ValueError: tf.function-decorated function tried to create variables on non-

当你在构建自定义keras类时候，如果call中使用已有的layer，需要在__init__中提前定义

比如：

    def __init__(self, pre_lens=28, window=180, is_teachers=False, name='mul', **kwargs):
        super(multcn, self).__init__(name=name,**kwargs)
        self.tcn = TCN(
            kernel_size=7,
            use_skip_connections=False,
            use_batch_norm=True,
            use_weight_norm=False,
            use_layer_norm=False
        )
        self.units = 32
        self.pre_lens = pre_lens
        self.window = window
        self.is_teachers = is_teachers
        self.dense_linear = Dense(self.pre_lens, activation='linear')
        self.dense_relu = Dense(self.pre_lens, activation='relu')
        self.dense_sigmoid = Dense(self.pre_lens, activation='sigmoid')
        self.dense_tanh = Dense(self.pre_lens, activation='tanh')

定义好后 可在call function中自由使用