def test_evaluate_fn_with_tuple_of_trainable_variables(self):
loss_builder = tf.keras.losses.MeanSquaredError
metrics_builder = lambda: [tf.keras.metrics.MeanSquaredError()]
def tff_model_fn():
return tff.learning.from_keras_model(
keras_model=model_builder(),
input_spec=get_input_spec(),
loss=loss_builder(),
metrics=metrics_builder())
iterative_process = tff.learning.build_federated_averaging_process(
tff_model_fn, client_optimizer_fn=tf.keras.optimizers.SGD)
state = iterative_process.initialize()
test_dataset = create_tf_dataset_for_client(1)
reference_model = tff.learning.ModelWeights(
trainable=tuple(state.model.trainable),
non_trainable=tuple(state.model.non_trainable))
evaluate_fn = training_utils.build_evaluate_fn(test_dataset, model_builder,
loss_builder,
metrics_builder)
test_metrics = evaluate_fn(reference_model)
self.assertIn('loss', test_metrics)
def run_federated(
iterative_process_builder: Callable[...,
tff.templates.IterativeProcess],
client_epochs_per_round: int,
client_batch_size: int,
clients_per_round: int,
schedule: Optional[str] = 'none',
beta: Optional[float] = 0.,
max_batches_per_client: Optional[int] = -1,
client_datasets_random_seed: Optional[int] = None,
crop_size: Optional[int] = 24,
total_rounds: Optional[int] = 1500,
experiment_name: Optional[str] = 'federated_cifar100',
root_output_dir: Optional[str] = '/tmp/fed_opt',
max_eval_batches: Optional[int] = None,
**kwargs):
"""Runs an iterative process on the CIFAR-100 classification task.
This method will load and pre-process dataset and construct a model used for
the task. It then uses `iterative_process_builder` to create an iterative
process that it applies to the task, using
`federated_research.utils.training_loop`.
We assume that the iterative process has the following functional type
signatures:
* `initialize`: `( -> S@SERVER)` where `S` represents the server state.
* `next`: `<S@SERVER, {B*}@CLIENTS> -> <S@SERVER, T@SERVER>` where `S`
represents the server state, `{B*}` represents the client datasets,
and `T` represents a python `Mapping` object.
Moreover, the server state must have an attribute `model` of type
`tff.learning.ModelWeights`.
Args:
iterative_process_builder: A function that accepts a no-arg `model_fn`, and
returns a `tff.templates.IterativeProcess`. The `model_fn` must return a
`tff.learning.Model`.
client_epochs_per_round: An integer representing the number of epochs of
training performed per client in each training round.
client_batch_size: An integer representing the batch size used on clients.
clients_per_round: An integer representing the number of clients
participating in each round.
max_batches_per_client: An optional int specifying the number of batches
taken by each client at each round. If `-1`, the entire client dataset is
used.
client_datasets_random_seed: An optional int used to seed which clients are
sampled at each round. If `None`, no seed is used.
crop_size: An optional integer representing the resulting size of input
images after preprocessing.
total_rounds: The number of federated training rounds.
experiment_name: The name of the experiment being run. This will be appended
to the `root_output_dir` for purposes of writing outputs.
root_output_dir: The name of the root output directory for writing
experiment outputs.
max_eval_batches: If set to a positive integer, evaluation datasets are
capped to at most that many batches. If set to None or a nonpositive
integer, the full evaluation datasets are used.
**kwargs: Additional arguments configuring the training loop. For details
on supported arguments, see
`federated_research/utils/training_utils.py`.
"""
crop_shape = (crop_size, crop_size, 3)
cifar_train, _, fed_test_data = cifar100_dataset.get_federated_cifar100(
client_epochs_per_round=client_epochs_per_round,
train_batch_size=client_batch_size,
crop_shape=crop_shape,
max_batches_per_client=max_batches_per_client)
_, cifar_test = cifar100_dataset.get_centralized_datasets(
train_batch_size=client_batch_size,
max_test_batches=max_eval_batches,
crop_shape=crop_shape)
input_spec = cifar_train.create_tf_dataset_for_client(
cifar_train.client_ids[0]).element_spec
model_builder = functools.partial(resnet_models.create_resnet18,
input_shape=crop_shape,
num_classes=NUM_CLASSES)
loss_builder = tf.keras.losses.SparseCategoricalCrossentropy
metrics_builder = lambda: [tf.keras.metrics.SparseCategoricalAccuracy()]
def tff_model_fn() -> tff.learning.Model:
return tff.learning.from_keras_model(keras_model=model_builder(),
input_spec=input_spec,
loss=loss_builder(),
metrics=metrics_builder())
training_process = iterative_process_builder(tff_model_fn)
evaluate_fn = training_utils.build_evaluate_fn(
eval_dataset=cifar_test,
model_builder=model_builder,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
test_fn = training_utils.build_unweighted_test_fn(
federated_eval_dataset=fed_test_data,
model_builder=model_builder,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
logging.info('Training model:')
logging.info(model_builder().summary())
try:
var = kwargs['hparam_dict']['var_q_clients']
q_client = np.load(
f'/home/monica/AVAIL_VECTORS/q_client_{var}_cifar.npy')
except:
logging.info(
'Could not load q_client - initializing random availabilities')
q_client = None
if schedule == 'none':
client_datasets_fn = training_utils.build_client_datasets_fn(
train_dataset=cifar_train,
train_clients_per_round=clients_per_round,
random_seed=client_datasets_random_seed,
var_q_clients=kwargs['hparam_dict']['var_q_clients'],
f_mult=kwargs['hparam_dict']['f_mult'],
f_intercept=kwargs['hparam_dict']['f_intercept'],
sine_wave=kwargs['hparam_dict']['sine_wave'],
use_p=True,
q_client=q_client,
)
training_loop.run(iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=test_fn,
total_rounds=total_rounds,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
elif schedule == 'loss':
if 'loss_pool_size' in kwargs['hparam_dict'] and kwargs['hparam_dict'][
'loss_pool_size'] is not None:
loss_pool_size = kwargs['hparam_dict']['loss_pool_size']
