我试图用我自己的Keras基于CNN的模型重新实现the github tutorial。但是评估时出现错误。
from __future__ import absolute_import, division, print_function
import collections
from six.moves import range
import numpy as np
import tensorflow as tf
from tensorflow.python.keras.optimizer_v2 import gradient_descent
from tensorflow_federated import python as tff
emnist_train, emnist_test = tff.simulation.datasets.emnist.load_data()
example_dataset = emnist_train.create_tf_dataset_for_client(
emnist_train.client_ids[0])
NUM_EPOCHS = 10
BATCH_SIZE = 20
SHUFFLE_BUFFER = 500
def preprocess(dataset):
def element_fn(element):
return collections.OrderedDict([
('x', tf.reshape(element['pixels'], [-1])),
('y', tf.reshape(element['label'], [1])),
])
return dataset.repeat(NUM_EPOCHS).map(element_fn).shuffle(
SHUFFLE_BUFFER).batch(BATCH_SIZE)
preprocessed_example_dataset = preprocess(example_dataset)
sample_batch = nest.map_structure(
lambda x: x.numpy(), iter(preprocessed_example_dataset).next())
def make_federated_data(client_data, client_ids):
return [preprocess(client_data.create_tf_dataset_for_client(x))
for x in client_ids]
NUM_CLIENTS = 3
sample_clients = emnist_train.client_ids[0:NUM_CLIENTS]
federated_train_data = make_federated_data(emnist_train, sample_clients)
len(federated_train_data), federated_train_data[0]
def create_compiled_keras_model():
model = tf.keras.models.Sequential([
tf.keras.layers.Reshape((28,28,1), input_shape=(784,)),
tf.keras.layers.Conv2D(32, kernel_size=(5,5), activation="relu", padding = "same", strides = 1),
tf.keras.layers.MaxPooling2D(pool_size=2, strides=2, padding='valid'),
tf.keras.layers.Conv2D(64, kernel_size=(5,5), activation="relu", padding = "same", strides = 1),
tf.keras.layers.MaxPooling2D(pool_size=2, strides=2, padding='valid'),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(512, activation="relu"),
tf.keras.layers.Dense(10, activation="softmax"),
])
def loss_fn(y_true, y_pred):
return tf.reduce_mean(tf.keras.losses.sparse_categorical_crossentropy(
y_true, y_pred))
model.compile(
loss=loss_fn,
optimizer=gradient_descent.SGD(learning_rate=0.02),
metrics=[tf.keras.metrics.SparseCategoricalAccuracy()])
return model
def model_fn():
keras_model = create_compiled_keras_model()
return tff.learning.from_compiled_keras_model(keras_model, sample_batch)
iterative_process = tff.learning.build_federated_averaging_process(model_fn)
state = iterative_process.initialize()
for round_num in range(1,10):
state, metrics = iterative_process.next(state, federated_train_data)
print('round {:2d}, metrics={}'.format(round_num, metrics))
##Evaluation of the model
#This function doesn't work
evaluation = tff.learning.build_federated_evaluation(model_fn)
federated_test_data = make_federated_data(emnist_test, sample_clients)
test_metrics = evaluation(state.model, federated_test_data)
我希望评估测试数据，但实际输出是以下错误：
---------------------------------------------------------------------------
_FallbackException Traceback (most recent call last)
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/ops/gen_functional_ops.py in stateful_partitioned_call(args, Tout, f, config, config_proto, executor_type, name)
482 "Tout", Tout, "f", f, "config", config, "config_proto", config_proto,
--> 483 "executor_type", executor_type)
484 return _result
_FallbackException: This function does not handle the case of the path where all inputs are not already EagerTensors.
During handling of the above exception, another exception occurred:
AttributeError Traceback (most recent call last)
<ipython-input-23-6e9c77f70201> in <module>()
----> 1 evaluation = tff.learning.build_federated_evaluation(model_fn)
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/learning/federated_evaluation.py in build_federated_evaluation(model_fn)
83 @tff.federated_computation(
84 tff.FederatedType(model_weights_type, tff.SERVER, all_equal=True),
---> 85 tff.FederatedType(tff.SequenceType(batch_type), tff.CLIENTS))
86 def server_eval(server_model_weights, federated_dataset):
87 client_outputs = tff.federated_map(
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/core/impl/computation_wrapper.py in <lambda>(fn)
406 args = (args,)
407 arg_type = computation_types.to_type(args[0])
--> 408 return lambda fn: _wrap(fn, arg_type, self._wrapper_fn)
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/core/impl/computation_wrapper.py in _wrap(fn, parameter_type, wrapper_fn)
94 function_utils.wrap_as_zero_or_one_arg_callable(fn, parameter_type),
95 parameter_type,
---> 96 name=fn_name)
97 py_typecheck.check_type(concrete_fn, function_utils.ConcreteFunction,
98 'value returned by the wrapper')
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/core/impl/computation_wrapper_instances.py in _federated_computation_wrapper_fn(target_fn, parameter_type, name)
52 parameter_type,
53 ctx_stack,
---> 54 suggested_name=name))
55 return computation_impl.ComputationImpl(target_lambda.proto, ctx_stack)
56
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/core/impl/federated_computation_utils.py in zero_or_one_arg_fn_to_building_block(fn, parameter_name, parameter_type, context_stack, suggested_name)
73 value_impl.ValueImpl(
74 computation_building_blocks.Reference(
---> 75 parameter_name, parameter_type), context_stack))
76 else:
77 result = fn()
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/core/impl/function_utils.py in <lambda>(arg)
551 # and to force any parameter bindings to be resolved now.
