Use otimizadores TFF no processo iterativo personalizado

Ver no TensorFlow.org Executar no Google Colab Ver fonte no GitHub Baixar caderno

Esta é uma alternativa para a construir o seu próprio Federated Aprendizagem Algoritmo tutorial eo simple_fedavg exemplo para construir um processo iterativo personalizado para o cálculo da média federado algoritmo. Este tutorial vai usar otimizadores TFF em vez de otimizadores Keras. A abstração do otimizador TFF foi projetada para ser state-in-state-out para ser mais fácil de ser incorporada em um processo iterativo TFF. Os tff.learning APIs também aceitam otimizadores TFF como argumento de entrada.

Antes de começarmos

Antes de começar, execute o seguinte para se certificar de que seu ambiente está configurado corretamente. Se você não vê uma saudação, consulte a instalação guia para obter instruções.

!pip install --quiet --upgrade tensorflow-federated-nightly
!pip install --quiet --upgrade nest-asyncio

import nest_asyncio
nest_asyncio.apply()
import functools
import attr
import numpy as np
import tensorflow as tf
import tensorflow_federated as tff

Preparando dados e modelo

O processamento de dados e EMNIST modelo são muito semelhantes ao simple_fedavg exemplo.

only_digits=True

# Load dataset.
emnist_train, emnist_test = tff.simulation.datasets.emnist.load_data(only_digits)

# Define preprocessing functions.
def preprocess_fn(dataset, batch_size=16):

  def batch_format_fn(element):
    return (tf.expand_dims(element['pixels'], -1), element['label'])

  return dataset.batch(batch_size).map(batch_format_fn)

# Preprocess and sample clients for prototyping.
train_client_ids = sorted(emnist_train.client_ids)
train_data = emnist_train.preprocess(preprocess_fn)
central_test_data = preprocess_fn(
    emnist_train.create_tf_dataset_for_client(train_client_ids[0]))

# Define model.
def create_keras_model():
  """The CNN model used in https://arxiv.org/abs/1602.05629."""
  data_format = 'channels_last'
  input_shape = [28, 28, 1]

  max_pool = functools.partial(
      tf.keras.layers.MaxPooling2D,
      pool_size=(2, 2),
      padding='same',
      data_format=data_format)
  conv2d = functools.partial(
      tf.keras.layers.Conv2D,
      kernel_size=5,
      padding='same',
      data_format=data_format,
      activation=tf.nn.relu)

  model = tf.keras.models.Sequential([
      conv2d(filters=32, input_shape=input_shape),
      max_pool(),
      conv2d(filters=64),
      max_pool(),
      tf.keras.layers.Flatten(),
      tf.keras.layers.Dense(512, activation=tf.nn.relu),
      tf.keras.layers.Dense(10 if only_digits else 62),
  ])

  return model

# Wrap as `tff.learning.Model`.
def model_fn():
  keras_model = create_keras_model()
  return tff.learning.from_keras_model(
      keras_model,
      input_spec=central_test_data.element_spec,
      loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True))

Processo iterativo personalizado

Em muitos casos, os algoritmos federados têm 4 componentes principais:

  1. Uma etapa de transmissão de servidor para cliente.
  2. Uma etapa de atualização do cliente local.
  3. Uma etapa de upload de cliente para servidor.
  4. Uma etapa de atualização do servidor.

Em TFF, que geralmente representam algoritmos federados como tff.templates.IterativeProcess (aos quais nos referimos como apenas um IterativeProcess em todo). Esta é uma classe que contém initialize e next funções. Aqui, initialize é usado para inicializar o servidor, e next realizará uma rodada de comunicação do algoritmo federado.

Apresentaremos diferentes componentes para construir o algoritmo de média federada (FedAvg), que usará um otimizador na etapa de atualização do cliente e outro otimizador na etapa de atualização do servidor. As lógicas básicas de atualizações de cliente e servidor podem ser expressas como blocos TF puros.

Blocos TF: atualização de cliente e servidor

Em cada cliente, um local de client_optimizer é inicializado e usado para atualizar os pesos modelo cliente. No servidor, server_optimizer usará o estado da ronda anterior, e atualizar o estado para a próxima rodada.

