# Copyright 2018 The TensorFlow Authors. 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. # ============================================================================== # pylint: disable=unidiomatic-typecheck """Prototype decorator for defining legacy-graph-mode functions.""" import weakref from tensorflow.core.function.polymorphism import function_type as function_type_lib from tensorflow.core.protobuf import meta_graph_pb2 from tensorflow.core.protobuf import struct_pb2 from tensorflow.python.eager import context from tensorflow.python.eager import function from tensorflow.python.eager import lift_to_graph from tensorflow.python.eager.polymorphic_function import atomic_function from tensorflow.python.framework import composite_tensor from tensorflow.python.framework import func_graph from tensorflow.python.framework import importer from tensorflow.python.framework import ops from tensorflow.python.framework import sparse_tensor from tensorflow.python.framework import tensor as tensor_lib from tensorflow.python.framework import tensor_shape from tensorflow.python.framework import tensor_util from tensorflow.python.ops import resource_variable_ops from tensorflow.python.ops import variable_scope from tensorflow.python.platform import tf_logging as logging from tensorflow.python.saved_model import nested_structure_coder from tensorflow.python.trackable import data_structures from tensorflow.python.util import nest from tensorflow.python.util.tf_export import tf_export class VariableHolder(object): """Holds variables for a python function.""" def __init__(self, fn=None, share_variables=False): self._fn = fn self._share_variables = share_variables self._variables_by_name = data_structures.Mapping() @property def variables(self): return self._variables_by_name def variable_creator_scope(self, next_creator, **kwargs): """Creates variables & adds them to collections to match legacy code.""" collections = kwargs.pop("collections", None) v = None # Get expected variable name. with ops.name_scope( kwargs.get("name", None), "Variable", skip_on_eager=False) as name: variable_name = ops.name_from_scope_name(name) kwargs["name"] = name if self._share_variables: v = self._variables_by_name.get(variable_name, None) if v is None: v = next_creator(**kwargs) self._variables_by_name[variable_name] = v if collections is None: collections = [ops.GraphKeys.GLOBAL_VARIABLES] if v.trainable and ops.GraphKeys.TRAINABLE_VARIABLES not in collections: collections = list(collections) + [ops.GraphKeys.TRAINABLE_VARIABLES] ops.add_to_collections(collections, v) return v def __call__(self, *args, **kwargs): return self.call_with_variable_creator_scope(self._fn)(*args, **kwargs) def call_with_variable_creator_scope(self, fn): def wrapped(*args, **kwargs): with variable_scope.variable_creator_scope(self.variable_creator_scope): return fn(*args, **kwargs) return wrapped def _get_element_from_tensor_info(tensor_info, graph): """Simplified copy of the deprecated `get_tensor_from_tensor_info`.""" encoding = tensor_info.WhichOneof("encoding") if encoding == "name": # We may get operations here in some cases. TensorInfo is a bit of a # misnomer if so. return graph.as_graph_element(tensor_info.name) elif encoding == "coo_sparse": return sparse_tensor.SparseTensor( graph.get_tensor_by_name(tensor_info.coo_sparse.indices_tensor_name), graph.get_tensor_by_name(tensor_info.coo_sparse.values_tensor_name), graph.get_tensor_by_name( tensor_info.coo_sparse.dense_shape_tensor_name)) elif encoding == "composite_tensor": spec_proto = struct_pb2.StructuredValue( type_spec_value=tensor_info.composite_tensor.type_spec) spec = nested_structure_coder.decode_proto(spec_proto) components = [graph.get_tensor_by_name(component.name) for component in tensor_info.composite_tensor.components] return spec._from_components(components) # pylint: disable=protected-access else: raise ValueError(f"Invalid TensorInfo.encoding: {encoding}. Valid " "encodings are 'name', 'coo_sparse', and " "'composite_tensor'.") def _lift_single_variable(old_variable, graph, variable_holder): """Lifts `old_variable` out of the `FuncGraph` `graph`.""" new_variable = resource_variable_ops.UninitializedVariable( shape=old_variable.shape, dtype=old_variable.dtype, name=old_variable.op.name, trainable=old_variable.trainable, extra_handle_data=old_variable.handle) new_variable._initializer_op = old_variable._initializer_op # pylint: disable=protected-access graph.add_capture(new_variable.handle, old_variable.handle) # Now that we've added the new variable to graph.captures, # graph.capture will use that cached value and do some post-processing # on the capture like recording it on the tape. graph.capture(new_variable.handle) # pylint: disable=protected-access variable_name = new_variable.name.split(":")[0] variable_holder._variables_by_name[variable_name] = new_variable graph._weak_variables.append(weakref.ref(new_variable)) # pylint: enable=protected-access graph.watch_variable(new_variable) return new_variable def _lift_unlifted_variables(graph, variable_holder): """Finds resource variables and lifts them into the outer context. When we import a GraphDef inside a wrap_function, no Python graph building code runs. This means we get VarHandleOps which create variable resources, but no corresponding Python objects. Leaving them like this works but gives the user no way to interact with or modify the variables outside the graph. This method searches for variables and lifts them out as regular variable objects when possible, indicating to the FuncGraph that they are captures. Args: graph: The FuncGraph to lift variables from. variable_holder: A VariableHolder to record the lifted variables in. """ with graph.as_default(): global_collection_variables = ops.get_collection( ops.GraphKeys.GLOBAL_VARIABLES) local_collection_variables = ops.get_collection( ops.GraphKeys.LOCAL_VARIABLES) existing_captures = {id(c) for c in graph.internal_captures} lifted_variables = {} def _should_lift_variable(v): return ((v._in_graph_mode # pylint: disable=protected-access and v.graph.building_function) and isinstance(v, resource_variable_ops.BaseResourceVariable) and id(v.handle) not in existing_captures) for old_variable in global_collection_variables: if _should_lift_variable(old_variable): new_variable = _lift_single_variable( old_variable, graph, variable_holder) lifted_variables[id(old_variable)] = new_variable existing_captures.add(id(old_variable.handle)) for old_variable in local_collection_variables: if _should_lift_variable(old_variable): new_variable = _lift_single_variable( old_variable, graph, variable_holder) lifted_variables[id(old_variable)] = new_variable existing_captures.add(id(old_variable.handle)) if new_variable._in_graph_mode: # pylint: disable=protected-access outer_graph = new_variable.graph # Variables are added to the global collection by default. In this # case we only want the variable in the local collection, so we'll pop # it out. global_collection = outer_graph.get_collection_ref( ops.GraphKeys.GLOBAL_VARIABLES) global_collection.remove(new_variable) outer_graph.add_to_collection( ops.GraphKeys.LOCAL_VARIABLES, new_variable) # Update the FuncGraph's collections, partly for the user and partly so this # function is idempotent when it runs again in prune() calls. for collection_name in [ ops.GraphKeys.GLOBAL_VARIABLES, ops.GraphKeys.LOCAL_VARIABLES ]: mutable_collection = ops.get_collection_ref(collection_name) for index, current in enumerate(mutable_collection): mutable_collection[index] = lifted_variables.get(id(current), current) if not resource_variable_ops.is_resource_variable( mutable_collection[index]): logging.log_first_n( logging.WARN, "Unable to create a python object for variable {} because it is " "a reference variable. It may not be visible to training APIs. " "If this is a problem, consider rebuilding the SavedModel after " "running tf.compat.v1.enable_resource_variables().".format( mutable_collection[index]), 5) # TODO(allenl): make this trackable class WrappedFunction(function.ConcreteFunction): """Wraps a tf V1 piece of code in a function.""" def __init__(self, fn_graph, variable_holder, attrs=None, signature=None): self._variable_holder = variable_holder _lift_unlifted_variables(fn_graph, variable_holder) # We call __init__ after lifting variables so that the function's signature # properly reflects the new captured inputs. for f in fn_graph.as_graph_def(use_pybind11_proto=True).library.function: context.context().add_function_def(f) self._signature = signature function_type = function_type_lib.from_structured_signature( fn_graph.structured_input_signature, fn_graph.structured_outputs, fn_graph.function_captures.capture_types, ) atomic_fn = atomic_function.from_func_graph( function._inference_name(fn_graph.name), fn_graph, attrs, function_type ) super().