Tensorflow 2.0.0-beta1：'EagerTensor object is not callable'

Question

我正在尝试在 google colab 上使用 Tensorflow 和 Keras API 实施自定义训练。我使用 Tensorflow 2.0.0-beta1。

我的损失函数代码部分是：

model = tf.keras.Sequential([
    tf.keras.layers.Embedding(
        max_features, 32, 
        embeddings_initializer='random_uniform'
    ),
    tf.keras.layers.SimpleRNN(32, kernel_initializer='random_uniform'),  
    tf.keras.layers.Dense(1, activation=tf.nn.sigmoid,),  # input shape is required
])

predictions = model(input_train)
predictions = tf.reshape(predictions,[25000,])

loss_object = tf.keras.losses.binary_crossentropy(
    y_true=y_train, 
    y_pred=predictions
)

def loss(model, x, y):
    y_ = model(x)

    return loss_object(y_true=y, y_pred=y_)

l = loss(model, input_train, y_train)

产生这个错误：

 TypeError  Traceback (most recent call last) <ipython-input-17-675f7c1fd9d0> in <module>()
 return loss_object(y_true=y, y_pred=y_) 
 l = loss(model, input_train, y_train) 
 <ipython-input-17-675f7c1fd9d0> in 
 loss(model, x, y) y_ = model(x) 
return loss_object(y_true=y, y_pred=y_) l = loss(model, input_train, y_train) 
 TypeError: 'tensorflow.python.framework.ops.EagerTensor' object is not callable

Answer 1

您想要计算 model 给定输入 x、目标输出 y 和预测 y_ 的损失。所以 loss_object 应该是一个损失函数（而不是预先计算的损失），你可以用它来计算损失。因此，替换为：

loss_object = tf.keras.losses.binary_crossentropy(y_true=y_train, y_pred=predictions)

有了这个：

loss_object = tf.keras.losses.binary_crossentropy

Tensorflow 2.0.0-beta1：'EagerTensor object is not callable'

Tensorflow 2.0.0-beta1: 'EagerTensor object is not callable'

python

keras

tensorflow

loss-function

eager-execution