# Copyright 2020 Flower Labs GmbH. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Training history."""
import pprint
from functools import reduce
from typing import Dict, List, Tuple
from flwr.common.typing import Scalar
[docs]class History:
"""History class for training and/or evaluation metrics collection."""
def __init__(self) -> None:
self.losses_distributed: List[Tuple[int, float]] = []
self.losses_centralized: List[Tuple[int, float]] = []
self.metrics_distributed_fit: Dict[str, List[Tuple[int, Scalar]]] = {}
self.metrics_distributed: Dict[str, List[Tuple[int, Scalar]]] = {}
self.metrics_centralized: Dict[str, List[Tuple[int, Scalar]]] = {}
[docs] def add_loss_distributed(self, server_round: int, loss: float) -> None:
"""Add one loss entry (from distributed evaluation)."""
self.losses_distributed.append((server_round, loss))
[docs] def add_loss_centralized(self, server_round: int, loss: float) -> None:
"""Add one loss entry (from centralized evaluation)."""
self.losses_centralized.append((server_round, loss))
[docs] def add_metrics_distributed_fit(
self, server_round: int, metrics: Dict[str, Scalar]
) -> None:
"""Add metrics entries (from distributed fit)."""
for key in metrics:
# if not (isinstance(metrics[key], float) or isinstance(metrics[key], int)):
# continue # ignore non-numeric key/value pairs
if key not in self.metrics_distributed_fit:
self.metrics_distributed_fit[key] = []
self.metrics_distributed_fit[key].append((server_round, metrics[key]))
[docs] def add_metrics_distributed(
self, server_round: int, metrics: Dict[str, Scalar]
) -> None:
"""Add metrics entries (from distributed evaluation)."""
for key in metrics:
# if not (isinstance(metrics[key], float) or isinstance(metrics[key], int)):
# continue # ignore non-numeric key/value pairs
if key not in self.metrics_distributed:
self.metrics_distributed[key] = []
self.metrics_distributed[key].append((server_round, metrics[key]))
[docs] def add_metrics_centralized(
self, server_round: int, metrics: Dict[str, Scalar]
) -> None:
"""Add metrics entries (from centralized evaluation)."""
for key in metrics:
# if not (isinstance(metrics[key], float) or isinstance(metrics[key], int)):
# continue # ignore non-numeric key/value pairs
if key not in self.metrics_centralized:
self.metrics_centralized[key] = []
self.metrics_centralized[key].append((server_round, metrics[key]))
def __repr__(self) -> str:
"""Create a representation of History.
The representation consists of the following data (for each round) if present:
* distributed loss.
* centralized loss.
* distributed training metrics.
* distributed evaluation metrics.
* centralized metrics.
Returns
-------
representation : str
The string representation of the history object.
"""
rep = ""
if self.losses_distributed:
rep += "History (loss, distributed):\n" + reduce(
lambda a, b: a + b,
[
f"\tround {server_round}: {loss}\n"
for server_round, loss in self.losses_distributed
],
)
if self.losses_centralized:
rep += "History (loss, centralized):\n" + reduce(
lambda a, b: a + b,
[
f"\tround {server_round}: {loss}\n"
for server_round, loss in self.losses_centralized
],
)
if self.metrics_distributed_fit:
rep += (
"History (metrics, distributed, fit):\n"
+ pprint.pformat(self.metrics_distributed_fit)
+ "\n"
)
if self.metrics_distributed:
rep += (
"History (metrics, distributed, evaluate):\n"
+ pprint.pformat(self.metrics_distributed)
+ "\n"
)
if self.metrics_centralized:
rep += "History (metrics, centralized):\n" + pprint.pformat(
self.metrics_centralized
)
return rep