def get_broadcast_form_for_computation(
comp: computation_base.Computation,
grappler_config: tf.compat.v1.ConfigProto = _GRAPPLER_DEFAULT_CONFIG
) -> forms.BroadcastForm:
"""Constructs `tff.backends.mapreduce.BroadcastForm` given a computation.
Args:
comp: An instance of `tff.Computation` that is compatible with broadcast
form. Computations are only compatible if they take in a single value
placed at server, return a single value placed at clients, and do not
contain any aggregations.
grappler_config: An instance of `tf.compat.v1.ConfigProto` to configure
Grappler graph optimization of the Tensorflow graphs backing the resulting
`tff.backends.mapreduce.BroadcastForm`. These options are combined with a
set of defaults that aggressively configure Grappler. If
`grappler_config_proto` has
`graph_options.rewrite_options.disable_meta_optimizer=True`, Grappler is
bypassed.
Returns:
An instance of `tff.backends.mapreduce.BroadcastForm` equivalent to the
provided `tff.Computation`.
"""
py_typecheck.check_type(comp, computation_base.Computation)
_check_function_signature_compatible_with_broadcast_form(
comp.type_signature)
py_typecheck.check_type(grappler_config, tf.compat.v1.ConfigProto)
grappler_config = _merge_grappler_config_with_default(grappler_config)
bb = comp.to_building_block()
bb, _ = intrinsic_reductions.replace_intrinsics_with_bodies(bb)
bb = _replace_lambda_body_with_call_dominant_form(bb)
tree_analysis.check_contains_only_reducible_intrinsics(bb)
aggregations = tree_analysis.find_aggregations_in_tree(bb)
if aggregations:
raise ValueError(
f'`get_broadcast_form_for_computation` called with computation '
f'containing {len(aggregations)} aggregations, but broadcast form '
'does not allow aggregation. Full list of aggregations:\n{aggregations}'
)
before_broadcast, after_broadcast = _split_ast_on_broadcast(bb)
compute_server_context = _extract_compute_server_context(
before_broadcast, grappler_config)
client_processing = _extract_client_processing(after_broadcast,
grappler_config)
compute_server_context, client_processing = (
computation_wrapper_instances.building_block_to_computation(bb)
for bb in (compute_server_context, client_processing))
comp_param_names = structure.name_list_with_nones(
comp.type_signature.parameter)
server_data_label, client_data_label = comp_param_names
return forms.BroadcastForm(compute_server_context,
client_processing,
server_data_label=server_data_label,
client_data_label=client_data_label)
def test_passes_with_federated_map(self):
intrinsic = building_blocks.Intrinsic(
intrinsic_defs.FEDERATED_MAP.uri,
computation_types.FunctionType([
computation_types.FunctionType(tf.int32, tf.float32),
computation_types.FederatedType(tf.int32, placements.CLIENTS)
], computation_types.FederatedType(tf.float32, placements.CLIENTS)))
tree_analysis.check_contains_only_reducible_intrinsics(intrinsic)
def test_raises_with_federated_mean(self):
intrinsic = building_blocks.Intrinsic(
intrinsic_defs.FEDERATED_MEAN.uri,
computation_types.FunctionType(
computation_types.FederatedType(tf.int32, placements.CLIENTS),
computation_types.FederatedType(tf.int32, placements.SERVER)))
with self.assertRaisesRegex(ValueError, intrinsic.compact_representation()):
tree_analysis.check_contains_only_reducible_intrinsics(intrinsic)
def check_iterative_process_compatible_with_map_reduce_form(
ip: iterative_process.IterativeProcess):
"""Tests compatibility with `tff.backends.mapreduce.MapReduceForm`.
Note: the conditions here are specified in the documentation for
`get_map_reduce_form_for_iterative_process`. Changes to this function should
be propagated to that documentation.
Args:
ip: An instance of `tff.templates.IterativeProcess` to check for
compatibility with `tff.backends.mapreduce.MapReduceForm`.
Returns:
TFF-internal building-blocks representing the validated and simplified
`initialize` and `next` computations.
Raises:
TypeError: If the arguments are of the wrong types.
