将Keras函数-API转换为Keras子类模型

Question

我对Keras和Tensorflow比较陌生，我想学习基本的实现。为此，我想要构建一个能够学习/检测/预测手写数字的模型，因此我使用来自Keras的MNIST数据集。我已经用创建了这个模型，一切都很好。现在我想做同样的事情，但是这次我想构建一个Keras子类模型。问题是，当我使用Keras子类模型执行代码时，出现了一个错误。这是带有functional的模型的代码(工作正常，没有任何问题)：

import tensorflow as tf 
from tensorflow import keras
from tensorflow.keras import layers    
from keras.datasets import mnist
import numpy as np 

#Load MNIST-Dataset
(x_train_full, y_train_full), (x_test, y_test) = mnist.load_data()

#Create train- and validationdata
X_valid = x_train_full[:5000]/255.0         
X_train = x_train_full[5000:] / 255.0       
y_valid, y_train = y_train_full[:5000], y_train_full[5000:]   



#Create the model with the keras functional-API
inputs = keras.layers.Input(shape=(28, 28))                
flatten = keras.layers.Flatten(input_shape=(28, 28))(inputs)        
hidden1 = keras.layers.Dense(256, activation="relu")(flatten)
hidden2 = keras.layers.Dense(128, activation='relu')(hidden1)
outputs = keras.layers.Dense(10, activation='softmax')(hidden2)

model = keras.Model(inputs=[inputs], outputs=[outputs])

model.compile(loss="sparse_categorical_crossentropy", optimizer="sgd", metrics=["accuracy"])
h = model.fit(X_train, y_train, epochs=5, validation_data=(X_valid, y_valid))

#Evaluate the model with testdata
test_loss, test_acc = model.evaluate(x_test, y_test, verbose=2)
print('\nTest accuracy: ', test_acc)
print('\nTest loss: ', test_loss)

#Create Predictions:
myPrediction = model.predict(x_test)

#Prediction example of one testpicture
print(myPrediction[0])
print('Predicted Item: ', np.argmax(myPrediction[0]))
print('Actual Item: ', y_test[0])

下面是Keras子类模型的(不工作)代码，它应该与上面的代码完全相同：

import tensorflow as tf 
from tensorflow import keras
from tensorflow.keras import layers    
from keras.datasets import mnist
import numpy as np 

#Load MNIST-Dataset
(x_train_full, y_train_full), (x_test, y_test) = mnist.load_data()

#Create train- and validationdata
X_valid = x_train_full[:5000]/255.0         
X_train = x_train_full[5000:] / 255.0       
y_valid, y_train = y_train_full[:5000], y_train_full[5000:] 

#Create a keras-subclassing-model:
class MyModel(tf.keras.Model):

    def __init__(self):
        super(MyModel, self).__init__()
        #Define layers
        self.input_ = keras.layers.Input(shape=(28, 28))
        self.flatten = keras.layers.Flatten(input_shape=(28, 28))
        self.dense_1 = keras.layers.Dense(256, activation="relu")
        self.dense_2 = keras.layers.Dense(128, activation="relu")
        self.output_ = keras.layers.Dense(10, activation="softmax")

    def call(self, inputs):
        x = self.input_(inputs)
        x = self.flatten(x)
        x = self.dense_1(x)
        x = self.dense_2(x)
        x = self.output_(x)
        return x

model = MyModel()

model.compile(loss="sparse_categorical_crossentropy", optimizer="sgd", metrics=["accuracy"])

h = model.fit(X_train, y_train, epochs=10, validation_data=(X_valid, y_valid))

