdev_chan2 = assign_fallback(
self.cp[self.SECTION].getfloat("normalize_stdev_chan2"),
self._normalize_stdev[1])
stdev_chan3 = assign_fallback(
self.cp[self.SECTION].getfloat("normalize_stdev_chan3"),
self._normalize_stdev[2])
return np.array([stdev_chan1, stdev_chan2, stdev_chan3])
elif self.color_mode == "grayscale":
grayscale_stdev = np.mean(self._normalize_stdev)
return assign_fallback(
self.cp[self.SECTION].getfloat("normalize_stdev"),
grayscale_stdev)
def norm_mean(self):
if self.color_mode == "rgb" or self.color_mode == "hsv":
mean_chan1 = assign_fallback(
self.cp[self.SECTION].getfloat("normalize_mean_chan1"),
self._normalize_mean[0])
mean_chan2 = assign_fallback(
self.cp[self.SECTION].getfloat("normalize_mean_chan2"),
self._normalize_mean[1])
mean_chan3 = assign_fallback(
self.cp[self.SECTION].getfloat("normalize_mean_chan3"),
self._normalize_mean[2])
return np.array([mean_chan1, mean_chan2, mean_chan3])
elif self.color_mode == "grayscale":
grayscale_mean = np.mean(self._normalize_mean)
return assign_fallback(
self.cp[self.SECTION].getfloat("normalize_mean"),
grayscale_mean)
def __init__(self, cp):
super().__init__(cp=cp)
self._initial_learning_rate = assign_fallback(
self.cp["TRAIN"].getfloat("initial_learning_rate"),
self._initial_learning_rate)
config_train_steps = self.cp["TRAIN"].get("train_steps")
if config_train_steps != "auto" and config_train_steps is not None:
try:
config_train_steps = int(config_train_steps)
except ValueError:
raise ValueError(
"** train_steps: {} is invalid,please use 'auto' or integer."
.format(config_train_steps))
else:
config_train_steps = None
config_validation_steps = assign_fallback(
self.cp["TRAIN"].get("validation_steps"), self._validation_steps)
if config_validation_steps != "auto" and config_validation_steps is not None:
try:
config_validation_steps = int(config_validation_steps)
except ValueError:
raise ValueError(
"** validation_steps: {} is invalid,please use 'auto' or integer."
.format(config_validation_steps))
else:
config_validation_steps = None
config_test_steps = assign_fallback(self.cp["TEST"].get("steps"),
self._test_steps)
if config_test_steps != "auto" and config_test_steps is not None:
try:
config_test_steps = int(config_test_steps)
except ValueError:
raise ValueError(
"** test_steps: {} is invalid,please use 'auto' or integer."
.format(config_test_steps))
else:
config_test_steps = None
self._train_steps = assign_fallback(config_train_steps,
self._train_steps)
self._validation_steps = assign_fallback(config_validation_steps,
self._validation_steps)
self._test_steps = assign_fallback(config_test_steps, self._test_steps)
def gpu(self):
return assign_fallback(self.cp["TRAIN"].getint("gpu"), self._gpu)
def verbosity(self):
return assign_fallback(self.cp["DEFAULT"].getint("verbosity"),
self._verbosity)
def model_name(self):
return assign_fallback(self.cp[self.SECTION].get("model_name"), self._model_name)
def random_state(self):
return assign_fallback(
self.cp["DATASET"].getint("split_dataset_random_state"),
self._random_state)
def dev_augmentation(self):
return assign_fallback(
self.cp[self.SECTION].getboolean("dev_augmentation"),
self._dev_augmentation)
def random_vert_flip(self):
return assign_fallback(
self.cp[self.SECTION].getboolean("random_vert_flip"),
self._random_vert_flip)
def histogram_freq(self):
return assign_fallback(self.cp["TRAIN"].getint("histogram_freq"),
self._histogram_freq)
def scale(self):
return assign_fallback(self.cp["IMAGE"].getfloat("scale"), self._scale)
def progress_test_verbosity(self):
return assign_fallback(self.cp["TEST"].getint("progress_verbosity"),
self._progress_verbosity)
def enable_grad_cam(self):
return assign_fallback(self.cp["TEST"].getboolean("enable_grad_cam"),
self._enable_grad_cam)
def progress_train_verbosity(self):
return assign_fallback(self.cp["TRAIN"].getint("progress_verbosity"),
self._progress_verbosity)
def color_mode(self):
return assign_fallback(self.cp["IMAGE"].get("color_mode"),
self._color_mode)
def final_activation(self):
return assign_fallback(self.cp["DATASET"].get("final_activation"),
self._final_activation)
def epochs(self):
return assign_fallback(self.cp["TRAIN"].getint("epochs"), self._epochs)
def write_graph(self):
return assign_fallback(self.cp["TRAIN"].getboolean("write_graph"),
self._write_graph)
def img_dim(self):
return assign_fallback(self.cp["IMAGE"].getint("image_dimension"),
self._img_dim)
def write_images(self):
return assign_fallback(self.cp["TRAIN"].getboolean("write_images"),
self._write_images)
def train_augmentation(self):
return assign_fallback(
self.cp[self.SECTION].getboolean("train_augmentation"),
self._train_augmentation)
def embeddings_freq(self):
return assign_fallback(self.cp["TRAIN"].getint("embeddings_freq"),
self._embeddings_freq)
def flip_prob(self):
return assign_fallback(self.cp[self.SECTION].getfloat("flip_prob"),
self._flip_prob)
def use_best_weights(self):
return assign_fallback(self.cp[self.SECTION].getboolean("use_best_weights"), self._use_trained_model_weights)
def __init__(self, cp):
super().__init__(cp)
self._use_trained_model_weights = assign_fallback(self.cp[self.SECTION].getboolean("use_trained_model_weights"),
self._use_trained_model_weights)
def use_class_balancing(self):
return assign_fallback(
self.cp["DATASET"].getboolean("use_class_balancing"),
self._use_class_balancing)
def use_ext_base_model_weights(self):
return assign_fallback(self.cp[self.SECTION].getboolean("use_ext_base_model_weights"),
self._use_ext_base_model_weights)
def use_default_split(self):
return assign_fallback(
self.cp["DATASET"].getboolean("use_default_split"),
self._use_default_split)
def show_model_summary(self):
return assign_fallback(self.cp[self.SECTION].getboolean("show_model_summary"), self._show_model_summary)
def patience_reduce_lr(self):
return assign_fallback(self.cp["TRAIN"].getint("patience_reduce_lr"),
self._patience_reduce_lr)
