Date |
Person |
start |
leave |
hours |
status
|
2017/2/5
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/6
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/7
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/8
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
11:30 |
20:00 |
6+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/9
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/10
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/11
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
14:30 |
20:00 |
5+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/13
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/14
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/15
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/16
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/17
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/18
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/19
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/21
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
20:00 |
9+ |
- found the reason of loss not go down
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/22
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:00 |
20:00 |
9+ |
- use cos to computer alignments, and loss can go down
- replace original attention with cos attention, partly and fully train
|
Peilun Xiao |
|
|
|
|
Guli |
|
|
|
|
2017/2/27
|
Andy Zhang |
|
|
|
|
Shiyue Zhang |
9:30 |
19:00 |
8+ |
- find the tanh linear problem
- use 20*cos replace tanh, get 44.8 blue
|
Peilun Xiao |
|
|
|
|
2017/2/28
|
Andy Zhang |
12:00 |
19:00 |
7 |
- read theano code of NMT, try to find out how it generated encoder input masks
- did some coding of input masks on baseline_beam, need to be further tested
|
Shiyue Zhang |
9:30 |
19:00 |
8 + |
- add encoding and attention mask
- try to find the problem of mem attention
|
Peilun Xiao |
|
|
|
|