logging.info(f'Loss pool size: {loss_pool_size}')
client_datasets_fn = training_utils.build_client_datasets_fn(
train_dataset=cifar_train,
train_clients_per_round=loss_pool_size,
random_seed=client_datasets_random_seed,
var_q_clients=kwargs['hparam_dict']['var_q_clients'],
f_mult=kwargs['hparam_dict']['f_mult'],
f_intercept=kwargs['hparam_dict']['f_intercept'],
sine_wave=kwargs['hparam_dict']['sine_wave'],
use_p=True,
q_client=q_client,
)
training_loop_loss.run(iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=test_fn,
total_rounds=total_rounds,
total_clients=loss_pool_size,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
else:
raise ValueError('Loss pool size not specified')
else:
client_datasets_fn = training_utils.build_availability_client_datasets_fn(
train_dataset=cifar_train,
train_clients_per_round=clients_per_round,
random_seed=client_datasets_random_seed,
beta=beta,
var_q_clients=kwargs['hparam_dict']['var_q_clients'],
f_mult=kwargs['hparam_dict']['f_mult'],
f_intercept=kwargs['hparam_dict']['f_intercept'],
sine_wave=kwargs['hparam_dict']['sine_wave'],
q_client=q_client,
)
training_loop_importance.run(iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=test_fn,
total_rounds=total_rounds,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
def run_federated(
iterative_process_builder: Callable[...,
tff.templates.IterativeProcess],
client_epochs_per_round: int,
client_batch_size: int,
clients_per_round: int,
max_batches_per_client: Optional[int] = -1,
client_datasets_random_seed: Optional[int] = None,
total_rounds: Optional[int] = 1500,
experiment_name: Optional[str] = 'federated_emnist_ae',
root_output_dir: Optional[str] = '/tmp/fed_opt',
max_eval_batches: Optional[int] = None,
**kwargs):
"""Runs an iterative process on the EMNIST autoencoder task.
This method will load and pre-process dataset and construct a model used for
the task. It then uses `iterative_process_builder` to create an iterative
process that it applies to the task, using
`federated_research.utils.training_loop`.
We assume that the iterative process has the following functional type
signatures:
* `initialize`: `( -> S@SERVER)` where `S` represents the server state.
* `next`: `<S@SERVER, {B*}@CLIENTS> -> <S@SERVER, T@SERVER>` where `S`
represents the server state, `{B*}` represents the client datasets,
and `T` represents a python `Mapping` object.
Moreover, the server state must have an attribute `model` of type
`tff.learning.ModelWeights`.
Args:
iterative_process_builder: A function that accepts a no-arg `model_fn`, and
returns a `tff.templates.IterativeProcess`. The `model_fn` must return a
`tff.learning.Model`.
client_epochs_per_round: An integer representing the number of epochs of
training performed per client in each training round.
client_batch_size: An integer representing the batch size used on clients.
clients_per_round: An integer representing the number of clients
participating in each round.
max_batches_per_client: An optional int specifying the number of batches
taken by each client at each round. If `-1`, the entire client dataset is
used.
client_datasets_random_seed: An optional int used to seed which clients are
sampled at each round. If `None`, no seed is used.
total_rounds: The number of federated training rounds.
experiment_name: The name of the experiment being run. This will be appended
to the `root_output_dir` for purposes of writing outputs.
root_output_dir: The name of the root output directory for writing
experiment outputs.
max_eval_batches: If set to a positive integer, evaluation datasets are
capped to at most that many batches. If set to None or a nonpositive
integer, the full evaluation datasets are used.
**kwargs: Additional arguments configuring the training loop. For details
on supported arguments, see
`federated_research/utils/training_utils.py`.
"""
emnist_train, _ = emnist_ae_dataset.get_emnist_datasets(
client_batch_size=client_batch_size,
client_epochs_per_round=client_epochs_per_round,
max_batches_per_client=max_batches_per_client,
only_digits=False)
_, emnist_test = emnist_ae_dataset.get_centralized_datasets(
train_batch_size=client_batch_size,
max_test_batches=max_eval_batches,
only_digits=False)
input_spec = emnist_train.create_tf_dataset_for_client(
emnist_train.client_ids[0]).element_spec
model_builder = emnist_ae_models.create_autoencoder_model
loss_builder = functools.partial(tf.keras.losses.MeanSquaredError,
reduction=tf.keras.losses.Reduction.SUM)
metrics_builder = lambda: [tf.keras.metrics.MeanSquaredError()]
def tff_model_fn() -> tff.learning.Model:
return tff.learning.from_keras_model(keras_model=model_builder(),
input_spec=input_spec,
loss=loss_builder(),
metrics=metrics_builder())
training_process = iterative_process_builder(tff_model_fn)
client_datasets_fn = training_utils.build_client_datasets_fn(
train_dataset=emnist_train,
train_clients_per_round=clients_per_round,
random_seed=client_datasets_random_seed)
evaluate_fn = training_utils.build_evaluate_fn(
eval_dataset=emnist_test,
model_builder=model_builder,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
logging.info('Training model:')
logging.info(model_builder().summary())
training_loop.run(iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=evaluate_fn,
total_rounds=total_rounds,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
def main(argv):
if len(argv) > 1:
raise app.UsageError('Expected no command-line arguments, '
'got: {}'.format(argv))
tff.backends.native.set_local_execution_context(max_fanout=10)
model_builder = functools.partial(
stackoverflow_models.create_recurrent_model,
vocab_size=FLAGS.vocab_size,
embedding_size=FLAGS.embedding_size,
latent_size=FLAGS.latent_size,
num_layers=FLAGS.num_layers,
shared_embedding=FLAGS.shared_embedding)
loss_builder = functools.partial(
tf.keras.losses.SparseCategoricalCrossentropy, from_logits=True)
special_tokens = stackoverflow_dataset.get_special_tokens(FLAGS.vocab_size)
pad_token = special_tokens.pad
oov_tokens = special_tokens.oov
eos_token = special_tokens.eos
def metrics_builder():
