thinkphp3.2.3 找不到自定义模型_TF2 6.tf.GradientTape自定义模型训练

自动求导机制：

梯度求解利器：tf.GradientTape

GradientTape是eager模式下计算梯度用的，而eager模式是TensorFlow2.0的默认模式。

x = tf.constant(3.0)

with tf.GradientTape() as g:
    g.watch(x)
    y = x * x
dy_dx = g.gradient(y, x)  # y’ = 2*x = 2*3 = 6

详细看下apply_gradients：

---

案例1、模型自动求导

构建模型（神经网络的前向传播） --> 定义损失函数 --> 定义优化函数 --> 定义tape --> 模型得到预测值 --> 前向传播得到loss --> 反向传播 --> 用优化函数将计算出来的梯度更新到变量上面去。

class MyModel(tf.keras.Model):

    def __init__(self, num_classes=10):
        super(MyModel, self).__init__(name='my_model')
        self.num_classes = num_classes
        # 定义自己需要的层
        self.dense_1 = tf.keras.layers.Dense(32, activation='relu') #隐藏层
        self.dense_2 = tf.keras.layers.Dense(num_classes)#输出层

    def call(self, inputs):
        #定义前向传播
        # 使用在 (in `__init__`)定义的层
        x = self.dense_1(inputs)
        return self.dense_2(x)

import numpy as np
# 10分类问题
data = np.random.random((1000, 32))
labels = np.random.random((1000, 10))

model = MyModel(num_classes=10)

loss_object = tf.keras.losses.CategoricalCrossentropy(from_logits=True)
optimizer = tf.keras.optimizers.Adam()


with tf.GradientTape() as tape:
    predictions = model(data)
    loss = loss_object(labels, predictions)
    
gradients = tape.gradient(loss, model.trainable_variables) #求梯度

optimizer.apply_gradients(zip(gradients, model.trainable_variables)) # 

---

案例2：使用GradientTape自定义训练模型

class MyModel(tf.keras.Model):

    def __init__(self, num_classes=10):
        super(MyModel, self).__init__(name='my_model')
        self.num_classes = num_classes
        # 定义自己需要的层
        self.dense_1 = tf.keras.layers.Dense(32, activation='relu')
        self.dense_2 = tf.keras.layers.Dense(num_classes)

    def call(self, inputs):
        #定义前向传播
        # 使用在 (in `__init__`)定义的层
        x = self.dense_1(inputs)
        return self.dense_2(x)

import numpy as np

data = np.random.random((1000, 32))
labels = np.random.random((1000, 10))

model = MyModel(num_classes=10)

# Instantiate an optimizer.
optimizer = tf.keras.optimizers.SGD(learning_rate=1e-3)
# Instantiate a loss function.
loss_fn = tf.keras.losses.CategoricalCrossentropy()

# Prepare the training dataset.
batch_size = 64
train_dataset = tf.data.Dataset.from_tensor_slices((data, labels))
train_dataset = train_dataset.shuffle(buffer_size=1024).batch(batch_size)

epochs = 3
for epoch in range(epochs):
    print('Start of epoch %d' % (epoch,))
    

    # 遍历数据集的batch_size
    for step, (x_batch_train, y_batch_train) in enumerate(train_dataset):

        
        # 打开GradientTape以记录正向传递期间运行的操作，这将启用自动区分。
        with tf.GradientTape() as tape:

            # 运行该模型的前向传播。 模型应用于其输入的操作将记录在GradientTape上。
            logits = model(x_batch_train, training=True)  # 这个minibatch的预测值

            # 计算这个minibatch的损失值
            loss_value = loss_fn(y_batch_train, logits)

        # 使用GradientTape自动获取可训练变量相对于损失的梯度。
        grads = tape.gradient(loss_value, model.trainable_weights)

        # 通过更新变量的值来最大程度地减少损失，从而执行梯度下降的一步。
        optimizer.apply_gradients(zip(grads, model.trainable_weights))

        # 每200 batches打印一次.
        if step % 200 == 0:
            print('Training loss (for one batch) at step %s: %s' % (step, float(loss_value)))
            print('Seen so far: %s samples' % ((step + 1) * 64))

---

案例3：使用GradientTape自定义训练模型进阶（加入评估函数）

让我们将metric添加到组合中。下面可以在从头开始编写的训练循环中随时使用内置指标（或编写的自定义指标）。流程如下：

• 在循环开始时初始化metrics
• metric.update_state():每batch之后更新
• metric.result():需要显示metrics的当前值时调用
• metric.reset_states():需要清除metrics状态时重置（通常在每个epoch的结尾）

class MyModel(tf.keras.Model):

    def __init__(self, num_classes=10):
        super(MyModel, self).__init__(name='my_model')
        self.num_classes = num_classes
        # 定义自己需要的层
        self.dense_1 = tf.keras.layers.Dense(32, activation='relu')
        self.dense_2 = tf.keras.layers.Dense(num_classes)
    
    def call(self, inputs):
        #定义前向传播
        # 使用在 (in `__init__`)定义的层
        x = self.dense_1(inputs)
        return self.dense_2(x)


import numpy as np
x_train = np.random.random((1000, 32))
y_train = np.random.random((1000, 10))
x_val = np.random.random((200, 32))
y_val = np.random.random((200, 10))
x_test = np.random.random((200, 32))
y_test = np.random.random((200, 10))




