def __call__(self, *inputs, **kwargs):
if (len(inputs) > self._schema.MaxNumInput()
or len(inputs) < self._schema.MinNumInput()):
raise ValueError("Operator {} expects [{}, " +
"{}] inputs, but received {}".format(
type(self).__name__, self._schema.MinNumInput(
), self._schema.MaxNumInput(), len(inputs)))
op_instance = _OperatorInstance(inputs, self, **kwargs)
outputs = {}
feature_names = []
features = []
for i, (feature_name, feature) in enumerate(self._features.items()):
t_name = "_TFRecordReader" + "_id_" + str(
op_instance.id) + "_output_" + str(i)
t = TensorReference(t_name, self._device, op_instance)
op_instance.spec.AddOutput(t.name, t.device)
op_instance.append_output(t)
outputs[feature_name] = t
feature_names.append(feature_name)
features.append(feature)
op_instance.spec.AddArg("feature_names", feature_names)
op_instance.spec.AddArg("features", features)
return outputs
def generate_outputs(self):
# Add outputs
if self._op.device == "gpu" or self._op.device == "mixed":
output_device = "gpu"
else:
output_device = "cpu"
num_output = self._op.schema.CalculateOutputs(self._spec) + self._op.schema.CalculateAdditionalOutputs(self._spec)
for i in range(num_output):
t_name = type(self._op).__name__ + "_id_" + str(self.id) + "_output_" + str(i)
t = TensorReference(t_name, output_device, self)
self._spec.AddOutput(t.name, t.device)
self.append_output(t)