return [
keras_metrics.MaskedCategoricalAccuracy(name='accuracy_with_oov',
masked_tokens=[pad_token]),
keras_metrics.MaskedCategoricalAccuracy(name='accuracy_no_oov',
masked_tokens=[pad_token] +
oov_tokens),
# Notice BOS never appears in ground truth.
keras_metrics.MaskedCategoricalAccuracy(
name='accuracy_no_oov_or_eos',
masked_tokens=[pad_token, eos_token] + oov_tokens),
keras_metrics.NumBatchesCounter(),
keras_metrics.NumTokensCounter(masked_tokens=[pad_token]),
]
datasets = stackoverflow_dataset.construct_word_level_datasets(
FLAGS.vocab_size, FLAGS.client_batch_size,
FLAGS.client_epochs_per_round, FLAGS.sequence_length,
FLAGS.max_elements_per_user, FLAGS.num_validation_examples)
train_dataset, validation_dataset, test_dataset = datasets
if FLAGS.uniform_weighting:
def client_weight_fn(local_outputs):
del local_outputs
return 1.0
else:
def client_weight_fn(local_outputs):
return tf.cast(tf.squeeze(local_outputs['num_tokens']), tf.float32)
def model_fn():
return tff.learning.from_keras_model(
model_builder(),
loss_builder(),
input_spec=validation_dataset.element_spec,
metrics=metrics_builder())
if FLAGS.noise_multiplier is not None:
if not FLAGS.uniform_weighting:
raise ValueError(
'Differential privacy is only implemented for uniform weighting.'
)
dp_query = tff.utils.build_dp_query(
clip=FLAGS.clip,
noise_multiplier=FLAGS.noise_multiplier,
expected_total_weight=FLAGS.clients_per_round,
adaptive_clip_learning_rate=FLAGS.adaptive_clip_learning_rate,
target_unclipped_quantile=FLAGS.target_unclipped_quantile,
clipped_count_budget_allocation=FLAGS.
clipped_count_budget_allocation,
expected_clients_per_round=FLAGS.clients_per_round,
per_vector_clipping=FLAGS.per_vector_clipping,
model=model_fn())
weights_type = tff.learning.framework.weights_type_from_model(model_fn)
aggregation_process = tff.utils.build_dp_aggregate_process(
weights_type.trainable, dp_query)
else:
aggregation_process = None
server_optimizer_fn = optimizer_utils.create_optimizer_fn_from_flags(
'server')
client_optimizer_fn = optimizer_utils.create_optimizer_fn_from_flags(
'client')
iterative_process = tff.learning.build_federated_averaging_process(
model_fn=model_fn,
server_optimizer_fn=server_optimizer_fn,
client_weight_fn=client_weight_fn,
client_optimizer_fn=client_optimizer_fn,
aggregation_process=aggregation_process)
client_datasets_fn = training_utils.build_client_datasets_fn(
train_dataset, FLAGS.clients_per_round)
evaluate_fn = training_utils.build_evaluate_fn(
model_builder=model_builder,
eval_dataset=validation_dataset,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
test_fn = training_utils.build_evaluate_fn(
model_builder=model_builder,
# Use both val and test for symmetry with other experiments, which
# evaluate on the entire test set.
eval_dataset=validation_dataset.concatenate(test_dataset),
loss_builder=loss_builder,
metrics_builder=metrics_builder)
logging.info('Training model:')
logging.info(model_builder().summary())
hparam_dict = utils_impl.lookup_flag_values(utils_impl.get_hparam_flags())
training_loop_dict = utils_impl.lookup_flag_values(training_loop_flags)
training_loop.run(iterative_process=iterative_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=test_fn,
hparam_dict=hparam_dict,
**training_loop_dict)
def run_federated(
iterative_process_builder: Callable[...,
tff.templates.IterativeProcess],
client_epochs_per_round: int,
client_batch_size: int,
clients_per_round: int,
max_batches_per_client: Optional[int] = -1,
client_datasets_random_seed: Optional[int] = None,
sequence_length: Optional[int] = 80,
total_rounds: Optional[int] = 1500,
experiment_name: Optional[str] = 'federated_shakespeare',
root_output_dir: Optional[str] = '/tmp/fed_opt',
max_eval_batches: Optional[int] = None,
**kwargs):
"""Runs an iterative process on a Shakespeare next character prediction task.
This method will load and pre-process dataset and construct a model used for
the task. It then uses `iterative_process_builder` to create an iterative
process that it applies to the task, using
`federated_research.utils.training_loop`.
We assume that the iterative process has the following functional type
signatures:
* `initialize`: `( -> S@SERVER)` where `S` represents the server state.
* `next`: `<S@SERVER, {B*}@CLIENTS> -> <S@SERVER, T@SERVER>` where `S`
represents the server state, `{B*}` represents the client datasets,
and `T` represents a python `Mapping` object.
