def next(self, state, data):
state, metrics = self._iterative_process.next(state, data)
state = self._server_state_from_tff_result(state)
metrics = metrics._asdict(recursive=True)
metrics.update(state.delta_aggregate_state)
outputs = None
return adapters.IterationResult(state, metrics, outputs)
def next(
self,
state: ServerState,
data: Collection[tf.data.Dataset],
) -> adapters.IterationResult:
state, metrics = self._iterative_process.next(state, data)
outputs = None
return adapters.IterationResult(state, metrics, outputs)
def next(self, state, data):
state, initial_metrics, metrics = self._iterative_process.next(
state, data)
total_metrics = {
'before_training': initial_metrics,
'during_training': metrics
}
outputs = None
return adapters.IterationResult(state, total_metrics, outputs)
def next(
self,
state: ServerState,
data: Collection[tf.data.Dataset],
) -> adapters.IterationResult:
state, metrics = self._iterative_process.next(state, data)
state = ServerState.from_tff_result(state)
metrics = metrics._asdict(recursive=True)
outputs = None
return adapters.IterationResult(state, metrics, outputs)
def next(self, state, data):
state, initial_metrics, metrics = self._iterative_process.next(
state, data)
state = ServerState.from_tff_result(state,
self._from_anon_client_callback,
self._from_anon_server_callback)
initial_metrics = initial_metrics._asdict(recursive=True)
metrics = metrics._asdict(recursive=True)
total_metrics = {
'before_training': initial_metrics,
'during_training': metrics
}
outputs = None
return adapters.IterationResult(state, total_metrics, outputs)
def next(self, state, data):
state, metrics = self._iterative_process.next(state, data)
if self._adaptive_clipping:
if self._per_vector_clipping:
metrics.update({('clip_' + str(i)):
self._get_clip(vector_state)
for i, vector_state in enumerate(
state.delta_aggregate_state)})
else:
metrics.update(
{'clip': self._get_clip(state.delta_aggregate_state)})
outputs = None
return adapters.IterationResult(state, metrics, outputs)
def next(self, state, data):
state, metrics = self._iterative_process.next(state, data)
python_state = self._server_state_from_tff_result(state)
metrics = metrics._asdict(recursive=True)
if self._adaptive_clipping:
if self._per_vector_clipping:
metrics.update({('clip_' + str(i)):
self._get_clip(vector_state)
for i, vector_state in enumerate(
state.delta_aggregate_state)})
else:
metrics.update(
{'clip': self._get_clip(state.delta_aggregate_state)})
outputs = None
return adapters.IterationResult(python_state, metrics, outputs)
def next(self, state, data):
state, metrics = self._iterative_process.next(state, data)
state = _from_tff_result(state)
metrics = metrics._asdict(recursive=True)
outputs = None
return adapters.IterationResult(state, metrics, outputs)
def next(self, state, data):
state, metrics = self._iterative_process.next(state, data)
outputs = None
return adapters.IterationResult(state, metrics, outputs)