# 优化器
optimizer = tf.keras.optimizers.SGD(learning_rate=1e-3)
# 损失函数
loss_fn = tf.keras.losses.CategoricalCrossentropy(from_logits=True)

# 准备metrics函数
train_acc_metric = tf.keras.metrics.CategoricalAccuracy()
val_acc_metric = tf.keras.metrics.CategoricalAccuracy()

# 准备训练数据集
batch_size = 64
train_dataset = tf.data.Dataset.from_tensor_slices((x_train, y_train))
train_dataset = train_dataset.shuffle(buffer_size=1024).batch(batch_size)

# 准备测试数据集
val_dataset = tf.data.Dataset.from_tensor_slices((x_val, y_val))
val_dataset = val_dataset.batch(64)

进行几个epoch运行训练循环：

model = MyModel(num_classes=10)
epochs = 3
for epoch in range(epochs):
    print('Start of epoch %d' % (epoch,))

    # 遍历数据集的batch_size
    for step, (x_batch_train, y_batch_train) in enumerate(train_dataset):
        
        
        #一个batch
        with tf.GradientTape() as tape:
            logits = model(x_batch_train)
            loss_value = loss_fn(y_batch_train, logits)
        grads = tape.gradient(loss_value, model.trainable_weights)
        optimizer.apply_gradients(zip(grads, model.trainable_weights))####

        # 更新训练集的metrics
        train_acc_metric(y_batch_train, logits)     
            
            
    # 在每个epoch结束时显示metrics。
    train_acc = train_acc_metric.result()
    print('Training acc over epoch: %s' % (float(train_acc),))
    # 在每个epoch结束时重置训练指标
    train_acc_metric.reset_states()#!!!!!!!!!!!!!!!

    # 在每个epoch结束时运行一个验证集。
    for x_batch_val, y_batch_val in val_dataset:
        val_logits = model(x_batch_val)
        # 更新验证集merics
        val_acc_metric(y_batch_val, val_logits)
    val_acc = val_acc_metric.result()
    print('Validation acc: %s' % (float(val_acc),))
    val_acc_metric.reset_states()
    
    #显示测试集

---

鸢尾花自定义模型：

import tensorflow as tf
from sklearn import datasets #导入数据集
import numpy as np

iris = datasets.load_iris()
data = iris.data
labels = iris.target
data = np.concatenate((data,labels.reshape(150,1)),axis=-1)
#将数据打乱：神经网络的拟合能力太强了，如果不乱序的话，同一个组合的batch反复出现，模型有可能会“记住”这些样本的次序，从而影响泛化能力。
np.random.shuffle(data)

#将数据拆分为数据和标签
Y = data[:,-1]
X = data[:,:4]

#子类模型
class MyModel(tf.keras.Model):
    def __init__(self):
        super(MyModel,self).__init__()
        self.D1 = tf.keras.layers.Dense(32,activation='relu')
        self.D2 = tf.keras.layers.Dense(32,activation='relu')
        self.D3 = tf.keras.layers.Dense(3,activation='softmax')
    def call(self,inputs):
        x = self.D1(inputs)
        x = self.D2(x)
        x = self.D3(x)
        return x
model = MyModel()

#2.定义损失函数
loss_object = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
#3.定义优化函数
optimizer = tf.keras.optimizers.SGD(learning_rate=0.01)
batch_size = 32
train_dataset = tf.data.Dataset.from_tensor_slices((X, Y))
train_dataset = train_dataset.shuffle(buffer_size=64).batch(batch_size)
train_acc_metric = tf.keras.metrics.SparseCategoricalAccuracy()

epochs = 100
for epoch in range(epochs):
    for step,(x_batch_train,y_batch_train) in enumerate(train_dataset):
        #定义tape计算梯度
        with tf.GradientTape() as tape:

            # 运行该模型的前向传播。 模型应用于其输入的操作将记录在GradientTape上。
            logits = model(x_batch_train)  # 这个minibatch的预测值

            # 计算这个minibatch的损失值
            loss_value = loss_object(y_batch_train, logits)
            
        grads = tape.gradient(loss_value, model.trainable_weights)
        optimizer.apply_gradients(zip(grads, model.trainable_weights))

        # 更新训练集的metrics
        train_acc_metric(y_batch_train, logits)
        
    train_acc = train_acc_metric.result()
    print('Training acc over epoch: %s' % (float(train_acc),))