Moreover, the server state must have an attribute `model` of type
`tff.learning.ModelWeights`.
Args:
iterative_process_builder: A function that accepts a no-arg `model_fn`, and
a `client_weight_fn`, and returns a `tff.templates.IterativeProcess`. The
`model_fn` must return a `tff.learning.Model`.
client_epochs_per_round: An integer representing the number of epochs of
training performed per client in each training round.
client_batch_size: An integer representing the batch size used on clients.
clients_per_round: An integer representing the number of clients
participating in each round.
max_batches_per_client: An optional int specifying the number of batches
taken by each client at each round. If `-1`, the entire client dataset is
used.
client_datasets_random_seed: An optional int used to seed which clients are
sampled at each round. If `None`, no seed is used.
sequence_length: An int specifying the length of the character sequences
used for prediction.
total_rounds: The number of federated training rounds.
experiment_name: The name of the experiment being run. This will be appended
to the `root_output_dir` for purposes of writing outputs.
root_output_dir: The name of the root output directory for writing
experiment outputs.
max_eval_batches: If set to a positive integer, evaluation datasets are
capped to at most that many batches. If set to None or a nonpositive
integer, the full evaluation datasets are used.
**kwargs: Additional arguments configuring the training loop. For details
on supported arguments, see
`federated_research/utils/training_utils.py`.
"""
train_clientdata = shakespeare_dataset.construct_character_level_datasets(
client_batch_size=client_batch_size,
client_epochs_per_round=client_epochs_per_round,
sequence_length=sequence_length,
max_batches_per_client=max_batches_per_client)
_, test_dataset = shakespeare_dataset.get_centralized_datasets(
train_batch_size=client_batch_size,
max_test_batches=max_eval_batches,
sequence_length=sequence_length)
model_builder = functools.partial(create_shakespeare_model,
sequence_length=sequence_length)
loss_builder = functools.partial(
tf.keras.losses.SparseCategoricalCrossentropy, from_logits=True)
input_spec = train_clientdata.create_tf_dataset_for_client(
train_clientdata.client_ids[0]).element_spec
def client_weight_fn(local_outputs):
# Num_tokens is a tensor with type int64[1], to use as a weight need
# a float32 scalar.
return tf.cast(tf.squeeze(local_outputs['num_tokens']), tf.float32)
def tff_model_fn() -> tff.learning.Model:
return tff.learning.from_keras_model(keras_model=model_builder(),
input_spec=input_spec,
loss=loss_builder(),
metrics=metrics_builder())
training_process = iterative_process_builder(
tff_model_fn, client_weight_fn=client_weight_fn)
client_datasets_fn = training_utils.build_client_datasets_fn(
train_dataset=train_clientdata,
train_clients_per_round=clients_per_round,
random_seed=client_datasets_random_seed)
evaluate_fn = training_utils.build_evaluate_fn(
eval_dataset=test_dataset,
model_builder=model_builder,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
logging.info('Training model:')
logging.info(model_builder().summary())
training_loop.run(iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=evaluate_fn,
total_rounds=total_rounds,
total_clients=clients_per_round,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
def main(argv):
if len(argv) > 1:
raise app.UsageError('Expected no command-line arguments, '
'got: {}'.format(argv))
emnist_train, emnist_test = emnist_dataset.get_emnist_datasets(
FLAGS.client_batch_size,
FLAGS.client_epochs_per_round,
only_digits=False)
if FLAGS.model == 'cnn':
model_builder = functools.partial(
emnist_models.create_conv_dropout_model, only_digits=False)
elif FLAGS.model == '2nn':
model_builder = functools.partial(
emnist_models.create_two_hidden_layer_model, only_digits=False)
else:
raise ValueError('Cannot handle model flag [{!s}].'.format(
FLAGS.model))
loss_builder = tf.keras.losses.SparseCategoricalCrossentropy
metrics_builder = lambda: [tf.keras.metrics.SparseCategoricalAccuracy()]
if FLAGS.uniform_weighting:
def client_weight_fn(local_outputs):
del local_outputs
return 1.0
else:
client_weight_fn = None # Defaults to the number of examples per client.
def model_fn():
return tff.learning.from_keras_model(
model_builder(),
loss_builder(),
input_spec=emnist_test.element_spec,
metrics=metrics_builder())
if FLAGS.noise_multiplier is not None:
if not FLAGS.uniform_weighting:
raise ValueError(
'Differential privacy is only implemented for uniform weighting.'
)
dp_query = tff.utils.build_dp_query(
clip=FLAGS.clip,
noise_multiplier=FLAGS.noise_multiplier,
expected_total_weight=FLAGS.clients_per_round,
adaptive_clip_learning_rate=FLAGS.adaptive_clip_learning_rate,
target_unclipped_quantile=FLAGS.target_unclipped_quantile,
clipped_count_budget_allocation=FLAGS.
