“Dongxu Zhang 2015-09-28”版本间的差异

来自cslt Wiki
跳转至: 导航搜索
(以“Last Week: 1.Initialing pooling method with pretrained transE start with a low performance before training since prediction method is different. So it still needs it...”为内容创建页面)
 
 
第1行: 第1行:
 
Last Week:
 
Last Week:
1.Initialing pooling method with pretrained transE start with a low performance before training since prediction method is different. So it still needs iterations to  before divergence.(one iteration needs 1.5 hour, tens of iterations needed.) Working on speed up pooling method and also produce the attention method.  
+
* Initialing pooling method with pretrained transE start with a low performance before training since prediction method is different. So it still needs iterations to  before divergence.(one iteration needs 1.5 hour, tens of iterations needed.) Working on speed up pooling method and also produce the attention method.  
2. Help huilian do the poem prediction.
+
* Help huilian do the poem prediction.
  
 
This week:
 
This week:
1. Run the program to see if the idea is doable.(running)
+
* Run the program to see if the idea is doable.(running)
2. fix the speed issue.
+
* fix the speed issue.

2015年9月28日 (一) 02:15的最后版本

Last Week:

  • Initialing pooling method with pretrained transE start with a low performance before training since prediction method is different. So it still needs iterations to before divergence.(one iteration needs 1.5 hour, tens of iterations needed.) Working on speed up pooling method and also produce the attention method.
  • Help huilian do the poem prediction.

This week:

  • Run the program to see if the idea is doable.(running)
  • fix the speed issue.