hkxIron

Results 5 issues of hkxIron

https://github.com/yaroslavvb/stuff/blob/8b3cb66cd2bedd90c521e2402cca66afd8a766bd/ericyue-slowreader/benchmark.py#L59 just as following _, serialized_example = reader.read_up_to(filename_queue, enqueue_many_size *batch_size)

1.看了整个代码,感觉作者并没有将 last_slot_label信息加入 decoder的解码中 2.这个与Attention-Based Recurrent Neural Network Models for Joint Intent Detection and Slot Filling的实现好像不太一样,感觉作者的方法还是encoder-decoder-attention,即论文中的第二种方法

https://github.com/gutouyu/ML_CIA/blob/37838eb655d3e432393cee7dda11ea693217eb42/DeepFM/DeepFM.py#L232 deepfm中的二阶特征交叉后,并没有直接相加,而是后面与一阶特征组合起来,乘以w,所以我理解二阶特征的维度是 m*(m-1)/2,其中m为feature_group的个数,望回复,谢谢

你好,论文能上传一下么? 网上无法下载

I have implement a tensor version based on numpy.ndarray, code: https://github.com/hkxIron/tensorgrad