clipped_count_budget_allocation,
expected_clients_per_round=FLAGS.clients_per_round,
per_vector_clipping=FLAGS.per_vector_clipping,
model=model_fn())
weights_type = tff.learning.framework.weights_type_from_model(model_fn)
aggregation_process = tff.utils.build_dp_aggregate_process(
weights_type.trainable, dp_query)
else:
aggregation_process = None
server_optimizer_fn = optimizer_utils.create_optimizer_fn_from_flags(
'server')
client_optimizer_fn = optimizer_utils.create_optimizer_fn_from_flags(
'client')
iterative_process = tff.learning.build_federated_averaging_process(
model_fn=model_fn,
server_optimizer_fn=server_optimizer_fn,
client_weight_fn=client_weight_fn,
client_optimizer_fn=client_optimizer_fn,
aggregation_process=aggregation_process)
client_datasets_fn = training_utils.build_client_datasets_fn(
emnist_train, FLAGS.clients_per_round)
evaluate_fn = training_utils.build_evaluate_fn(
eval_dataset=emnist_test,
model_builder=model_builder,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
logging.info('Training model:')
logging.info(model_builder().summary())
hparam_dict = utils_impl.lookup_flag_values(utils_impl.get_hparam_flags())
training_loop_dict = utils_impl.lookup_flag_values(training_loop_flags)
training_loop.run(iterative_process=iterative_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
hparam_dict=hparam_dict,
**training_loop_dict)
def run_federated(
iterative_process_builder: Callable[...,
tff.templates.IterativeProcess],
client_epochs_per_round: int,
client_batch_size: int,
clients_per_round: int,
schedule: Optional[str] = 'none',
beta: Optional[float] = 0.,
max_batches_per_client: Optional[int] = -1,
client_datasets_random_seed: Optional[int] = None,
model: Optional[str] = 'cnn',
total_rounds: Optional[int] = 1500,
experiment_name: Optional[str] = 'federated_synthetic',
root_output_dir: Optional[str] = '/tmp/fed_opt',
max_eval_batches: Optional[int] = None,
alpha: Optional[float] = 0.,
beta_data: Optional[float] = 0.,
iid: Optional[int] = 0,
num_users: Optional[int] = 100,
**kwargs):
"""Runs an iterative process on the EMNIST character recognition task.
This method will load and pre-process dataset and construct a model used for
the task. It then uses `iterative_process_builder` to create an iterative
process that it applies to the task, using
`federated_research.utils.training_loop`.
We assume that the iterative process has the following functional type
signatures:
* `initialize`: `( -> S@SERVER)` where `S` represents the server state.
* `next`: `<S@SERVER, {B*}@CLIENTS> -> <S@SERVER, T@SERVER>` where `S`
represents the server state, `{B*}` represents the client datasets,
and `T` represents a python `Mapping` object.
Moreover, the server state must have an attribute `model` of type
`tff.learning.ModelWeights`.
Args:
iterative_process_builder: A function that accepts a no-arg `model_fn`, and
returns a `tff.templates.IterativeProcess`. The `model_fn` must return a
`tff.learning.Model`.
client_epochs_per_round: An integer representing the number of epochs of
training performed per client in each training round.
client_batch_size: An integer representing the batch size used on clients.
clients_per_round: An integer representing the number of clients
participating in each round.
max_batches_per_client: An optional int specifying the number of batches
taken by each client at each round. If `-1`, the entire client dataset is
used.
client_datasets_random_seed: An optional int used to seed which clients are
sampled at each round. If `None`, no seed is used.
model: A string specifying the model used for character recognition.
Can be one of `cnn` and `2nn`, corresponding to a CNN model and a densely
connected 2-layer model (respectively).
total_rounds: The number of federated training rounds.
experiment_name: The name of the experiment being run. This will be appended
to the `root_output_dir` for purposes of writing outputs.
root_output_dir: The name of the root output directory for writing
experiment outputs.
max_eval_batches: If set to a positive integer, evaluation datasets are
capped to at most that many batches. If set to None or a nonpositive
integer, the full evaluation datasets are used.
**kwargs: Additional arguments configuring the training loop. For details
on supported arguments, see
`federated_research/utils/training_utils.py`.
"""
logging.info(f' DATA PARAMS: ')
logging.info(f' Num Users: {num_users}')
logging.info(f' alpha: {alpha}')
logging.info(f' beta: {beta_data}')
logging.info(f' iid: {iid}')
train_data, test_data, federated_test = synthetic_dataset.generate_federated_softmax_data(
batch_size=client_batch_size,
client_epochs_per_round=client_epochs_per_round,
test_batch_size=100,
alpha=alpha,
beta=beta_data,
iid=iid,
num_users=num_users)
input_spec = train_data.create_tf_dataset_for_client(
train_data.client_ids[0]).element_spec
model_builder = functools.partial(create_logistic_regression_model)
loss_builder = tf.keras.losses.SparseCategoricalCrossentropy
metrics_builder = lambda: [tf.keras.metrics.SparseCategoricalAccuracy()]
def tff_model_fn() -> tff.learning.Model:
return tff.learning.from_keras_model(keras_model=model_builder(),
input_spec=input_spec,
loss=loss_builder(),
metrics=metrics_builder())
training_process = iterative_process_builder(tff_model_fn)
evaluate_fn = training_utils.build_evaluate_fn(
eval_dataset=test_data,
model_builder=model_builder,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
test_fn = training_utils.build_unweighted_test_fn(
federated_eval_dataset=federated_test,
model_builder=model_builder,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
logging.info('Training model:')
logging.info(model_builder().summary())
try:
var = kwargs['hparam_dict']['var_q_clients']
print(f'Variance: {var}')
q_client = np.load(
f'/home/monica/AVAIL_VECTORS/q_client_{var}_synthetic.npy')
#if q_client is None:
#logging.info('Could not load q_client - initializing random availabilities')
#q_client=None