__init__(atomic_fn) def _call_impl(self, args, kwargs): if self._arg_keywords is None: if kwargs: raise NotImplementedError( "Keyword arguments are not supported when calling a " f"wrap_function-decorated function. Got {kwargs}.") if self._signature is not None: args = list(args) for i, arg in enumerate(args): if isinstance(self._signature[i], tensor_lib.DenseSpec): args[i] = ops.convert_to_tensor(arg, self._signature[i].dtype) return self._call_flat(args, self.captured_inputs) else: return super()._call_impl(args, kwargs) def prune(self, feeds, fetches, name=None, input_signature=None): """Extract a subgraph of this function's underlying graph. Wraps the subgraph in a new `WrappedFunction` object. Args: feeds: Input tensors to the subgraph to extract, as `Tensor` objects. fetches: Possibly-nested Python data structure containing information about outputs of the target subgraph. Each entry can either be a `Tensor` object (for data outputs), an `Operation` object (for control outputs), or a `TensorInfo` proto. Any additional shape/dtype information provided in a `TensorInfo` and not present in the original graph will be added to the returned subgraph. name: (optional) Name to give to the underlying `FuncGraph` of the returned object. If no name is provided, the graph's name will be `"pruned"`. input_signature: (optional) possibly-nested Python data structure containing `TensorSpec` objects, with which to populate the returned functions's `FuncGraph`'s `structured_input_signature` field. Returns: A new `WrappedFunction` object containing a copy of the portion of this object's graph that goes from `feeds` to `fetches`. """ # TODO(b/129646028): Add support for CompositeTensors. name = name or "pruned" flat_feeds = nest.flatten(feeds, expand_composites=True) flat_feeds = [self.graph.as_graph_element(t) for t in flat_feeds] for f in flat_feeds: if not isinstance(f, tensor_lib.Tensor): raise ValueError( "All members of argument `feeds` must be tensors. " f"Got {f} with type {type(f)}." ) # Ignoring all feeds that are captures allows prune to be called # using wrapped_func.inputs even when it uses variables internal_captures = {id(c) for c in self.graph.internal_captures} flat_feeds = [f for f in flat_feeds if id(f) not in internal_captures] operation_fetches = [] tensor_fetches = [] tensor_infos = [] def _fetch_preprocessing_callback(fetch): """Extract out lists of ops, tensors, and tensor type info. Turns TensorInfos into Tensors in the original `fetches` structure. Also extracts ops from `fetches`. Args: fetch: The fetch to preprocess: Tensor, TensorInfo, or Operation, or string identifying a Tensor or Operation. Returns: `fetch` converted to a Tensor. """ if isinstance(fetch, ops.Operation): operation_fetches.append(fetch) return fetch elif isinstance(fetch, meta_graph_pb2.TensorInfo): tensor_infos.append(fetch) decoded = _get_element_from_tensor_info(fetch, self._func_graph) if (tensor_util.is_tf_type(decoded) or isinstance(decoded, composite_tensor.CompositeTensor)): tensor_fetches.append(decoded) else: operation_fetches.append(decoded) return decoded elif isinstance( fetch, (tensor_lib.Tensor, composite_tensor.CompositeTensor)): tensor_fetches.append(fetch) return fetch else: graph_element = self.graph.as_graph_element(fetch) return _fetch_preprocessing_callback(graph_element) fetches = nest.map_structure(_fetch_preprocessing_callback, fetches) # Expand composite tensors into their component dense Tensors. tensor_fetches = nest.flatten(tensor_fetches, expand_composites=True) for f in flat_feeds + tensor_fetches + operation_fetches: if f.graph is not self._func_graph: raise ValueError("Can only prune function whose feeds and fetches " f"from graph {self._func_graph}. Input " f"{f} is from a different graph {f.graph}.") with self._func_graph.as_default(): pruned_graph = func_graph.FuncGraph(name) lift_map = lift_to_graph.lift_to_graph( operation_fetches + tensor_fetches, pruned_graph, sources=flat_feeds + self.graph.internal_captures, base_graph=self._func_graph) # Note that we add the component tensors of any composite tensors to the # returned function's outputs list; the list must contain these component # tensors, or the function's sparse outputs won't work properly. pruned_graph.outputs.extend(lift_map[x] for x in tensor_fetches) pruned_graph.control_outputs.extend( [lift_map[operation] for operation in operation_fetches]) pruned_graph.inputs.extend(lift_map[x] for x in flat_feeds) for external_capture, internal_capture in self.graph.captures: pruned_graph.add_capture(external_capture, lift_map[internal_capture]) for ti in tensor_infos: if ti.WhichOneof("encoding") == "name": # Dense tensors only t = pruned_graph.as_graph_element(ti.name) if tensor_util.is_tf_type(t): t.set_shape(tensor_shape.TensorShape(ti.tensor_shape)) # pylint: disable=protected-access for f in self.graph._functions.values(): pruned_graph._add_function(f) # pylint: enable=protected-access pruned_graph.variables = self.graph.variables def _structured_output_mapping(fetched): """callback for `nest.map_structure()`""" lifted = lift_map[fetched] if isinstance(lifted, ops.Operation): return None return lifted # expand_composites=True here causes composite tensors to be expanded # into their component dense Tensors, mapped to the new graph, and then # reconstituted into their original composite form. pruned_graph.structured_outputs = nest.map_structure( _structured_output_mapping, fetches, expand_composites=True) if