def __init__(self, concurrency, output, num_categories, stats_interval,
stats_type, conf_thresh):
self._stats = CategoricalStatistics(stats_interval, stats_type,
num_categories, conf_thresh)
self._concurrency = concurrency
self._active = 0
self._condition = threading.Condition()
self._output = output
class SavedModel(Model):
def __init__(self, output_file, model_dir, signature_def, stats_interval,
stats_type, conf_thresh, conf_percent):
super(SavedModel, self).__init__(model_dir, signature_def)
self._output_file = output_file
# loads the metagraphdef(s) into the provided session
# restores variables, gets assets, initializes the assets into the main function
self._sess = tf.Session()
# For now, we only set the default tf_serving tag_set
tag_set = "serve"
tf.saved_model.loader.load(self._sess, [tag_set], self._model_dir)
graph = tf.get_default_graph()
self._input_node = graph.get_tensor_by_name(self.get_input_name())
self._model = graph.get_tensor_by_name(self.get_output_name())
num_categories = self.get_output_shape()
self._stats = CategoricalStatistics(stats_interval, stats_type,
num_categories, conf_thresh,
conf_percent)
def infer(self, sample, label):
inference = self._sess.run(self._model, {self._input_node: [sample]})
prediction = self._stats.infer_stats(sample, label, inference)
self._output_file.write("{}\n".format(prediction))
def __del__(self):
self._sess.close()
del self._stats
super(SavedModel, self).__del__()
class _CallbackHandler(object):
"""Maintains concurrency level and records results."""
def __init__(self, concurrency, output, num_categories, stats_interval,
stats_type, conf_thresh):
self._stats = CategoricalStatistics(stats_interval, stats_type,
num_categories, conf_thresh)
self._concurrency = concurrency
self._active = 0
self._condition = threading.Condition()
self._output = output
def record_stats(self, sample, label, inference):
prediction = self._stats.infer_stats(sample, label, inference)
self._output_file.write("{}\n".format(prediction))
def dec_active(self):
with self._condition:
self._active -= 1
self._condition.notify()
def throttle(self):
with self._condition:
while self._active == self._concurrency:
self._condition.wait()
self._active += 1
def __init__(self, output_file, model_dir, signature_def, stats_interval,
stats_type, conf_thresh):
super(SavedModel, self).__init__(model_dir, signature_def)
self._output_file = output_file
# loads the metagraphdef(s) into the provided session
# restores variables, gets assets, initializes the assets into the main function
self._sess = tf.Session()
# For now, we only set the default tf_serving tag_set
tag_set = "serve"
tf.saved_model.loader.load(self._sess, [tag_set], self._model_dir)
graph = tf.get_default_graph()
self._input_node = graph.get_tensor_by_name(self.get_input_name())
self._model = graph.get_tensor_by_name(self.get_output_name())
num_categories = self.get_output_shape()
self._stats = CategoricalStatistics(stats_interval, stats_type,
num_categories, conf_thresh)