yangalan123 / anhp-andtt Goto Github PK

你好，请教个关于代码细节的问题。

在 manager.py 中加载数据的时候 add_bos=False, 但是在模型中却强制写成 add_bos=True，我改成 add_bos=False会报错。

dataset = NHPDataset(_data[_split], event_types, concurrent=False, add_bos=False, add_eos=False)
line 35，https://github.com/yangalan123/anhp-andtt/blob/master/anhp/esm/manager.py

self.add_bos = True
line 64，https://github.com/yangalan123/anhp-andtt/blob/master/anhp/model/xfmr_nhp_fast.py

从我个人理解来看，model.add_bos 和 dataset.add_bos 含义不一样对吗？ model.add_bos 是指每次预测下一个点，所以label序列的长度比输入序列的长度要少一个，后面代码会因此有特殊的处理。

如果我的理解正确的话，那是不是所有的模型 model.add_bos 都应该是 True，这样代码里面一些 if else 的判断就不需要了，比如
line 158-165, https://github.com/yangalan123/anhp-andtt/blob/master/anhp/model/xfmr_nhp_fast.py，这部分实际上都不会执行。

非常感谢。

Hi Alan,

Thank you for your response.

I quite enjoyed the paper you recommended.
I believe that I have some level of understaning about the thinning algorithm, but still having a hard time understanding the mathmetical details behind it.

Also, It got me wondering about the difference between the method used in Neural Hawkes Process where you directly approximate the conditional expectation of t_i and the thinnin algorithm approach.
What would be the benefit of using the thinning algorithm?

Again, thank you for sharing a great work!

Hi authors,
Thank you for sharing the great work along with fixed versions for the previous works: THP and SAHP.
While I was reproducing the results for THP, I ran into a problem.

• NLL: I was able to reproduce the results on stackoverflow but the results on MIMIC easily go down to ~5, which is far below than the reported number ~8. I have only changed lr from 1e-4 to 1e-3. Is it expected or am I missing something here?
• RMSE: it seems like the code for RMSE is not fixed the current version of the code. I believe all the dummy values for event_time and prediction need to be ruled out here:
  

  anhp-andtt/thp/thp_training/Utils.py

  Line 103 in b907f38

  def time_loss(prediction, event_time):

  
  Could you confirm that it has not been fixed yet? If so, do you have a plan to fix it anytime soon?

Lastly, according to the datasets in google drive, each dataset is divided into several fold. Have you used all the folds together to report the numbers in Figure2? Or used a particular fold?
Thank you in advance.