Adding skip-thoughts model #29
Conversation
nadvornix
commented
May 18, 2016
nadvornix
commented
- Adding skipthoughts model.
- Running it on wang/yodaqa/para/semeval-sts/sick2014 datasets.
| @@ -0,0 +1,194 @@ | |||
| """ | |||
| A simple model based on skipthoughts sentence embeddings. | |||
There was a problem hiding this comment.
Can you please add a literature reference, and an appropriate README entry?
There was a problem hiding this comment.
Also, can you think of some single-sentence summary of what the skipthoughts model does? Like "A previously proposed model based on bidi-RNN-with-memory trained to predict preceding and followup words", is that remotely correct? My memory is a bit hazy.
|
Brilliant, I'm super-glad that we have this PR here now. :) However, I think in order to merge this in a maintainable form, we should discuss what the motivation is behind not using standard ptscorers but rather a custom compiled model. Can you please expand on that? What would happen if we used the standard ptscorer with absdiff merge_mode? I'm afraid that this way, the code will really go quickly out of date and break, as you note in the xxx comment, so I'm trying to gauge the level of effort required to offload this back to ptscorer. |
|
Jirka, just a soft ping if you'd find a few moments to take a look. :) |
|
Hi. I am sorry but I am rather busy over summer. I think I will not be able to find time to do this. |