except:
logging.info(
'Could not load q_client - initializing random availabilities')
q_client = None
if schedule == 'none':
client_datasets_fn = training_utils.build_client_datasets_fn(
train_dataset=train_data,
train_clients_per_round=clients_per_round,
random_seed=client_datasets_random_seed,
min_clients=kwargs['hparam_dict']['min_clients'],
var_q_clients=kwargs['hparam_dict']['var_q_clients'],
f_mult=kwargs['hparam_dict']['f_mult'],
f_intercept=kwargs['hparam_dict']['f_intercept'],
sine_wave=kwargs['hparam_dict']['sine_wave'],
use_p=True,
q_client=q_client,
)
training_loop.run(iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=test_fn,
total_rounds=total_rounds,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
elif schedule == 'loss':
if 'loss_pool_size' in kwargs['hparam_dict'] and kwargs['hparam_dict'][
'loss_pool_size'] is not None:
loss_pool_size = kwargs['hparam_dict']['loss_pool_size']
logging.info(f'Loss pool size: {loss_pool_size}')
client_datasets_fn = training_utils.build_client_datasets_fn(
train_dataset=train_data,
train_clients_per_round=loss_pool_size,
random_seed=client_datasets_random_seed,
min_clients=kwargs['hparam_dict']['min_clients'],
var_q_clients=kwargs['hparam_dict']['var_q_clients'],
f_mult=kwargs['hparam_dict']['f_mult'],
f_intercept=kwargs['hparam_dict']['f_intercept'],
sine_wave=kwargs['hparam_dict']['sine_wave'],
use_p=True,
q_client=q_client)
training_loop_loss.run(iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=test_fn,
total_rounds=total_rounds,
total_clients=loss_pool_size,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
else:
raise ValueError('Loss pool size not specified')
else:
init_p = kwargs['hparam_dict']['initialize_p']
logging.info(f'Initializing as p = {init_p}')
client_datasets_fn = training_utils.build_availability_client_datasets_fn(
train_dataset=train_data,
train_clients_per_round=clients_per_round,
beta=beta,
min_clients=kwargs['hparam_dict']['min_clients'],
var_q_clients=kwargs['hparam_dict']['var_q_clients'],
f_mult=kwargs['hparam_dict']['f_mult'],
f_intercept=kwargs['hparam_dict']['f_intercept'],
sine_wave=kwargs['hparam_dict']['sine_wave'],
q_client=q_client,
initialize_p=init_p,
)
training_loop_importance.run(iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=test_fn,
total_rounds=total_rounds,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
def run_federated(
iterative_process_builder: Callable[..., tff.templates.IterativeProcess],
client_epochs_per_round: int,
client_batch_size: int,
clients_per_round: int,
max_batches_per_client: Optional[int] = -1,
client_datasets_random_seed: Optional[int] = None,
vocab_size: Optional[int] = 10000,
num_oov_buckets: Optional[int] = 1,
sequence_length: Optional[int] = 20,
max_elements_per_user: Optional[int] = 1000,
num_validation_examples: Optional[int] = 10000,
embedding_size: Optional[int] = 96,
latent_size: Optional[int] = 670,
num_layers: Optional[int] = 1,
shared_embedding: Optional[bool] = False,
total_rounds: Optional[int] = 1500,
experiment_name: Optional[str] = 'federated_so_nwp',
root_output_dir: Optional[str] = '/tmp/fed_opt',
max_eval_batches: Optional[int] = None,
**kwargs):
"""Runs an iterative process on the Stack Overflow next word prediction task.
This method will load and pre-process dataset and construct a model used for
the task. It then uses `iterative_process_builder` to create an iterative
process that it applies to the task, using
`federated_research.utils.training_loop`.
We assume that the iterative process has the following functional type
signatures:
* `initialize`: `( -> S@SERVER)` where `S` represents the server state.
* `next`: `<S@SERVER, {B*}@CLIENTS> -> <S@SERVER, T@SERVER>` where `S`
represents the server state, `{B*}` represents the client datasets,
and `T` represents a python `Mapping` object.
Moreover, the server state must have an attribute `model` of type
`tff.learning.ModelWeights`.
Args:
iterative_process_builder: A function that accepts a no-arg `model_fn`, a
`client_weight_fn` and returns a `tff.templates.IterativeProcess`. The
`model_fn` must return a `tff.learning.Model`.
client_epochs_per_round: An integer representing the number of epochs of
training performed per client in each training round.
client_batch_size: An integer representing the batch size used on clients.
clients_per_round: An integer representing the number of clients
participating in each round.
max_batches_per_client: An optional int specifying the number of batches
taken by each client at each round. If `-1`, the entire client dataset is
used.
client_datasets_random_seed: An optional int used to seed which clients are
sampled at each round. If `None`, no seed is used.
vocab_size: Integer dictating the number of most frequent words to use in
the vocabulary.
num_oov_buckets: The number of out-of-vocabulary buckets to use.
sequence_length: The maximum number of words to take for each sequence.
max_elements_per_user: The maximum number of elements processed for each
client's dataset.
num_validation_examples: The number of test examples to use for validation.
embedding_size: The dimension of the word embedding layer.
latent_size: The dimension of the latent units in the recurrent layers.
num_layers: The number of stacked recurrent layers to use.
shared_embedding: Boolean indicating whether to tie input and output
embeddings.
total_rounds: The number of federated training rounds.
experiment_name: The name of the experiment being run. This will be appended
to the `root_output_dir` for purposes of writing outputs.
root_output_dir: The name of the root output directory for writing
experiment outputs.
max_eval_batches: If set to a positive integer, evaluation datasets are
capped to at most that many batches. If set to None or a nonpositive
integer, the full evaluation datasets are used.
**kwargs: Additional arguments configuring the training loop. For details
on supported arguments, see
`federated_research/utils/training_utils.py`.