552 # pylint: disable=unnecessary-lambda,undefined-variable
--> 553 return (lambda fn, at, kt: lambda arg: _unpack_and_call(fn, at, kt, arg))(
554 fn, arg_types, kwarg_types)
555 # pylint: enable=unnecessary-lambda,undefined-variable
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/core/impl/function_utils.py in _unpack_and_call(fn, arg_types, kwarg_types, arg)
545 name, str(expected_type), str(actual_type)))
546 kwargs[name] = element_value
--> 547 return fn(*args, **kwargs)
548
549 # Deliberate wrapping to isolate the caller from the underlying function
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/learning/federated_evaluation.py in server_eval(server_model_weights, federated_dataset)
88 client_eval,
89 [tff.federated_broadcast(server_model_weights), federated_dataset])
---> 90 return model.federated_output_computation(client_outputs.local_outputs)
91
92 return server_eval
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/learning/model_utils.py in federated_output_computation(self)
531 @property
532 def federated_output_computation(self):
--> 533 return self._model.federated_output_computation
534
535
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow_federated/python/learning/model_utils.py in federated_output_computation(self)
406 def federated_output_computation(self):
407 metric_variable_type_dict = nest.map_structure(tf.TensorSpec.from_tensor,
--> 408 self.report_local_outputs())
409 federated_local_outputs_type = tff.FederatedType(
410 metric_variable_type_dict, tff.CLIENTS, all_equal=False)
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/eager/def_function.py in __call__(self, *args, **kwds)
314 if not self._created_variables:
315 # If we did not create any variables the trace we have is good enough.
--> 316 return self._concrete_stateful_fn._filtered_call(canon_args, canon_kwds) # pylint: disable=protected-access
317
318 def fn_with_cond(*inner_args, **inner_kwds):
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/eager/function.py in _filtered_call(self, args, kwargs)
382 """
383 return self._call_flat(
--> 384 (t for t in nest.flatten((args, kwargs))
385 if isinstance(
386 t, (ops.Tensor, resource_variable_ops.ResourceVariable))))
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/eager/function.py in _call_flat(self, args)
431 # Only need to override the gradient in graph mode and when we have outputs.
432 if context.executing_eagerly() or not self.outputs:
--> 433 outputs = self._inference_function.call(ctx, args)
434 else:
435 if not self._gradient_name:
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/eager/function.py in call(self, ctx, args)
267 executing_eagerly=executing_eagerly,
268 config=function_call_options.config_proto_serialized,
--> 269 executor_type=function_call_options.executor_type)
270
271 if executing_eagerly:
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/ops/functional_ops.py in partitioned_call(args, f, tout, executing_eagerly, config, executor_type)
1081 outputs = gen_functional_ops.stateful_partitioned_call(
1082 args=args, Tout=tout, f=f, config_proto=config,
-> 1083 executor_type=executor_type)
1084 else:
1085 outputs = gen_functional_ops.partitioned_call(
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/ops/gen_functional_ops.py in stateful_partitioned_call(args, Tout, f, config, config_proto, executor_type, name)
487 return stateful_partitioned_call_eager_fallback(
488 args, Tout=Tout, f=f, config=config, config_proto=config_proto,
--> 489 executor_type=executor_type, name=name, ctx=_ctx)
490 except _core._SymbolicException:
491 pass # Add nodes to the TensorFlow graph.
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/ops/gen_functional_ops.py in stateful_partitioned_call_eager_fallback(args, Tout, f, config, config_proto, executor_type, name, ctx)
548 executor_type = ""
549 executor_type = _execute.make_str(executor_type, "executor_type")
--> 550 _attr_Tin, args = _execute.convert_to_mixed_eager_tensors(args, _ctx)
551 _inputs_flat = list(args)
552 _attrs = ("Tin", _attr_Tin, "Tout", Tout, "f", f, "config", config,
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/eager/execute.py in convert_to_mixed_eager_tensors(values, ctx)
207 def convert_to_mixed_eager_tensors(values, ctx):
208 v = [ops.internal_convert_to_tensor(t, ctx=ctx) for t in values]
--> 209 types = [t._datatype_enum() for t in v] # pylint: disable=protected-access
210 return types, v
211
/Library/Frameworks/Python.framework/Versions/3.6/lib/python3.6/site-packages/tensorflow/python/eager/execute.py in <listcomp>(.0)
207 def convert_to_mixed_eager_tensors(values, ctx):
208 v = [ops.internal_convert_to_tensor(t, ctx=ctx) for t in values]
--> 209 types = [t._datatype_enum() for t in v] # pylint: disable=protected-access
210 return types, v
211
AttributeError: 'Tensor' object has no attribute '_datatype_enum'
答案 0 :(得分：1)
努里亚：应该只是fixed earlier today。如果您不想等待下一个版本（即将推出），我建议您仅从源代码构建本地pip包。您可以找到说明in the install guide。
答案 1 :(得分：1)
作为后续操作，这里发布了TFF 0.4.0，其中包含此错误修正。