mohamad maheri
|
e9474085ef
|
change hyper parameters
|
2 years ago |
mohamad maheri
|
de53dda45f
|
Implementing Robustify Negative sampling and undo prepossessing
|
2 years ago |
mohamad maheri
|
dbbec9facc
|
repair bug of skipping in sequence
|
2 years ago |
mohamad maheri
|
52f474670f
|
using GRU instead of LSTM
|
2 years ago |
mohamad maheri
|
703b6e570b
|
could skip sequence (p=0.5)
|
2 years ago |
mohamad maheri
|
372e75db43
|
hyper parameter tuning including neg_size
|
3 years ago |
mohamad maheri
|
c89c660a45
|
hyper parameter tuning including neg_size
|
3 years ago |
mohamad maheri
|
8432f281de
|
implement top1
|
3 years ago |
mohamad maheri
|
dca86acb3a
|
implement bpr and bpr_max loss
|
3 years ago |
mohamad maheri
|
703d2aaf76
|
bpr loss + 5X NS
|
3 years ago |
mohamad maheri
|
6af0db43ad
|
Distribute NS to all member of the sequence (except last item of seq in support-set)
|
3 years ago |
mohamad maheri
|
2a680a294f
|
increase NS size (10X) + choose K most scored for NS
|
3 years ago |
mohamad maheri
|
de111c73dc
|
negative sampling by choosing from other user's items
|
3 years ago |
mohamad maheri
|
b79f900501
|
make sampler to choose only sequenced items + write results
|
3 years ago |
mohamad maheri
|
8e21c390fc
|
cuda setting fixed
|
3 years ago |
mohamad maheri
|
5c17c5eb38
|
bug fixing for hyper parameter tuning
|
3 years ago |
mohamad maheri
|
6029ca93a1
|
using ray for hyper-parameter tanning
|
3 years ago |
mohamad maheri
|
9cbfd2aff6
|
input all sequence of items to a LSTM (instead of averaging pairs)
|
3 years ago |
mohamad maheri
|
1c015e3b35
|
using LSTM instead of MLP
|
3 years ago |
mohamad maheri
|
f3bf760148
|
first commit
|
3 years ago |
mohamad maheri
|
9e1c651cfe
|
first commit
|
3 years ago |