TF 訓練簡單梯度下降模型

tf.GrandientTape 訓練簡單的線性梯度下降模型

添加引用

首先需要添加TensorFlow引用和Eager Execution，如下：

from __future__ import absolute_import, division, print_function
import tensorflow as tf

tf.enable_eager_execution()
tfe = tf.contrib.eager

構造模型和樣本數(shù)據(jù)

構造一個樣本數(shù)=1000眼坏，權重=3扣溺，偏差=2的模型砂心，并且加入噪點天揖，如下：

NUM_EXAMPLES = 1000   # 樣本數(shù)1000
training_inputs = tf.random_normal([NUM_EXAMPLES])   # 輸出正態(tài)分布的隨機值
noise = tf.random_normal([NUM_EXAMPLES])
training_outputs = training_inputs * 3 + 2 + noise   # 構造標簽樣本（權重=3，偏差=2瘸爽，并且加入噪點）

定義函數(shù)

定義預測函數(shù)展运、損失函數(shù)和梯度下降函數(shù)活逆，如下：

# 預測函數(shù)
def prediction(input, weight, bias):
    return input * weight + bias

# 使用“均方差”的損失函數(shù)
def loss(weights, biases):
    error = prediction(training_inputs, weights, biases) - training_outputs
    return tf.reduce_mean(tf.square(error))

# 執(zhí)行梯度下降精刷，返回下降之后的權重和偏差
def grad(weights, biases):
    with tf.GradientTape() as tape:
        loss_value = loss(weights,biases)
    return tape.gradient(loss_value,[weights,biases])

訓練模型

初始化

初始化訓練次數(shù)、學習率蔗候、權重和偏差怒允，如下：

train_steps = 200
learning_rate = 0.01
# 權重和偏差設置任意值
W = tfe.Variable(5.)
B = tfe.Variable(10.)

訓練

for i in range(train_steps):
    dW, dB = grad(W, B)
    W.assign_sub(dW * learning_rate)
    B.assign_sub(dB * learning_rate)
    print("Loss at step {:03d}: {:.3f}".format(i, loss(W, B)))

得到訓練后的權重和方差，即我們的模型锈遥。

print("Final loss: {:.3f}".format(loss(W, B)))
print("W = {}, B = {}".format(W.numpy(), B.numpy()))

image

權重=3.018173... 偏差=2.17251... 與我們設定的3和2基本相符纫事。

完整實例代碼如下：

方式一：

from __future__ import absolute_import, division, print_function
import tensorflow as tf

tf.enable_eager_execution()
tfe = tf.contrib.eager

NUM_EXAMPLES = 1000   # 樣本數(shù)1000
training_inputs = tf.random_normal([NUM_EXAMPLES])    # 輸出正態(tài)分布的隨機值
noise = tf.random_normal([NUM_EXAMPLES])
training_outputs = training_inputs * 3 + 2 + noise    # 構造標簽樣本（權重=3，偏差=2所灸，并且加入噪點）

# 預測函數(shù)
def prediction(input, weight, bias):
    return input * weight + bias

# 使用“均方差”的損失函數(shù)
def loss(weights, biases):
    error = prediction(training_inputs, weights, biases) - training_outputs
    return tf.reduce_mean(tf.square(error))

# 執(zhí)行梯度下降丽惶，返回下降之后的權重和偏差
def grad(weights, biases):
    with tf.GradientTape() as tape:
        loss_value = loss(weights,biases)
    return tape.gradient(loss_value,[weights,biases])

train_steps = 200
learning_rate = 0.01
# 權重和偏差設置任意值
W = tfe.Variable(5.)
B = tfe.Variable(10.)

print("Initial loss: {:.3f}".format(loss(W, B)))

for i in range(train_steps):
    dW, dB = grad(W, B)
    W.assign_sub(dW * learning_rate)
    B.assign_sub(dB * learning_rate)
    print("Loss at step {:03d}: {:.3f}".format(i, loss(W, B)))


print("Final loss: {:.3f}".format(loss(W, B)))
print("W = {}, B = {}".format(W.numpy(), B.numpy()))

方式二（通過將 tfe.Variable 與 tf.GradientTape 結合使用可以更好地封裝模型參數(shù)）：

from __future__ import absolute_import, division, print_function
import tensorflow as tf
tf.enable_eager_execution()
tfe = tf.contrib.eager

class Model(tf.keras.Model):
    '''自定義模型'''
    def __init__(self):
        super(Model, self).__init__()
        self.W = tfe.Variable(5., name = 'weight')
        self.B = tfe.Variable(10., name = 'bias')

    def call(self, inputs):
        return inputs * self.W + self.B

# 構造樣本數(shù)據(jù)
NUM_EXAMPLES = 2000
training_inputs = tf.random_normal([NUM_EXAMPLES])  # 隨機正態(tài)分布的值
nosie = tf.random_normal([NUM_EXAMPLES])   # 噪點
training_outputs = training_inputs * 3 + 2 + nosie

# 定義損失和梯度函數(shù)
def loss(model, inputs, targets):
    error = model(inputs) - targets
    return tf.reduce_mean(tf.square(error))

def grad(model, inputs, targets):
    with tf.GradientTape() as tape:
        loss_value = loss(model, inputs, targets)
    return tape.gradient(loss_value, [model.W, model.B])

# 訓練
# 1. 實例化模型
# 2. 模型損失函數(shù)的導數(shù)
# 3. 更新變量的策略
model = Model()
optimizer = tf.train.GradientDescentOptimizer(learning_rate = 0.01)   # 優(yōu)化器

print("Initial loss: {:.3f}".format(loss(model, training_inputs, training_outputs).numpy()))
    
# 訓練循環(huán)
for i in range(300):
    grads = grad(model, training_inputs, training_outputs)
    optimizer.apply_gradients(zip(grads, [model.W, model.B]),global_step=tf.train.get_or_create_global_step())
    if i % 20 == 0:
            print('Loss at step {:03d}: {:.3f}'.format(i,loss(model, training_inputs,training_outputs).numpy()))

print('Final loss: {:.3f}'.format(loss(model, training_inputs, training_outputs).numpy()))
print('W = {}, B = {}'.format(model.W.numpy(), model.B.numpy()))