缺少必需的位置参数：

Question

我尝试基于 LSTM 方法实现联邦学习。

def create_keras_model():
    model = Sequential()
    model.add(LSTM(32, input_shape=(3,1)))
    model.add(Dense(1))
    return model

def model_fn():
    keras_model = create_keras_model()
    return tff.learning.from_keras_model(
      keras_model,
      input_spec=(look_back, 1),
      loss=tf.keras.losses.mean_squared_error(),
      metrics=[tf.keras.metrics.mean_squared_error()])

但是当我想定义 iterative_process 时出现此错误。

iterative_process = tff.learning.build_federated_averaging_process(
    model_fn,
    client_optimizer_fn=lambda: tf.keras.optimizers.SGD(learning_rate=0.001),
    server_optimizer_fn=lambda: tf.keras.optimizers.SGD(learning_rate=1.0))

TypeError: Missing required positional argument

我如何解决它？

Answer 1

提供的与环回参数匹配的输入要求可以由客户端训练数据要求代替。 ( TensorSpec )联合

您可以通过不同类型的输入参数并行工作。

[ 样本 ]：

import tensorflow as tf
import tensorflow_federated as tff

# Load simulation data.
source, _ = tff.simulation.datasets.emnist.load_data()

def client_data(n):
    return source.create_tf_dataset_for_client(source.client_ids[n]).map(
    lambda e: (tf.reshape(e['pixels'], [-1]), e['label'])
    ).repeat(10).batch(20)

train_data = [client_data(n) for n in range(3)]


def create_keras_model():
    model = tf.keras.models.Sequential([ ])
    model.add(tf.keras.layers.InputLayer(input_shape=( 784 )))
    model.add(tf.keras.layers.Reshape((784, 1)))
    model.add(tf.keras.layers.LSTM(32))
    model.add(tf.keras.layers.Dense(1))
    return model

def model_fn():
    keras_model = create_keras_model()
    return tff.learning.from_keras_model( keras_model, 
    input_spec=train_data[0].element_spec,
    loss=tf.keras.losses.MeanSquaredError(reduction=tf.keras.losses.Reduction.AUTO), 
    metrics=[tf.keras.metrics.MeanSquaredError( ), tf.keras.metrics.Accuracy( )])

# Simulate a few rounds of training with the selected client devices.
trainer = tff.learning.build_federated_averaging_process( model_fn, client_optimizer_fn=lambda : tf.keras.optimizers.SGD(0.1),
    server_optimizer_fn=lambda : tf.keras.optimizers.SGD(1.0))

state = trainer.initialize()
for _ in range(50):
    state, metrics = trainer.next(state, train_data)
    #print(metrics['train']['loss'])
    print(metrics)

[ 输出 ]：

OrderedDict([('broadcast', ()), ('aggregation', OrderedDict([('value_sum_process', ()), ('weight_sum_process', ())])), ('train', OrderedDict([('mean_squared_error', 8.502816), ('accuracy', 0.0), ('loss', 8.5030365)]))])
OrderedDict([('broadcast', ()), ('aggregation', OrderedDict([('value_sum_process', ()), ('weight_sum_process', ())])), ('train', OrderedDict([('mean_squared_error', 8.500688), ('accuracy', 0.0), ('loss', 8.500914)]))])
OrderedDict([('broadcast', ()), ('aggregation', OrderedDict([('value_sum_process', ()), ('weight_sum_process', ())])), ('train', OrderedDict([('mean_squared_error', 8.498711), ('accuracy', 0.0), ('loss', 8.498943)]))])