@tf.function
def client_update(model, dataset, server_weights, client_optimizer):
  """Performs local training on the client's dataset."""
  # Initialize the client model with the current server weights.
  client_weights = model.trainable_variables
  # Assign the server weights to the client model.
  tf.nest.map_structure(lambda x, y: x.assign(y),
                        client_weights, server_weights)
  # Initialize the client optimizer.
  trainable_tensor_specs = tf.nest.map_structure(
          lambda v: tf.TensorSpec(v.shape, v.dtype), client_weights)
  optimizer_state = client_optimizer.initialize(trainable_tensor_specs)
  # Use the client_optimizer to update the local model.
  for batch in iter(dataset):
    with tf.GradientTape() as tape:
      # Compute a forward pass on the batch of data.
      outputs = model.forward_pass(batch)
    # Compute the corresponding gradient.
    grads = tape.gradient(outputs.loss, client_weights)
    # Apply the gradient using a client optimizer.
    optimizer_state, updated_weights = client_optimizer.next(
        optimizer_state, client_weights, grads)
    tf.nest.map_structure(lambda a, b: a.assign(b), 
                          client_weights, updated_weights)
  # Return model deltas.
  return tf.nest.map_structure(tf.subtract, client_weights, server_weights)
@attr.s(eq=False, frozen=True, slots=True)
class ServerState(object):
  trainable_weights = attr.ib()
  optimizer_state = attr.ib()

@tf.function
def server_update(server_state, mean_model_delta, server_optimizer):
  """Updates the server model weights."""
  # Use aggregated negative model delta as pseudo gradient. 
  negative_weights_delta = tf.nest.map_structure(
      lambda w: -1.0 * w, mean_model_delta)
  new_optimizer_state, updated_weights = server_optimizer.next(
      server_state.optimizer_state, server_state.trainable_weights, 
      negative_weights_delta)
  return tff.structure.update_struct(
      server_state,
      trainable_weights=updated_weights,
      optimizer_state=new_optimizer_state)

Blocos TFF: tff.tf_computation e tff.federated_computation

Agora usamos TFF para orquestração e construímos o processo iterativo para FedAvg. Temos que quebrar os blocos TF definidos acima, com tff.tf_computation e métodos de uso TFF tff.federated_broadcast , tff.federated_map , tff.federated_mean em um tff.federated_computation função. É fácil usar o tff.learning.optimizers.Optimizer APIs com initialize e next funções ao definir um processo iterativo personalizado.

# 1. Server and client optimizer to be used.
server_optimizer = tff.learning.optimizers.build_sgdm(
    learning_rate=0.05, momentum=0.9)
client_optimizer = tff.learning.optimizers.build_sgdm(
    learning_rate=0.01)

# 2. Functions return initial state on server. 
@tff.tf_computation
def server_init():
  model = model_fn()
  trainable_tensor_specs = tf.nest.map_structure(
        lambda v: tf.TensorSpec(v.shape, v.dtype), model.trainable_variables)
  optimizer_state = server_optimizer.initialize(trainable_tensor_specs)
  return ServerState(
      trainable_weights=model.trainable_variables,
      optimizer_state=optimizer_state)

@tff.federated_computation
def server_init_tff():
  return tff.federated_value(server_init(), tff.SERVER)

# 3. One round of computation and communication.
server_state_type = server_init.type_signature.result
print('server_state_type:\n', 
      server_state_type.formatted_representation())
trainable_weights_type = server_state_type.trainable_weights
print('trainable_weights_type:\n', 
      trainable_weights_type.formatted_representation())

# 3-1. Wrap server and client TF blocks with `tff.tf_computation`.
@tff.tf_computation(server_state_type, trainable_weights_type)
def server_update_fn(server_state, model_delta):
  return server_update(server_state, model_delta, server_optimizer)

whimsy_model = model_fn()
tf_dataset_type = tff.SequenceType(whimsy_model.input_spec)
print('tf_dataset_type:\n', 
      tf_dataset_type.formatted_representation())
@tff.tf_computation(tf_dataset_type, trainable_weights_type)
def client_update_fn(dataset, server_weights):
  model = model_fn()
  return client_update(model, dataset, server_weights, client_optimizer)

