求助！tensorflow转pytorch！！

最新推荐文章于 2023-05-25 10:59:09 发布

置顶 qq_41869108

最新推荐文章于 2023-05-25 10:59:09 发布

阅读量369

点赞数

文章标签： pytorch tensorflow 深度学习神经网络

本文链接：https://blog.csdn.net/qq_41869108/article/details/128151188

版权

有没有哪位大佬指导一下这段代码用pytorch怎么实现？代码小白的脑子都要想炸了。
class Filter(Model, ABC):
    def __init__(self, emb_size, reg_term):     # emb_size嵌入维度   reg_term正则化参数
        super(Filter, self).__init__()
        self.emb_size = emb_size
        self.reg_term = reg_term
        self.mlp_layers = []
        self.bn_layers = []
        self.film_alpha_layers = []
        self.film_beta_layers = []
        for i in range(3):
            self.mlp_layers.append(
                Dense(self.emb_size, activation='relu'))
            self.film_alpha_layers.append(
                Dense(self.emb_size, activation=None,
                      activity_regularizer=L2Regularizer_to_one(self.reg_term),        
                      use_bias=False))
            self.film_beta_layers.append(
                Dense(self.emb_size, activation=None,
                      activity_regularizer=L2(self.reg_term), use_bias=False))
            self.bn_layers.append(BatchNormalization())

    def call(self, inputs, training):       # inputs是利用nn.Embedding编码好的向量
        embedding = inputs
        for i in range(3):
            filter_layer = self.mlp_layers[i]
            film_alpha_layer = self.film_alpha_layers[i]
            film_beta_layer = self.film_beta_layers[i]
            bn_layer = self.bn_layers[i]
            embedding = filter_layer(embedding)
            alpha = film_alpha_layer(inputs)
            beta = film_beta_layer(inputs)
            embedding = tf.multiply(embedding, alpha) + beta
            embedding = bn_layer(embedding, training=training)
            embedding = tf.nn.leaky_relu(embedding)
        return embedding