"""
model_builder = functools.partial(
stackoverflow_models.create_recurrent_model,
vocab_size=vocab_size,
num_oov_buckets=num_oov_buckets,
embedding_size=embedding_size,
latent_size=latent_size,
num_layers=num_layers,
shared_embedding=shared_embedding)
loss_builder = functools.partial(
tf.keras.losses.SparseCategoricalCrossentropy, from_logits=True)
special_tokens = stackoverflow_dataset.get_special_tokens(
vocab_size, num_oov_buckets)
pad_token = special_tokens.pad
oov_tokens = special_tokens.oov
eos_token = special_tokens.eos
def metrics_builder():
return [
keras_metrics.MaskedCategoricalAccuracy(
name='accuracy_with_oov', masked_tokens=[pad_token]),
keras_metrics.MaskedCategoricalAccuracy(
name='accuracy_no_oov', masked_tokens=[pad_token] + oov_tokens),
# Notice BOS never appears in ground truth.
keras_metrics.MaskedCategoricalAccuracy(
name='accuracy_no_oov_or_eos',
masked_tokens=[pad_token, eos_token] + oov_tokens),
keras_metrics.NumBatchesCounter(),
keras_metrics.NumTokensCounter(masked_tokens=[pad_token])
]
train_clientdata, _, _ = tff.simulation.datasets.stackoverflow.load_data()
# TODO(b/161914546): consider moving evaluation to use
# `tff.learning.build_federated_evaluation` to get metrics over client
# distributions, as well as the example weight means from this centralized
# evaluation.
_, validation_dataset, test_dataset = stackoverflow_dataset.get_centralized_datasets(
vocab_size=vocab_size,
max_seq_len=sequence_length,
train_batch_size=client_batch_size,
max_validation_batches=max_eval_batches,
max_test_batches=max_eval_batches,
num_validation_examples=num_validation_examples,
num_oov_buckets=num_oov_buckets)
train_dataset_preprocess_comp = stackoverflow_dataset.create_train_dataset_preprocess_fn(
vocab=stackoverflow_dataset.create_vocab(vocab_size),
num_oov_buckets=num_oov_buckets,
client_batch_size=client_batch_size,
client_epochs_per_round=client_epochs_per_round,
max_seq_len=sequence_length,
max_training_elements_per_user=max_elements_per_user,
max_batches_per_user=max_batches_per_client)
input_spec = train_dataset_preprocess_comp.type_signature.result.element
def tff_model_fn() -> tff.learning.Model:
return tff.learning.from_keras_model(
keras_model=model_builder(),
input_spec=input_spec,
loss=loss_builder(),
metrics=metrics_builder())
def client_weight_fn(local_outputs):
# Num_tokens is a tensor with type int64[1], to use as a weight need
# a float32 scalar.
return tf.cast(tf.squeeze(local_outputs['num_tokens']), tf.float32)
training_process = iterative_process_builder(
tff_model_fn, client_weight_fn=client_weight_fn)
training_process = tff.simulation.compose_dataset_computation_with_iterative_process(
train_dataset_preprocess_comp, training_process)
client_datasets_fn = training_utils.build_client_datasets_fn(
train_dataset=train_clientdata,
train_clients_per_round=clients_per_round,
random_seed=client_datasets_random_seed)
evaluate_fn = training_utils.build_evaluate_fn(
model_builder=model_builder,
eval_dataset=validation_dataset,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
test_fn = training_utils.build_evaluate_fn(
model_builder=model_builder,
# Use both val and test for symmetry with other experiments, which
# evaluate on the entire test set.
eval_dataset=validation_dataset.concatenate(test_dataset),
loss_builder=loss_builder,
metrics_builder=metrics_builder)
logging.info('Training model:')
logging.info(model_builder().summary())
training_loop.run(
iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=test_fn,
total_rounds=total_rounds,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
def run_federated(
iterative_process_builder: Callable[...,
tff.templates.IterativeProcess],
client_epochs_per_round: int,
client_batch_size: int,
clients_per_round: int,
max_batches_per_client: Optional[int] = -1,
client_datasets_random_seed: Optional[int] = None,
vocab_tokens_size: Optional[int] = 10000,
vocab_tags_size: Optional[int] = 500,
max_elements_per_user: Optional[int] = 1000,
num_validation_examples: Optional[int] = 10000,
total_rounds: Optional[int] = 1500,
experiment_name: Optional[str] = 'federated_so_lr',
root_output_dir: Optional[str] = '/tmp/fed_opt',
max_eval_batches: Optional[int] = None,
**kwargs):
"""Runs an iterative process on the Stack Overflow logistic regression task.
This method will load and pre-process dataset and construct a model used for
the task. It then uses `iterative_process_builder` to create an iterative
process that it applies to the task, using
`federated_research.utils.training_loop`.
We assume that the iterative process has the following functional type
signatures:
* `initialize`: `( -> S@SERVER)` where `S` represents the server state.
* `next`: `<S@SERVER, {B*}@CLIENTS> -> <S@SERVER, T@SERVER>` where `S`
represents the server state, `{B*}` represents the client datasets,
and `T` represents a python `Mapping` object.
Moreover, the server state must have an attribute `model` of type
`tff.learning.ModelWeights`.
Args:
iterative_process_builder: A function that accepts a no-arg `model_fn`, and
returns a `tff.templates.IterativeProcess`. The `model_fn` must return a
`tff.learning.Model`.
client_epochs_per_round: An integer representing the number of epochs of
training performed per client in each training round.
client_batch_size: An integer representing the batch size used on clients.
clients_per_round: An integer representing the number of clients
participating in each round.
max_batches_per_client: An optional int specifying the number of batches
taken by each client at each round. If `-1`, the entire client dataset is
used.
client_datasets_random_seed: An optional int used to seed which clients are
sampled at each round. If `None`, no seed is used.
vocab_tokens_size: Integer dictating the number of most frequent words to
use in the vocabulary.
vocab_tags_size: Integer dictating the number of most frequent tags to use
in the label creation.
max_elements_per_user: The maximum number of elements processed for each
client's dataset.
num_validation_examples: The number of test examples to use for validation.
total_rounds: The number of federated training rounds.
experiment_name: The name of the experiment being run. This will be appended
to the `root_output_dir` for purposes of writing outputs.
root_output_dir: The name of the root output directory for writing
experiment outputs.
max_eval_batches: If set to a positive integer, evaluation datasets are
capped to at most that many batches. If set to None or a nonpositive
integer, the full evaluation datasets are used.