# 3-2. Orchestration with `tff.federated_computation`.
federated_server_type = tff.FederatedType(server_state_type, tff.SERVER)
federated_dataset_type = tff.FederatedType(tf_dataset_type, tff.CLIENTS)
@tff.federated_computation(federated_server_type, federated_dataset_type)
def run_one_round(server_state, federated_dataset):
  # Server-to-client broadcast.
  server_weights_at_client = tff.federated_broadcast(
      server_state.trainable_weights)
  # Local client update.
  model_deltas = tff.federated_map(
      client_update_fn, (federated_dataset, server_weights_at_client))
  # Client-to-server upload and aggregation.
  mean_model_delta = tff.federated_mean(model_deltas)
  # Server update.
  server_state = tff.federated_map(
      server_update_fn, (server_state, mean_model_delta))
  return server_state

# 4. Build the iterative process for FedAvg.
fedavg_process = tff.templates.IterativeProcess(
    initialize_fn=server_init_tff, next_fn=run_one_round)
print('type signature of `initialize`:\n', 
      fedavg_process.initialize.type_signature.formatted_representation())
print('type signature of `next`:\n', 
      fedavg_process.next.type_signature.formatted_representation())
server_state_type:
 <
  trainable_weights=<
    float32[5,5,1,32],
    float32[32],
    float32[5,5,32,64],
    float32[64],
    float32[3136,512],
    float32[512],
    float32[512,10],
    float32[10]
  >,
  optimizer_state=<
    float32[5,5,1,32],
    float32[32],
    float32[5,5,32,64],
    float32[64],
    float32[3136,512],
    float32[512],
    float32[512,10],
    float32[10]
  >
>
trainable_weights_type:
 <
  float32[5,5,1,32],
  float32[32],
  float32[5,5,32,64],
  float32[64],
  float32[3136,512],
  float32[512],
  float32[512,10],
  float32[10]
>
tf_dataset_type:
 <
  float32[?,28,28,1],
  int32[?]
>*
type signature of `initialize`:
 ( -> <
  trainable_weights=<
    float32[5,5,1,32],
    float32[32],
    float32[5,5,32,64],
    float32[64],
    float32[3136,512],
    float32[512],
    float32[512,10],
    float32[10]
  >,
  optimizer_state=<
    float32[5,5,1,32],
    float32[32],
    float32[5,5,32,64],
    float32[64],
    float32[3136,512],
    float32[512],
    float32[512,10],
    float32[10]
  >
>@SERVER)
type signature of `next`:
 (<
  server_state=<
    trainable_weights=<
      float32[5,5,1,32],
      float32[32],
      float32[5,5,32,64],
      float32[64],
      float32[3136,512],
      float32[512],
      float32[512,10],
      float32[10]
    >,
    optimizer_state=<
      float32[5,5,1,32],
      float32[32],
      float32[5,5,32,64],
      float32[64],
      float32[3136,512],
      float32[512],
      float32[512,10],
      float32[10]
    >
  >@SERVER,
  federated_dataset={<
    float32[?,28,28,1],
    int32[?]
  >*}@CLIENTS
> -> <
  trainable_weights=<
    float32[5,5,1,32],
    float32[32],
    float32[5,5,32,64],
    float32[64],
    float32[3136,512],
    float32[512],
    float32[512,10],
    float32[10]
  >,
  optimizer_state=<
    float32[5,5,1,32],
    float32[32],
    float32[5,5,32,64],
    float32[64],
    float32[3136,512],
    float32[512],
    float32[512,10],
    float32[10]
  >
>@SERVER)

Avaliando o algoritmo

Avaliamos o desempenho em um conjunto de dados de avaliação centralizado.

def evaluate(server_state):
  keras_model = create_keras_model()
  tf.nest.map_structure(
      lambda var, t: var.assign(t),
      keras_model.trainable_weights, server_state.trainable_weights)
  metric = tf.keras.metrics.SparseCategoricalAccuracy()
  for batch in iter(central_test_data):
    preds = keras_model(batch[0], training=False)
    metric.update_state(y_true=batch[1], y_pred=preds)
  return metric.result().numpy()
server_state = fedavg_process.initialize()
acc = evaluate(server_state)
print('Initial test accuracy', acc)

# Evaluate after a few rounds
CLIENTS_PER_ROUND=2
sampled_clients = train_client_ids[:CLIENTS_PER_ROUND]
sampled_train_data = [
    train_data.create_tf_dataset_for_client(client)
    for client in sampled_clients]
for round in range(20):
  server_state = fedavg_process.next(server_state, sampled_train_data)
acc = evaluate(server_state)
print('Test accuracy', acc)
Initial test accuracy 0.09677419
Test accuracy 0.13978495