tech
Defaultzh

TensorFlow

谷歌机器学习框架 - 生产就绪、TF Serving、移动部署、大型生态

TL;DR

是什么：谷歌的开源机器学习框架。

为什么用：生产就绪、TensorFlow Serving、移动部署、大型生态系统。

Quick Start

安装：

pip install tensorflow

Hello TensorFlow：

import tensorflow as tf

# 检查版本
print(tf.__version__)

# 简单张量操作
a = tf.constant([[1, 2], [3, 4]])
b = tf.constant([[5, 6], [7, 8]])
print(tf.matmul(a, b))

# 检查 GPU
print("GPU available:", tf.config.list_physical_devices('GPU'))

Cheatsheet

操作代码
创建张量tf.constant([1, 2, 3])
零/一tf.zeros((3, 3))
随机tf.random.normal((3, 3))
形状tensor.shape
重塑tf.reshape(x, (2, 3))
转 numpytensor.numpy()

Gotchas

Sequential 模型

import tensorflow as tf
from tensorflow.keras import layers, models

model = models.Sequential([
    layers.Dense(128, activation='relu', input_shape=(784,)),
    layers.Dropout(0.2),
    layers.Dense(64, activation='relu'),
    layers.Dense(10, activation='softmax')
])

model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',
    metrics=['accuracy']
)

model.summary()

训练

# 加载数据
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.mnist.load_data()
x_train = x_train.reshape(-1, 784).astype('float32') / 255.0
x_test = x_test.reshape(-1, 784).astype('float32') / 255.0

# 训练
history = model.fit(
    x_train, y_train,
    epochs=10,
    batch_size=32,
    validation_split=0.2
)

# 评估
loss, accuracy = model.evaluate(x_test, y_test)
print(f'Test accuracy: {accuracy}')

函数式 API

inputs = tf.keras.Input(shape=(784,))
x = layers.Dense(128, activation='relu')(inputs)
x = layers.Dense(64, activation='relu')(x)
outputs = layers.Dense(10, activation='softmax')(x)

model = tf.keras.Model(inputs=inputs, outputs=outputs)

自定义训练循环

@tf.function
def train_step(model, optimizer, x, y, loss_fn):
    with tf.GradientTape() as tape:
        predictions = model(x, training=True)
        loss = loss_fn(y, predictions)
    gradients = tape.gradient(loss, model.trainable_variables)
    optimizer.apply_gradients(zip(gradients, model.trainable_variables))
    return loss

# 训练循环
for epoch in range(epochs):
    for x_batch, y_batch in dataset:
        loss = train_step(model, optimizer, x_batch, y_batch, loss_fn)

保存和加载

# 保存整个模型
model.save('model.keras')

# 加载模型
loaded_model = tf.keras.models.load_model('model.keras')

# 仅保存权重
model.save_weights('weights.h5')
model.load_weights('weights.h5')

# SavedModel 格式（用于服务）
tf.saved_model.save(model, 'saved_model/')

Next Steps