**kwargs: Additional arguments configuring the training loop. For details
on supported arguments, see
`federated_research/utils/training_utils.py`.
"""
stackoverflow_train, _, _ = stackoverflow_lr_dataset.get_stackoverflow_datasets(
vocab_tokens_size=vocab_tokens_size,
vocab_tags_size=vocab_tags_size,
client_batch_size=client_batch_size,
client_epochs_per_round=client_epochs_per_round,
max_training_elements_per_user=max_elements_per_user,
max_batches_per_user=max_batches_per_client,
num_validation_examples=num_validation_examples)
_, stackoverflow_validation, stackoverflow_test = stackoverflow_lr_dataset.get_centralized_datasets(
train_batch_size=client_batch_size,
vocab_tokens_size=vocab_tokens_size,
vocab_tags_size=vocab_tags_size,
num_validation_examples=num_validation_examples,
max_validation_batches=max_eval_batches,
max_test_batches=max_eval_batches)
input_spec = stackoverflow_train.create_tf_dataset_for_client(
stackoverflow_train.client_ids[0]).element_spec
model_builder = functools.partial(
stackoverflow_lr_models.create_logistic_model,
vocab_tokens_size=vocab_tokens_size,
vocab_tags_size=vocab_tags_size)
loss_builder = functools.partial(tf.keras.losses.BinaryCrossentropy,
from_logits=False,
reduction=tf.keras.losses.Reduction.SUM)
def tff_model_fn() -> tff.learning.Model:
return tff.learning.from_keras_model(keras_model=model_builder(),
input_spec=input_spec,
loss=loss_builder(),
metrics=metrics_builder())
training_process = iterative_process_builder(tff_model_fn)
client_datasets_fn = training_utils.build_client_datasets_fn(
train_dataset=stackoverflow_train,
train_clients_per_round=clients_per_round,
random_seed=client_datasets_random_seed)
evaluate_fn = training_utils.build_evaluate_fn(
model_builder=model_builder,
eval_dataset=stackoverflow_validation,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
test_fn = training_utils.build_evaluate_fn(
model_builder=model_builder,
# Use both val and test for symmetry with other experiments, which
# evaluate on the entire test set.
eval_dataset=stackoverflow_validation.concatenate(stackoverflow_test),
loss_builder=loss_builder,
metrics_builder=metrics_builder)
logging.info('Training model:')
logging.info(model_builder().summary())
training_loop.run(iterative_process=training_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
test_fn=test_fn,
total_rounds=total_rounds,
experiment_name=experiment_name,
root_output_dir=root_output_dir,
**kwargs)
def run_experiment():
"""Data preprocessing and experiment execution."""
emnist_train, emnist_test = emnist_dataset.get_emnist_datasets(
FLAGS.client_batch_size,
FLAGS.client_epochs_per_round,
only_digits=FLAGS.only_digits)
example_dataset = emnist_train.create_tf_dataset_for_client(
emnist_train.client_ids[0])
input_spec = example_dataset.element_spec
client_datasets_fn = training_utils.build_client_datasets_fn(
emnist_train, FLAGS.clients_per_round)
evaluate_fn = training_utils.build_evaluate_fn(
eval_dataset=emnist_test,
model_builder=model_builder,
loss_builder=loss_builder,
metrics_builder=metrics_builder)
client_optimizer_fn = functools.partial(
utils_impl.create_optimizer_from_flags, 'client')
server_optimizer_fn = functools.partial(
utils_impl.create_optimizer_from_flags, 'server')
def tff_model_fn():
keras_model = model_builder()
return tff.learning.from_keras_model(
keras_model,
input_spec=input_spec,
loss=loss_builder(),
metrics=metrics_builder())
if FLAGS.use_compression:
# We create a `MeasuredProcess` for broadcast process and a
# `MeasuredProcess` for aggregate process by providing the
# `_broadcast_encoder_fn` and `_mean_encoder_fn` to corresponding utilities.
# The fns are called once for each of the model weights created by
# tff_model_fn, and return instances of appropriate encoders.
encoded_broadcast_process = (
tff.learning.framework.build_encoded_broadcast_process_from_model(
tff_model_fn, _broadcast_encoder_fn))
encoded_mean_process = (
tff.learning.framework.build_encoded_mean_process_from_model(
tff_model_fn, _mean_encoder_fn))
else:
encoded_broadcast_process = None
encoded_mean_process = None
iterative_process = tff.learning.build_federated_averaging_process(
model_fn=tff_model_fn,
client_optimizer_fn=client_optimizer_fn,
server_optimizer_fn=server_optimizer_fn,
aggregation_process=encoded_mean_process,
broadcast_process=encoded_broadcast_process)
hparam_dict = utils_impl.lookup_flag_values(utils_impl.get_hparam_flags())
training_loop_dict = utils_impl.lookup_flag_values(training_loop_flags)
training_loop.run(
iterative_process=iterative_process,
client_datasets_fn=client_datasets_fn,
validation_fn=evaluate_fn,
hparam_dict=hparam_dict,
**training_loop_dict)
