Created by: shenshuijing

网络结构如下：

user_input = data_layer(name='input_fea', size=input_dim)
mask_input = data_layer(name='mask_layer', size=mask_dim)
label = data_layer(name="label", size=num_classes)

user clusters

user_hidden = fc_layer(input=user_input, size=64, act=ReluActivation(), param_attr=ParamAttr(sparse_update=True), layer_attr=ExtraLayerAttribute(drop_rate=0.3))
user_hidden = fc_layer(input=user_hidden, size=64, act=ReluActivation())
user_hidden = fc_layer(input=user_hidden, size=32, act=ReluActivation())
user_hidden = fc_layer(input=user_hidden, size=16, act=ReluActivation())

customer clusters

customer_hidden = fc_layer(input=user_hidden, size=mask_dim, act=ReluActivation())
for i in range(3):
    mix = mixed_layer(size=mask_input, input=[dotmul_projection(input=[customer_hidden,mask_input])])
    customer_hidden = fc_layer(input=mix, size=mask_dim, act=ReluActivation())

prediction = fc_layer(input=customer_hidden, size=num_classes, act=SoftmaxActivation())
if not is_predict:
    pred_cost = cross_entropy(input=prediction, label=label)
    evaluator = auc_evaluator(input=prediction, label=label)
    outputs(pred_cost)
else:
    outputs([prediction])

我把网络结构改成上述这样， 生成binary conf的时候错误信息为： 我没有设置过param_attr， 怎么回assert失败， 另外我强制在dotmul_projection中输入param_attr=None, 也是报这个错误