input_signature: # canonicalize the signature before setting args, kwargs = input_signature args = () if args is None else args input_signature = (args, kwargs) pruned_graph.structured_input_signature = input_signature pruned_fn = WrappedFunction( pruned_graph, variable_holder=self._variable_holder) pruned_fn._num_positional_args = len(flat_feeds) # pylint: disable=protected-access # TODO(kathywu): Enable keyword arguments if an input signature is specified pruned_fn._arg_keywords = [tensor.op.name for tensor in flat_feeds] # pylint: disable=protected-access return pruned_fn def _filter_returned_ops(fn): """Filtering out any ops returned by function. Args: fn: a function Returns: A tuple of ( Wrapped function that returns `None` in place of any ops, dict that maps the index in the flat output structure to the returned op ) """ returned_ops = {} def wrap_and_filter_returned_ops(*args, **kwargs): outputs = fn(*args, **kwargs) flat_outputs = nest.flatten(outputs) for n in range(len(flat_outputs)): output = flat_outputs[n] if isinstance(output, ops.Operation): returned_ops[n] = output flat_outputs[n] = None return nest.pack_sequence_as(outputs, flat_outputs) return wrap_and_filter_returned_ops, returned_ops class WrappedGraph(object): """Class for wrapping multiple TF 1.X functions in a single graph. Maintains a dictionary mapping names to wrapped functions. See `tf.compat.v1.wrap_function` to learn more about wrapping V1 functions. Functions wrapped using this class have access to variables and collections created in other wrapped functions, using the standard TF 1.X API ( `tf.compat.v1.get_variable` or `tf.compat.v1.get_default_graph().get_collection(...)`) Outside a function, variables and collections may be accessed using the `variables` and `graph` properties. Example: ``` def add_v1(x): with tf.compat.v1.variable_scope('vars', reuse=tf.compat.v1.AUTO_REUSE): v = tf.compat.v1.get_variable('v', shape=[], dtype=tf.int32) return v + x def increment_var_v1(x): with tf.compat.v1.variable_scope('vars', reuse=tf.compat.v1.AUTO_REUSE): v = tf.compat.v1.get_variable('v', shape=[], dtype=tf.int32) return v.assign_add(x) g = WrappedGraph() add = g.wrap_function(add_v1, [tf.TensorSpec([], tf.int32)]) increment_var = g.wrap_function(increment_var_v1, [tf.TensorSpec([], tf.int32)]) assert len(g.variables) == 1 assert g.variables[0].numpy() == 0 increment_var(tf.constant(5)) assert g.variables[0].numpy() == 5 ``` """ def __init__(self, variable_holder=None, **kwargs): self._variable_holder = ( variable_holder or VariableHolder(share_variables=True)) name = kwargs.pop("name", "wrapped_function_graph") # Always start with empty collections, unless otherwise specified. Setting # `collections=None` will copy the collections from the outer graph. collections = kwargs.pop("collections", {}) self.graph = func_graph.FuncGraph(name, collections=collections, **kwargs) self._wrapped_function = WrappedFunction(self.graph, self._variable_holder) self._functions = {} @property def functions(self): return self._functions @property def variables(self): return self._variable_holder.variables def wrap_function(self, fn, signature, name=None): """Wraps a TF 1.X function and returns an eager-compatible function. All functions wrapped in the same `WrappedGraph` will have access to the same graph (`tf.compat.v1.get_default_graph` to get the graph object within a function, or `WrappedGraph.graph` to get the graph outside a function). Variables created within the function will be added to the `variables` list. Function inputs: All inputs to the function must be tensors (nested ok), with their shapes and dtypes defined in the `signature` argument. Function outputs: * The 1.X function may return tensors, variables, and ops. The wrapped eager-compatible function will always return tensors in the same nested structure. * Variables are replaced with a tensor containing the latest read values. * Returned ops are executed, and replaced with None. * The order of op execution and variable reads in the return is nondeterministic. For example: ``` def update_var(x): v = tf.Variable(0) op = tf.compat.v1.assign(v, x).op return v, op g = WrappedGraph() fn = g.wrap_function(update_var) read_value, _ = fn(tf.constant(3)) print(read_value.numpy()) # could be 0 or 3 print(g.variables[0].numpy()) # always 3 ``` To ensure that ops in the function are executed (e.g. ops added to the `tf.GraphKeys.UPDATE_OPS` collection), include them in the function returns. Args: fn: a 1.X tensorflow function. signature: a possibly nested sequence of `TensorSpecs` specifying the shapes and dtypes of the arguments. name: an optional string name for the function. The function will be saved with key `name` in the `functions` dictionary. Returns: An eager-compatible function. """ return self._wrap_function(fn, signature=signature, name=name) def _wrap_function(self, fn, args=None, kwargs=None, signature=None, name=None): """Internal wrap function method with extended func_graph arguments.""" fn_with_filter_and_scope, returned_ops = _filter_returned_ops( self._variable_holder.call_with_variable_creator_scope(fn)) func_graph.func_graph_from_py_func( None, # Name is unused. fn_with_filter_and_scope, args=args, kwargs=kwargs, signature=signature, add_control_dependencies=False, func_graph=self.graph) # This code relies on questional behavior from `func_graph_from_py_func`. # If an existing FuncGraph is passed into the `func_graph` arg, the inputs # and structured outputs are overwritten. Pretty sure this is a bug, # because structured outputs doesn't match up with the outputs... fn_inputs = self.graph.inputs[:-len(self.graph.captures)] # Return filtered ops to the flattened outputs. flat_fn_outputs = nest.flatten(self.graph.structured_outputs) for index, op in returned_ops.items(): flat_fn_outputs[index] = op fn_outputs = nest.pack_sequence_as(self.graph.structured_outputs, flat_fn_outputs) name = name or fn.__name__ wrapped_function = self._wrapped_function.prune( fn_inputs, fn_outputs, name, self.graph.structured_input_signature) self._functions[name] = wrapped_function return wrapped_function @tf_export(v1=["wrap_function"]) def wrap_function(fn, signature, name=None): """Wraps the TF 1.x function fn into a graph function. The python function `fn` will be called once with symbolic arguments specified in the `signature`, traced, and turned into a graph function. Any variables created by `fn` will be owned by the object returned by `wrap_function`. The resulting graph function can be called with tensors which match the signature. ```python def f(x, do_add): v = tf.Variable(5.0) if do_add: op = v.assign_add(x) else: op = v.assign_sub(x) with tf.control_dependencies([op]): return v.read_value() f_add = tf.compat.v1.wrap_function(f, [tf.TensorSpec((), tf.float32), True]) assert float(f_add(1.0)) == 6.0 assert float(f_add(1.0)) == 7.0 # Can call tf.compat.v1.wrap_function again to get a new trace, a new set # of variables, and possibly different non-template arguments. f_sub= tf.compat.v1.wrap_function(f, [tf.TensorSpec((), tf.float32), False]) assert float(f_sub(1.0)) == 4.0 assert float(f_sub(1.0)) == 3.0 ``` Both `tf.compat.v1.wrap_function` and `tf.function` create a callable TensorFlow graph. But while `tf.function` runs all stateful operations (e.g. `tf.print`) and sequences operations to provide the same semantics as eager execution, `wrap_function` is closer to the behavior of `session.run` in TensorFlow 1.x. It will not run any operations unless they are required to compute the function's outputs, either through a data dependency or a control dependency. Nor will it sequence operations. Unlike `tf.function`, `wrap_function` will only trace the Python function once. As with placeholders in TF 1.x, shapes and dtypes must be provided to `wrap_function`'s `signature` argument. Since it is only traced once, variables and state may be created inside the function and owned by the function wrapper object. Args: fn: python function to be wrapped signature: the placeholder and python arguments to be passed to the wrapped function name: Optional. The name of the function. Returns: the wrapped graph function. """ holder = VariableHolder(fn) func_graph_name = "wrapped_function" if name is not None: func_graph_name = "wrapped_function_" + name return WrappedFunction( func_graph.func_graph_from_py_func( func_graph_name, holder, args=None, kwargs=None, signature=signature, add_control_dependencies=False, collections={}), variable_holder=holder, signature=signature) def function_from_graph_def(graph_def, inputs, outputs, captures=None): """Creates a ConcreteFunction from a GraphDef. Args: graph_def: A GraphDef to make a function out of. inputs: A Tensor name or nested structure of names in `graph_def` which should be inputs to the function. outputs: A Tensor name or nested structure of names in `graph_def` which should be outputs of the function. captures: (Optional) A dictionary mapping node names in `graph_def` that should be captured as inputs to tensors containing the value of the captured inputs. Returns: A ConcreteFunction. """ def _imports_graph_def(): importer.import_graph_def(graph_def, name="") graph = ops.get_default_graph() if captures is not None: for c in captures: graph.add_capture(captures[c], graph.get_tensor_by_name(str(c) + ":0")) wrapped_import = wrap_function(_imports_graph_def, []) import_graph = wrapped_import.graph return wrapped_import.prune( nest.map_structure(import_graph.as_graph_element, inputs), nest.map_structure(import_graph.as_graph_element, outputs))