"""
py_typecheck.check_type(ip, iterative_process.IterativeProcess)
initialize_tree = ip.initialize.to_building_block()
next_tree = ip.next.to_building_block()
init_type = initialize_tree.type_signature
_check_type_is_no_arg_fn(init_type, '`initialize`', TypeError)
if (not init_type.result.is_federated()
or init_type.result.placement != placements.SERVER):
raise TypeError(
'Expected `initialize` to return a single federated value '
'placed at server (type `T@SERVER`), found return type:\n'
f'{init_type.result}')
next_type = next_tree.type_signature
_check_type_is_fn(next_type, '`next`', TypeError)
if not next_type.parameter.is_struct() or len(next_type.parameter) != 2:
raise TypeError(
'Expected `next` to take two arguments, found parameter '
f' type:\n{next_type.parameter}')
if not next_type.result.is_struct() or len(next_type.result) != 2:
raise TypeError('Expected `next` to return two values, found result '
f'type:\n{next_type.result}')
initialize_tree, _ = intrinsic_reductions.replace_intrinsics_with_bodies(
initialize_tree)
next_tree, _ = intrinsic_reductions.replace_intrinsics_with_bodies(
next_tree)
next_tree = _replace_lambda_body_with_call_dominant_form(next_tree)
tree_analysis.check_contains_only_reducible_intrinsics(initialize_tree)
tree_analysis.check_contains_only_reducible_intrinsics(next_tree)
tree_analysis.check_broadcast_not_dependent_on_aggregate(next_tree)
return initialize_tree, next_tree
def test_generic_plus_reduces(self):
uri = intrinsic_defs.GENERIC_PLUS.uri
comp = building_blocks.Intrinsic(
uri, computation_types.FunctionType([tf.float32, tf.float32],
tf.float32))
count_before_reduction = _count_intrinsics(comp, uri)
reduced, modified = intrinsic_reductions.replace_intrinsics_with_bodies(
comp)
count_after_reduction = _count_intrinsics(reduced, uri)
self.assertTrue(modified)
self.assert_types_identical(comp.type_signature, reduced.type_signature)
self.assertGreater(count_before_reduction, 0)
self.assertEqual(count_after_reduction, 0)
tree_analysis.check_contains_only_reducible_intrinsics(reduced)
def test_generic_divide_reduces(self):
uri = intrinsic_defs.GENERIC_DIVIDE.uri
context_stack = context_stack_impl.context_stack
comp = building_blocks.Intrinsic(
uri,
computation_types.FunctionType([tf.float32, tf.float32],
tf.float32))
count_before_reduction = _count_intrinsics(comp, uri)
reduced, modified = value_transformations.replace_intrinsics_with_bodies(
comp, context_stack)
count_after_reduction = _count_intrinsics(reduced, uri)
self.assertGreater(count_before_reduction, 0)
self.assertEqual(count_after_reduction, 0)
tree_analysis.check_contains_only_reducible_intrinsics(reduced)
self.assertTrue(modified)
def test_raises_on_none(self):
with self.assertRaises(TypeError):
tree_analysis.check_contains_only_reducible_intrinsics(None)
def get_canonical_form_for_iterative_process(ip):
"""Constructs `tff.backends.mapreduce.CanonicalForm` given iterative process.
This function transforms computations from the input `ip` into
an instance of `tff.backends.mapreduce.CanonicalForm`.
Args:
ip: An instance of `tff.templates.IterativeProcess`.
Returns:
An instance of `tff.backends.mapreduce.CanonicalForm` equivalent to this
process.
Raises:
TypeError: If the arguments are of the wrong types.
transformations.CanonicalFormCompilationError: If the compilation
process fails.
"""
py_typecheck.check_type(ip, iterative_process.IterativeProcess)
initialize_comp = building_blocks.ComputationBuildingBlock.from_proto(
ip.initialize._computation_proto) # pylint: disable=protected-access
next_comp = building_blocks.ComputationBuildingBlock.from_proto(
ip.next._computation_proto) # pylint: disable=protected-access
_check_iterative_process_compatible_with_canonical_form(
initialize_comp, next_comp)
initialize_comp = _replace_intrinsics_with_bodies(initialize_comp)
next_comp = _replace_intrinsics_with_bodies(next_comp)
tree_analysis.check_contains_only_reducible_intrinsics(initialize_comp)
tree_analysis.check_contains_only_reducible_intrinsics(next_comp)
tree_analysis.check_broadcast_not_dependent_on_aggregate(next_comp)
if tree_analysis.contains_called_intrinsic(
next_comp, intrinsic_defs.FEDERATED_BROADCAST.uri):
before_broadcast, after_broadcast = (
transformations.force_align_and_split_by_intrinsics(
next_comp, [intrinsic_defs.FEDERATED_BROADCAST.uri]))
else:
before_broadcast, after_broadcast = (
_create_before_and_after_broadcast_for_no_broadcast(next_comp))
contains_federated_aggregate = tree_analysis.contains_called_intrinsic(
next_comp, intrinsic_defs.FEDERATED_AGGREGATE.uri)
contains_federated_secure_sum = tree_analysis.contains_called_intrinsic(
next_comp, intrinsic_defs.FEDERATED_SECURE_SUM.uri)
if not (contains_federated_aggregate or contains_federated_secure_sum):
raise ValueError(
'Expected an `tff.templates.IterativeProcess` containing at least one '
'`federated_aggregate` or `federated_secure_sum`, found none.')
