I am producing a iterative process via tff.learning.build_federated_averaging_process(). and receive the error:
Traceback (most recent call last):
File "B:\tools and software\Anaconda\envs\bookProjects\lib\site-packages\IPython\core\interactiveshell.py", line 3331, in run_code
exec(code_obj, self.user_global_ns, self.user_ns)
File "<ipython-input-2-47998fd56829>", line 1, in <module>
runfile('B:/projects/openProjects/githubprojects/BotnetTrafficAnalysisFederaedLearning/anomaly-detection/train_v04.py', args=['--experiment_name=temp', '--client_batch_size=20', '--client_optimizer=sgd', '--client_learning_rate=0.2', '--server_optimizer=sgd', '--server_learning_rate=1.0', '--total_rounds=200', '--rounds_per_eval=1', '--rounds_per_checkpoint=50', '--rounds_per_profile=0', '--root_output_dir=B:/projects/openProjects/githubprojects/BotnetTrafficAnalysisFederaedLearning/anomaly-detection/logs/fed_out/'], wdir='B:/projects/openProjects/githubprojects/BotnetTrafficAnalysisFederaedLearning/anomaly-detection')
File "B:\tools and software\PyCharm 2020.1\plugins\python\helpers\pydev\_pydev_bundle\pydev_umd.py", line 197, in runfile
pydev_imports.execfile(filename, global_vars, local_vars) # execute the script
File "B:\tools and software\PyCharm 2020.1\plugins\python\helpers\pydev\_pydev_imps\_pydev_execfile.py", line 18, in execfile
exec(compile(contents+"\n", file, 'exec'), glob, loc)
File "B:/projects/openProjects/githubprojects/BotnetTrafficAnalysisFederaedLearning/anomaly-detection/train_v04.py", line 306, in <module>
app.run(main)
File "B:\tools and software\Anaconda\envs\bookProjects\lib\site-packages\absl\app.py", line 299, in run
_run_main(main, args)
File "B:\tools and software\Anaconda\envs\bookProjects\lib\site-packages\absl\app.py", line 250, in _run_main
sys.exit(main(argv))
File "B:/projects/openProjects/githubprojects/BotnetTrafficAnalysisFederaedLearning/anomaly-detection/train_v04.py", line 299, in main
train_main()
File "B:/projects/openProjects/githubprojects/BotnetTrafficAnalysisFederaedLearning/anomaly-detection/train_v04.py", line 262, in train_main
server_optimizer_fn=server_optimizer_fn,
File "B:\tools and software\Anaconda\envs\bookProjects\lib\site-packages\tensorflow_federated\python\learning\federated_averaging.py", line 211, in build_federated_averaging_process
stateful_delta_aggregate_fn, stateful_model_broadcast_fn)
File "B:\tools and software\Anaconda\envs\bookProjects\lib\site-packages\tensorflow_federated\python\learning\framework\optimizer_utils.py", line 498, in build_model_delta_optimizer_process
py_typecheck.check_callable(model_fn)
File "B:\tools and software\Anaconda\envs\bookProjects\lib\site-packages\tensorflow_federated\python\common_libs\py_typecheck.py", line 106, in check_callable
type_string(type(target))))
TypeError: Expected a callable, found non-callable tensorflow_federated.python.learning.model_utils.EnhancedModel.
in build_federated_averaging_process
stateful_delta_aggregate_fn, stateful_model_broadcast_fn)
and
TypeError: Expected a callable, found non-callable tensorflow_federated.python.learning.model_utils.EnhancedModel.
model_fn=Callable[[], model_fn]
only creates a new error.iterative process:
model_fn = model_builder(input_dim=sysarg,
input_spec=input_spec)
iterative_process = tff.learning.build_federated_averaging_process(
model_fn=model_fn,
client_optimizer_fn=client_optimizer_fn,
server_optimizer_fn=server_optimizer_fn,
)
iterative_process = compression_process_adapter.CompressionProcessAdapter(iterative_process)```
model bulder:
def model_builder(input_dim, input_spec):
model = create_model(input_dim)
return tff.learning.from_keras_model(keras_model=model,
loss=tf.keras.losses.MeanSquaredError(),
input_spec=input_spec,
metrics=[tf.keras.metrics.Accuracy()],
)
def create_model(input_dim):
autoencoder = Sequential([
tf.keras.layers.Dense(int(0.75 * input_dim), activation="tanh", input_shape=(input_dim,)),
tf.keras.layers.Dense(int(0.5 * input_dim), activation="tanh"),
tf.keras.layers.Dense(int(0.33 * input_dim), activation="tanh"),
tf.keras.layers.Dense(int(0.25 * input_dim), activation="tanh"),
tf.keras.layers.Dense(int(0.33 * input_dim), activation="tanh"),
tf.keras.layers.Dense(int(0.5 * input_dim), activation="tanh"),
tf.keras.layers.Dense(int(0.75 * input_dim), activation="tanh"),
tf.keras.layers.Dense(input_dim)
])
The model_fn argument of tff.learning.build_federated_averaging_process needs to be a callable (What is a callable?) that takes no arguments and returns a tff.learning.Model.
From the code (reproduced here for readability):
def model_builder(input_dim, input_spec):
model = create_model(input_dim)
return tff.learning.from_keras_model(
keras_model=model,
loss=tf.keras.losses.MeanSquaredError(),
input_spec=input_spec,
metrics=[tf.keras.metrics.Accuracy()])
model_fn = model_builder(input_dim=sysarg, input_spec=input_spec)
iterative_process = tff.learning.build_federated_averaging_process(
model_fn=model_fn,
client_optimizer_fn=client_optimizer_fn,
server_optimizer_fn=server_optimizer_fn)
model_fn is actually an instance of tff.learning.Model, rather than a callable that constructs and returns a model.
model_builder should be passed to tff.learning.build_federated_averaging_process, but since it takes arguments this won't work as-is.
One alternative would be to use functools.partial:
iterative_process = tff.learning.build_federated_averaging_process(
model_fn=functools.partial(model_builder, input_dim=sysarg, input_spec=input_spec),
client_optimizer_fn=client_optimizer_fn,
server_optimizer_fn=server_optimizer_fn)
Or even use a no-arg lambda:
iterative_process = tff.learning.build_federated_averaging_process(
model_fn=lambda: model_builder(input_dim=sysarg, input_spec=input_spec),
client_optimizer_fn=client_optimizer_fn,
server_optimizer_fn=server_optimizer_fn)
The following question has a good discussion on the differences between lambdas and partials: Python: Why is functools.partial necessary?