TensorFlow Addons 优化器:LazyAdam

在 TensorFlow.org 上查看 在 Google Colab 中运行 在 GitHub 中查看源代码 {img1下载笔记本

概述

此笔记本将演示如何使用 Addons 包中的 Lazy Adam 优化器。

LazyAdam

LazyAdam 是 Adam 优化器的一种变体,可以更高效地处理稀疏更新。原始的 Adam 算法为每个可训练变量维护两个移动平均累加器,这些累加器在每一步都会更新。此类为稀疏变量提供了更加懒惰的梯度更新处理。它仅更新当前批次中出现的稀疏变量索引的移动平均累加器,而不是更新所有索引的累加器。与原始的 Adam 优化器相比,它可以大幅提高某些应用的模型训练吞吐量。但是,它的语义与原始的 Adam 算法略有不同,这可能会产生不同的实验结果。

设置

pip install -q -U tensorflow-addons
import tensorflow as tf
import tensorflow_addons as tfa
# Hyperparameters
batch_size=64
epochs=10

构建模型

model = tf.keras.Sequential([
    tf.keras.layers.Dense(64, input_shape=(784,), activation='relu', name='dense_1'),
    tf.keras.layers.Dense(64, activation='relu', name='dense_2'),
    tf.keras.layers.Dense(10, activation='softmax', name='predictions'),
])

准备数据

# Load MNIST dataset as NumPy arrays
dataset = {}
num_validation = 10000
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()

# Preprocess the data
x_train = x_train.reshape(-1, 784).astype('float32') / 255
x_test = x_test.reshape(-1, 784).astype('float32') / 255

训练和评估

只需用新的 TFA 优化器替换典型的 Keras 优化器

# Compile the model
model.compile(
    optimizer=tfa.optimizers.LazyAdam(0.001),  # Utilize TFA optimizer
    loss=tf.keras.losses.SparseCategoricalCrossentropy(),
    metrics=['accuracy'])

# Train the network
history = model.fit(
    x_train,
    y_train,
    batch_size=batch_size,
    epochs=epochs)
Epoch 1/10
938/938 [==============================] - 2s 2ms/step - loss: 0.3188 - accuracy: 0.9079
Epoch 2/10
938/938 [==============================] - 2s 2ms/step - loss: 0.1316 - accuracy: 0.9607
Epoch 3/10
938/938 [==============================] - 2s 2ms/step - loss: 0.0970 - accuracy: 0.9706
Epoch 4/10
938/938 [==============================] - 2s 2ms/step - loss: 0.0783 - accuracy: 0.9760
Epoch 5/10
938/938 [==============================] - 2s 2ms/step - loss: 0.0622 - accuracy: 0.9805
Epoch 6/10
938/938 [==============================] - 2s 2ms/step - loss: 0.0516 - accuracy: 0.9843
Epoch 7/10
938/938 [==============================] - 2s 2ms/step - loss: 0.0431 - accuracy: 0.9863
Epoch 8/10
938/938 [==============================] - 2s 2ms/step - loss: 0.0392 - accuracy: 0.9873
Epoch 9/10
938/938 [==============================] - 2s 2ms/step - loss: 0.0332 - accuracy: 0.9893
Epoch 10/10
938/938 [==============================] - 2s 2ms/step - loss: 0.0283 - accuracy: 0.9909
# Evaluate the network
print('Evaluate on test data:')
results = model.evaluate(x_test, y_test, batch_size=128, verbose = 2)
print('Test loss = {0}, Test acc: {1}'.format(results[0], results[1]))
Evaluate on test data:
79/79 - 0s - loss: 0.0843 - accuracy: 0.9765
Test loss = 0.08429064601659775, Test acc: 0.9764999747276306