if contains_federated_aggregate and contains_federated_secure_sum:
before_aggregate, after_aggregate = (
transformations.force_align_and_split_by_intrinsics(
after_broadcast, [
intrinsic_defs.FEDERATED_AGGREGATE.uri,
intrinsic_defs.FEDERATED_SECURE_SUM.uri,
]))
elif contains_federated_secure_sum:
assert not contains_federated_aggregate
before_aggregate, after_aggregate = (
_create_before_and_after_aggregate_for_no_federated_aggregate(
after_broadcast))
else:
assert contains_federated_aggregate and not contains_federated_secure_sum
before_aggregate, after_aggregate = (
_create_before_and_after_aggregate_for_no_federated_secure_sum(
after_broadcast))
type_info = _get_type_info(initialize_comp, before_broadcast, after_broadcast,
before_aggregate, after_aggregate)
initialize = transformations.consolidate_and_extract_local_processing(
initialize_comp)
_check_type_equal(initialize.type_signature, type_info['initialize_type'])
prepare = _extract_prepare(before_broadcast)
_check_type_equal(prepare.type_signature, type_info['prepare_type'])
work = _extract_work(before_aggregate)
_check_type_equal(work.type_signature, type_info['work_type'])
zero, accumulate, merge, report = _extract_federated_aggregate_functions(
before_aggregate)
_check_type_equal(zero.type_signature, type_info['zero_type'])
_check_type_equal(accumulate.type_signature, type_info['accumulate_type'])
_check_type_equal(merge.type_signature, type_info['merge_type'])
_check_type_equal(report.type_signature, type_info['report_type'])
bitwidth = _extract_federated_secure_sum_functions(before_aggregate)
_check_type_equal(bitwidth.type_signature, type_info['bitwidth_type'])
update = _extract_update(after_aggregate)
_check_type_equal(update.type_signature, type_info['update_type'])
return canonical_form.CanonicalForm(
computation_wrapper_instances.building_block_to_computation(initialize),
computation_wrapper_instances.building_block_to_computation(prepare),
computation_wrapper_instances.building_block_to_computation(work),
computation_wrapper_instances.building_block_to_computation(zero),
computation_wrapper_instances.building_block_to_computation(accumulate),
computation_wrapper_instances.building_block_to_computation(merge),
computation_wrapper_instances.building_block_to_computation(report),
computation_wrapper_instances.building_block_to_computation(bitwidth),
computation_wrapper_instances.building_block_to_computation(update))
def get_canonical_form_for_iterative_process(
ip: iterative_process.IterativeProcess,
grappler_config: Optional[
tf.compat.v1.ConfigProto] = _GRAPPLER_DEFAULT_CONFIG):
"""Constructs `tff.backends.mapreduce.CanonicalForm` given iterative process.
This function transforms computations from the input `ip` into
an instance of `tff.backends.mapreduce.CanonicalForm`.
Args:
ip: An instance of `tff.templates.IterativeProcess` that is compatible
with canonical form. Iterative processes are only compatible if: -
`initialize_fn` returns a single federated value placed at `SERVER`. -
`next` takes exactly two arguments. The first must be the state value
placed at `SERVER`. - `next` returns exactly two values.
grappler_config: An optional instance of `tf.compat.v1.ConfigProto` to
configure Grappler graph optimization of the TensorFlow graphs backing the
resulting `tff.backends.mapreduce.CanonicalForm`. These options are
combined with a set of defaults that aggressively configure Grappler. If
`None`, Grappler is bypassed.
Returns:
An instance of `tff.backends.mapreduce.CanonicalForm` equivalent to this
process.
Raises:
TypeError: If the arguments are of the wrong types.
transformations.CanonicalFormCompilationError: If the compilation
process fails.
