def _build_eval_spec(self):
invoke_args = self._resolve_input_fn_args(
self.customer_script.eval_input_fn)
eval_input_fn = lambda: self.customer_script.eval_input_fn(**
invoke_args)
if self.saves_training():
serving_input_receiver_fn = lambda: self.customer_script.serving_input_fn(
self.customer_params)
exporter = tf.estimator.LatestExporter(
'Servo', serving_input_receiver_fn=serving_input_receiver_fn)
else:
logger.warn(
'serving_input_fn not specified, model NOT saved, use checkpoints to reconstruct'
)
exporter = None
valid_eval_keys = ['start_delay_secs', 'throttle_secs']
eval_params = {
k: v
for k, v in self.customer_params.items() if k in valid_eval_keys
}
return tf.estimator.EvalSpec(eval_input_fn,
steps=self.eval_steps,
exporters=exporter,
**eval_params)
def _build_eval_spec(self):
# TODO: why is this different from the train_input_fn? investigate
input_dir = self.input_channels.get(self.DEFAULT_TRAINING_CHANNEL,
None)
params = self.customer_params
eval_input_fn = lambda: self.customer_script.eval_input_fn(
input_dir, params)
if self.saves_training():
serving_input_receiver_fn = lambda: self.customer_script.serving_input_fn(
params)
exporter = tf.estimator.LatestExporter(
'Servo', serving_input_receiver_fn=serving_input_receiver_fn)
else:
logger.warn(
'serving_input_fn not specified, model NOT saved, use checkpoints to reconstruct'
)
exporter = None
valid_eval_keys = ['start_delay_secs', 'throttle_secs']
eval_params = {
k: v
for k, v in self.customer_params.items() if k in valid_eval_keys
}
return tf.estimator.EvalSpec(eval_input_fn,
steps=self.eval_steps,
exporters=exporter,
**eval_params)
def _export_strategy():
if self.saves_training():
return [saved_model_export_utils.make_export_strategy(
serving_input_fn=_serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1)]
logger.warn("serving_input_fn not specified, model NOT saved, use checkpoints to reconstruct")
return None
def _export_strategy():
if self.saves_training():
return [saved_model_export_utils.make_export_strategy(
serving_input_fn=_serving_input_fn,
default_output_alternative_key=None,
exports_to_keep=1)]
logger.warn("serving_input_fn not specified, model NOT saved, use checkpoints to reconstruct")
return None
def _build_eval_spec(self):
invoke_args = self._resolve_input_fn_args(self.customer_script.eval_input_fn)
eval_input_fn = lambda: self.customer_script.eval_input_fn(**invoke_args)
if self.saves_training():
serving_input_receiver_fn = lambda: self.customer_script.serving_input_fn(self.customer_params)
exporter = tf.estimator.LatestExporter('Servo',
serving_input_receiver_fn=serving_input_receiver_fn)
else:
logger.warn('serving_input_fn not specified, model NOT saved, use checkpoints to reconstruct')
exporter = None
valid_eval_keys = ['start_delay_secs', 'throttle_secs']
eval_params = {k: v for k, v in self.customer_params.items() if k in valid_eval_keys}
return tf.estimator.EvalSpec(eval_input_fn, steps=self.eval_steps, exporters=exporter, **eval_params)