alimeituan

请教下这个loss的依据是啥啊，怎么y_true没有参与啊？有点奇怪啊 ``` def sampledsoftmaxloss(y_true, y_pred): return K.mean(y_pred) ```

在用户特征中为啥有个hist_len这个特征？历史观看的movie长度也是个特征吗？？ ``` >>> build_input_features(user_feature_columns) OrderedDict([('user_id', ), ('gender', ), ('age', ), ('occupation', ), ('zip', ), ('hist_movie_id', ), ('hist_len', )]) ``` 另外这些特征都是单个值的，假如有多标签或者说多值的话怎么办呢？

请教下，计算得到的user和item的vector是按照lbe编码的顺序的吗？能对应上吗？

And the validation's cost is large,about 50, Now step :5500, learning_rate: 0.000098

Warning : NaN or Inf found in input data. How to do ? 

    Not good

learning_rate reduce, but cost no reduce, What's the matter ?

After one day,   No update, No reduce but the learning_rate. And I no idea

The bug again:  Does it mean the training is over and will write the wav ? But the result is not good according to the valid_cost. I don't know...

Should I give up ? Or just go on ? But Who can help me ?