"""
py_typecheck.check_type(ip, iterative_process.IterativeProcess)
if grappler_config is not None:
py_typecheck.check_type(grappler_config, tf.compat.v1.ConfigProto)
overridden_grappler_config = tf.compat.v1.ConfigProto()
overridden_grappler_config.CopyFrom(_GRAPPLER_DEFAULT_CONFIG)
overridden_grappler_config.MergeFrom(grappler_config)
grappler_config = overridden_grappler_config
initialize_comp = building_blocks.ComputationBuildingBlock.from_proto(
ip.initialize._computation_proto) # pylint: disable=protected-access
next_comp = building_blocks.ComputationBuildingBlock.from_proto(
ip.next._computation_proto) # pylint: disable=protected-access
_check_iterative_process_compatible_with_canonical_form(
initialize_comp, next_comp)
initialize_comp = _replace_intrinsics_with_bodies(initialize_comp)
next_comp = _replace_intrinsics_with_bodies(next_comp)
next_comp = _replace_lambda_body_with_call_dominant_form(next_comp)
tree_analysis.check_contains_only_reducible_intrinsics(initialize_comp)
tree_analysis.check_contains_only_reducible_intrinsics(next_comp)
tree_analysis.check_broadcast_not_dependent_on_aggregate(next_comp)
if tree_analysis.contains_called_intrinsic(
next_comp, intrinsic_defs.FEDERATED_BROADCAST.uri):
before_broadcast, after_broadcast = (
transformations.force_align_and_split_by_intrinsics(
next_comp, [intrinsic_defs.FEDERATED_BROADCAST.uri]))
else:
before_broadcast, after_broadcast = (
_create_before_and_after_broadcast_for_no_broadcast(next_comp))
contains_federated_aggregate = tree_analysis.contains_called_intrinsic(
next_comp, intrinsic_defs.FEDERATED_AGGREGATE.uri)
contains_federated_secure_sum = tree_analysis.contains_called_intrinsic(
next_comp, intrinsic_defs.FEDERATED_SECURE_SUM.uri)
if not (contains_federated_aggregate or contains_federated_secure_sum):
raise ValueError(
'Expected an `tff.templates.IterativeProcess` containing at least one '
'`federated_aggregate` or `federated_secure_sum`, found none.')
if contains_federated_aggregate and contains_federated_secure_sum:
before_aggregate, after_aggregate = (
transformations.force_align_and_split_by_intrinsics(
after_broadcast, [
intrinsic_defs.FEDERATED_AGGREGATE.uri,
intrinsic_defs.FEDERATED_SECURE_SUM.uri,
]))
elif contains_federated_secure_sum:
assert not contains_federated_aggregate
before_aggregate, after_aggregate = (
_create_before_and_after_aggregate_for_no_federated_aggregate(
after_broadcast))
else:
assert contains_federated_aggregate and not contains_federated_secure_sum
before_aggregate, after_aggregate = (
_create_before_and_after_aggregate_for_no_federated_secure_sum(
after_broadcast))
type_info = _get_type_info(initialize_comp, before_broadcast,
after_broadcast, before_aggregate,
after_aggregate)
initialize = transformations.consolidate_and_extract_local_processing(
initialize_comp, grappler_config)
_check_type_equal(initialize.type_signature, type_info['initialize_type'])
prepare = _extract_prepare(before_broadcast, grappler_config)
_check_type_equal(prepare.type_signature, type_info['prepare_type'])
work = _extract_work(before_aggregate, grappler_config)
_check_type_equal(work.type_signature, type_info['work_type'])
zero, accumulate, merge, report = _extract_federated_aggregate_functions(
before_aggregate, grappler_config)
_check_type_equal(zero.type_signature, type_info['zero_type'])
_check_type_equal(accumulate.type_signature, type_info['accumulate_type'])
_check_type_equal(merge.type_signature, type_info['merge_type'])
_check_type_equal(report.type_signature, type_info['report_type'])
bitwidth = _extract_federated_secure_sum_functions(before_aggregate,
grappler_config)
_check_type_equal(bitwidth.type_signature, type_info['bitwidth_type'])
update = _extract_update(after_aggregate, grappler_config)
_check_type_equal(update.type_signature, type_info['update_type'])
next_parameter_names = (name for (
name, _) in structure.iter_elements(ip.next.type_signature.parameter))
server_state_label, client_data_label = next_parameter_names
return canonical_form.CanonicalForm(
computation_wrapper_instances.building_block_to_computation(
initialize),
computation_wrapper_instances.building_block_to_computation(prepare),
computation_wrapper_instances.building_block_to_computation(work),
computation_wrapper_instances.building_block_to_computation(zero),
computation_wrapper_instances.building_block_to_computation(
accumulate),
computation_wrapper_instances.building_block_to_computation(merge),
computation_wrapper_instances.building_block_to_computation(report),
computation_wrapper_instances.building_block_to_computation(bitwidth),
computation_wrapper_instances.building_block_to_computation(update),
server_state_label=server_state_label,
client_data_label=client_data_label)
