获取子类模型的权重

Question

我一直在研究电影镜头和协同过滤。我使用了一个子类模型，并希望获得经过训练的嵌入，以在二维等中可视化它们。不知何故，权重似乎是空的，但我不知道为什么。显然，模型在训练时变得更好（如果我绘制历史记录['loss']）。我不知道为什么权重是空的或如何获取它们如果我使用模型而不是子类它工作正常并且我可以获得权重。下面是我的代码

class RegressionModel(Model):
    def __init__(self, embedding_size, max_user_id, max_item_id):
        super().__init__()      
        self.user_embedding = Embedding(output_dim=embedding_size,
                                        input_dim=max_user_id + 1,
                                        input_length=1,
                                        name='user_embedding')
        self.item_embedding = Embedding(output_dim=embedding_size,
                                        input_dim=max_item_id + 1,
                                        input_length=1,
                                        name='item_embedding')
        
        self.flatten = Flatten()
        self.dot = Dot(axes=1)
     

    def call(self, inputs, training=False):
        user_inputs = inputs[0]
        item_inputs = inputs[1]
        
        user_vecs = self.flatten(self.user_embedding(user_inputs))
        item_vecs = self.flatten(self.item_embedding(item_inputs))
        
        y = self.dot([user_vecs, item_vecs])    
        return y
    
    
    
max_user_id = ratings['userId'].max()
max_item_id = ratings['movieId'].max()

thename = 'models/my_model'#cannot save as .h5 as it's a subclassed model
model = RegressionModel(64, max_user_id, max_item_id)
model.compile(optimizer="adam", loss='mae')

              
inputs_train = [np.array(ratings_train["userId"]),  np.array(ratings_train["movieId"])]
inputs_test = [np.array(ratings_test["userId"]),  np.array(ratings_test["movieId"])]

model.fit(inputs_train,
                                    np.array(ratings_train["rating"]),
                                    batch_size=64, 
                                    epochs=10, 
                                    validation_split=0.1,
                                    shuffle=True)


#save model
#model.save doesn't work for a subclassed model but save_weights works
def get_model():  
    return RegressionModel(s, max_user_id, max_item_id)

model.save_weights(thename, save_format='tf')


#load the model
model_new = get_model()  
model_new.compile(optimizer="adam", loss='mae') 
model_new.load_weights(thename)  

weights = model_new.get_weights()

#this is empty (even without saving and then loading the weights)
weights

Answer 1

实际上从tensorflow 2.2开始你可以保存自定义模型。
您需要指定格式，

# Save the model
model.save('path_to_my_model',save_format='tf')

# Recreate the exact same model purely from the file
new_model = keras.models.load_model('path_to_my_model')

尝试查看无法保存自定义子类模型