Can anyone reappear this project?

[DavidJett]

It seems there are so many limitations and mistakes in the project, the reappearing ability of this project is quite low.

[SimengSun]

I have finished training the model, use approximately 3.5 days.
But now I am having problem testing the model. While running run.lua, I always get this:

/home/simeng/torch/install/bin/luajit: ./summary/beam_search.lua:103: attempt to call field 'TemporalKMaxPooling' (a nil value)
stack traceback:
	./summary/beam_search.lua:103: in function 'generate'
	summary/run.lua:86: in function 'main'
	summary/run.lua:121: in main chunk
	[C]: in function 'dofile'
	...meng/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
	[C]: at 0x00406670

Is there anyone having the same problem?

[DavidJett]

Maybe a few modifications can help. My teammate has successfully finished training and testing.

[edorado93]

@SimengSun , check this out. #17
Also, @DavidJett , were you able to replicate the perplexity results as they are mentioned in the paper? The hyper-parameters in the repo are very different from the ones in the paper. I trained the BOW and attention models using the parameters in the paper and my numbers are a bit off.