每次我运行这段代码时都会遇到同样的错误。当调用fit(...)-method时出现此错误：

Traceback (most recent call last):
  File "c:/Users/MichaelM/Documents/PythonSkripte/MachineLearning/SubclassedModel.py", line 39, in <module>
    h = model.fit(X_train, y_train, epochs=10, validation_data=(X_valid, y_valid))
  File "C:\Python37\lib\site-packages\tensorflow_core\python\keras\engine\training.py", line 819, in fit
    use_multiprocessing=use_multiprocessing)
  File "C:\Python37\lib\site-packages\tensorflow_core\python\keras\engine\training_v2.py", line 235, in fit
    use_multiprocessing=use_multiprocessing)
  File "C:\Python37\lib\site-packages\tensorflow_core\python\keras\engine\training_v2.py", line 593, in _process_training_inputs
    use_multiprocessing=use_multiprocessing)
  File "C:\Python37\lib\site-packages\tensorflow_core\python\keras\engine\training_v2.py", line 646, in _process_inputs
    x, y, sample_weight=sample_weights)
  File "C:\Python37\lib\site-packages\tensorflow_core\python\keras\engine\training.py", line 2346, in _standardize_user_data
    all_inputs, y_input, dict_inputs = self._build_model_with_inputs(x, y)
  File "C:\Python37\lib\site-packages\tensorflow_core\python\keras\engine\training.py", line 2572, in _build_model_with_inputs
    self._set_inputs(cast_inputs)
  File "C:\Python37\lib\site-packages\tensorflow_core\python\keras\engine\training.py", line 2659, in _set_inputs
    outputs = self(inputs, **kwargs)
  File "C:\Python37\lib\site-packages\tensorflow_core\python\keras\engine\base_layer.py", line 773, in __call__
    outputs = call_fn(cast_inputs, *args, **kwargs)
  File "C:\Python37\lib\site-packages\tensorflow_core\python\autograph\impl\api.py", line 237, in wrapper
    raise e.ag_error_metadata.to_exception(e)
TypeError: in converted code:

    c:/Users/MichaelM/Documents/PythonSkripte/MachineLearning/SubclassedModel.py:28 call  *
        x = self.input_(inputs)
    C:\Python37\lib\site-packages\tensorflow_core\python\autograph\impl\api.py:447 converted_call
        f in m.__dict__.values() for m in (collections, pdb, copy, inspect, re)):
    C:\Python37\lib\site-packages\tensorflow_core\python\autograph\impl\api.py:447 <genexpr>
        f in m.__dict__.values() for m in (collections, pdb, copy, inspect, re)):
    C:\Python37\lib\site-packages\tensorflow_core\python\ops\math_ops.py:1351 tensor_equals
        return gen_math_ops.equal(self, other, incompatible_shape_error=False)
    C:\Python37\lib\site-packages\tensorflow_core\python\ops\gen_math_ops.py:3240 equal
        name=name)
    C:\Python37\lib\site-packages\tensorflow_core\python\framework\op_def_library.py:477 _apply_op_helper
        repr(values), type(values).__name__, err))

    TypeError: Expected float32 passed to parameter 'y' of op 'Equal', got 'collections' of type 'str' instead. Error: Expected float32, got 'collections' of type 'str' instead.

你能帮我解决这个问题吗?也许能解释一下为什么这个问题不起作用，因为我不知道这个错误到底意味着什么。然后，我可以调用函数API代码中的evaluate(...)和predict(...)方法吗？我使用以下配置：

• 以Python扩展为IDE的Visual代码
• Python-版本: 3.7.6
• TensorFlow-版本: 2.1.0
• Keras-版本: 2.2.4-tf

Answer 1

实际上，您不需要在call方法中实现输入，因为您将数据直接传递给子类。我更新了代码，它运行良好，就像预期的那样。请检查下面。

#Create a keras-subclassing-model:
class MyModel(tf.keras.Model):

    def __init__(self):
        super(MyModel, self).__init__()
        #Define layers
        #self.input_ = keras.layers.Input(shape=(28, 28))
        self.flatten = keras.layers.Flatten(input_shape=(28, 28))
        self.dense_1 = keras.layers.Dense(256, activation="relu")
        self.dense_2 = keras.layers.Dense(128, activation="relu")
        self.output_ = keras.layers.Dense(10, activation="softmax")

    def call(self, inputs):
        #x = self.input_(inputs)
        x = self.flatten(inputs)
        x = self.dense_1(x)
        x = self.dense_2(x)
        x = self.output_(x)
        return x

model = MyModel()

model.compile(loss="sparse_categorical_crossentropy", optimizer="sgd", metrics=["accuracy"])

h = model.fit(X_train, y_train, epochs=10, validation_data=(X_valid, y_valid))

输出如下

Epoch 1/10
1719/1719 [==============================] - 6s 3ms/step - loss: 0.6251 - accuracy: 0.8327 - val_loss: 0.3068 - val_accuracy: 0.9180
....
....
Epoch 10/10
1719/1719 [==============================] - 6s 3ms/step - loss: 0.1097 - accuracy: 0.9687 - val_loss: 0.1215 - val_accuracy: 0.9648

完整代码是这里。