def mli_accept_weights(self, weights: Weights):
try:
request_iterator = weights_to_iterator(weights, encode=False)
self.stub.SetWeights(request_iterator)
except grpc.RpcError as e:
_logger.exception(f"Failed to call SetWeights: {e}")
raise ConnectionError(f"GRPC error: {e}")
def test_iterator_and_back():
part_a = bytes(b"a" * WEIGHTS_PART_SIZE_BYTES)
part_b = bytes(b"b" * (WEIGHTS_PART_SIZE_BYTES - 2))
test_weights = part_a + part_b
weights = Weights(weights=test_weights)
iterator = weights_to_iterator(input_weights=weights, encode=False)
result = iterator_to_weights(request_iterator=iterator, decode=False)
assert result == weights
def test_weights_to_iterator_small_limit():
part_a = bytes(b"a" * WEIGHTS_PART_SIZE_BYTES)
test_weights = part_a
weights = Weights(weights=test_weights)
iterator = weights_to_iterator(input_weights=weights, encode=False)
val = next(iterator, b"")
assert isinstance(val, WeightsPart)
assert val.total_bytes == WEIGHTS_PART_SIZE_BYTES
assert val.byte_index == 0
assert bytes(val.weights) == part_a
val = next(iterator, b"")
assert val == b""
def test_weights_to_iterator_small():
part_a = bytes(b"a")
test_weights = part_a
weights = Weights(weights=test_weights)
iterator = weights_to_iterator(input_weights=weights, encode=False)
val = next(iterator, b"")
assert isinstance(val, WeightsPart)
assert val.total_bytes == 1
assert val.byte_index == 0
assert bytes(val.weights) == part_a
val = next(iterator, b"")
assert val == b""
def mli_test_weights(self, weights: Weights = None) -> ProposedWeights:
try:
if weights:
response = self.stub.TestWeights(
weights_to_iterator(weights, encode=False))
else:
raise Exception(
"mli_test_weights(None) is not currently supported")
return ProposedWeights(weights=weights,
vote_score=response.vote_score,
test_score=response.test_score,
vote=response.vote)
except grpc.RpcError as ex:
_logger.exception(f"Failed to test_model: {ex}")
raise ConnectionError(f"GRPC error: {ex}")
def ProposeWeights(self, request, context):
_count_propose.inc()
if not self._check_model(context):
return
self._learner_mutex.acquire()
try:
_logger.debug("Start training...")
weights = self.learner.mli_propose_weights()
_logger.debug("Training done!")
weights_part_iterator = weights_to_iterator(weights)
for wp in weights_part_iterator:
yield wp
except Exception as ex: # pylint: disable=W0703
_logger.exception(f"Exception in ProposeWeights: {ex} {type(ex)}")
context.set_code(grpc.StatusCode.INTERNAL)
context.set_details(str(ex))
_count_propose_err.inc()
finally:
self._learner_mutex.release()
