使用 Keras 的模型类将 Tensorflow 1.x 代码迁移到 Tensorflow 2.x

Question

我刚刚开始使用 Keras 2.3.1 和 Python 3.7.7 学习 Tensorflow 2.1.0。

我发现这个“Omniglot Character Set Classification Using Prototypical Network”github Jupyter Notebook，我认为它适用于 Tensorflow 1.x。

我的问题在于这段代码：

for epoch in range(num_epochs):

    for episode in range(num_episodes):

        # select 60 classes
        episodic_classes = np.random.permutation(no_of_classes)[:num_way]

        support = np.zeros([num_way, num_shot, img_height, img_width], dtype=np.float32)

        query = np.zeros([num_way, num_query, img_height, img_width], dtype=np.float32)


        for index, class_ in enumerate(episodic_classes):
            selected = np.random.permutation(num_examples)[:num_shot + num_query]
            support[index] = train_dataset[class_, selected[:num_shot]]

            # 5 querypoints per classs
            query[index] = train_dataset[class_, selected[num_shot:]]

        support = np.expand_dims(support, axis=-1)
        query = np.expand_dims(query, axis=-1)
        labels = np.tile(np.arange(num_way)[:, np.newaxis], (1, num_query)).astype(np.uint8)
        _, loss_, accuracy_ = sess.run([train, loss, accuracy], feed_dict={support_set: support, query_set: query, y:labels})

        if (episode+1) % 10 == 0:
            print('Epoch {} : Episode {} : Loss: {}, Accuracy: {}'.format(epoch+1, episode+1, loss_, accuracy_))

是否有任何教程或书籍或文章可以帮助我使用 Keras 的模型将此代码迁移到 Tensorflow 2.x 和 Keras？

我想从链接中编写代码，如下所示：

import numpy as np 
import os
import skimage.io as io
import skimage.transform as trans
import numpy as np
from keras.models import *
from keras.layers import *
from keras.optimizers import *
from keras.callbacks import ModelCheckpoint, LearningRateScheduler
from keras import backend as keras

def unet(pretrained_weights = None,input_size = (256,256,1)):
    inputs = Input(input_size)
    conv1 = Conv2D(64, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(inputs)
    conv1 = Conv2D(64, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv1)
    pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
    conv2 = Conv2D(128, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(pool1)
    conv2 = Conv2D(128, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv2)
    pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
    conv3 = Conv2D(256, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(pool2)
    conv3 = Conv2D(256, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv3)
    pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)
    conv4 = Conv2D(512, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(pool3)
    conv4 = Conv2D(512, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv4)
    drop4 = Dropout(0.5)(conv4)
    pool4 = MaxPooling2D(pool_size=(2, 2))(drop4)

    conv5 = Conv2D(1024, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(pool4)
    conv5 = Conv2D(1024, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv5)
    drop5 = Dropout(0.5)(conv5)

    up6 = Conv2D(512, 2, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(drop5))
    merge6 = concatenate([drop4,up6], axis = 3)
    conv6 = Conv2D(512, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(merge6)
    conv6 = Conv2D(512, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv6)

    up7 = Conv2D(256, 2, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(conv6))
    merge7 = concatenate([conv3,up7], axis = 3)
    conv7 = Conv2D(256, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(merge7)
    conv7 = Conv2D(256, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv7)

    up8 = Conv2D(128, 2, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(conv7))
    merge8 = concatenate([conv2,up8], axis = 3)
    conv8 = Conv2D(128, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(merge8)
    conv8 = Conv2D(128, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv8)

    up9 = Conv2D(64, 2, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(UpSampling2D(size = (2,2))(conv8))
    merge9 = concatenate([conv1,up9], axis = 3)
    conv9 = Conv2D(64, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(merge9)
    conv9 = Conv2D(64, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv9)
    conv9 = Conv2D(2, 3, activation = 'relu', padding = 'same', kernel_initializer = 'he_normal')(conv9)
    conv10 = Conv2D(1, 1, activation = 'sigmoid')(conv9)

    model = Model(input = inputs, output = conv10)

    model.compile(optimizer = Adam(lr = 1e-4), loss = 'binary_crossentropy', metrics = ['accuracy'])

    #model.summary()

    if(pretrained_weights):
        model.load_weights(pretrained_weights)

    return model

在train.py:

model = unet(...)
model.compile(...)
model.fit(...)

Answer 1

来自 Tensorflow 的 this tutorial 总结了一切。

最重要的是Sessions已经不存在了，应该使用tensorflow.keras.layers创建模型。

现在在训练模型时，您有 2 个选择，您可以使用 Keras 方式，也可以使用 GradientTape（这有点旧方式）。

这意味着您有两种选择，一种不会对您的代码产生太大影响（GradientTape），另一种只会让您改变一些事情（Keras）。

渐变胶带

GradientTape 用于您想要自己循环并计算所需的渐变，它有点像 Tensorflow 1.X。

• 使用 Keras API 构建模型：

import tensorflow as tf

def unet(...):
    inputs = tf.keras.layers.Input(shape_images)
    ...
    model = Model(input = inputs, output = conv10)

    model.compile(...)

    return model

...

model = unet(...)

• 定义损失

mse = tf.keras.losses.MeanSquaredError()

• 定义优化器

optimizer = tf.keras.optimizer.Adam(lr=1e-4)

然后，您可以像往常一样进行训练，只是将旧的 Session 机制替换为 GradientTape ：

for epoch in range(num_epochs):

    for episode in range(num_episodes):

        # select 60 classes
        episodic_classes = np.random.permutation(no_of_classes)[:num_way]

        support = np.zeros([num_way, num_shot, img_height, img_width], dtype=np.float32)

        query = np.zeros([num_way, num_query, img_height, img_width], dtype=np.float32)


        for index, class_ in enumerate(episodic_classes):
            selected = np.random.permutation(num_examples)[:num_shot + num_query]
            support[index] = train_dataset[class_, selected[:num_shot]]

            # 5 querypoints per classs
            query[index] = train_dataset[class_, selected[num_shot:]]

        support = np.expand_dims(support, axis=-1)
        query = np.expand_dims(query, axis=-1)
        labels = np.tile(np.arange(num_way)[:, np.newaxis], (1, num_query)).astype(np.uint8)

        # No session here but a Gradient computing

        with tf.GradientTape() as tape:
            prediction = model(support) # or whatever you need as input of model
            loss = mse(label, prediction)
        # apply gradient descent
        grads = tape.gradient(loss, model.trainable_weights)
        optimizer.apply_gradients(zip(grads, model.trainable_weights))

Keras

对于 keras，您需要更改提供数据的方式，因为您使用 fit，所以您不会有 for 循环，而您需要实现 Generator 或任何可以迭代的数据结构。这意味着您基本上需要(X, y) 的列表。 data_struct[0] 将为您提供第一个 X、Y 对。

一旦你有了这个数据结构，就很容易了。

• 像 GradientTape 一样定义模型

• 像 GradientTape 一样定义优化器

• 编译模型

model.compile(loss='binary_crossentropy', optimizer=optimizer, metrics=['accuracy']) # Or whatever you need as loss/metrics

• 使用您的 data_struct 拟合模型

model.fit(data_struct, epochs=500) # Add validation_data if you want, callback ...