Gumbel softmax trick with beam search
Open
enhancement
help wanted
topic: modules
- Dominant language
- Python
- Stars
- 2.4k
- Forks
- 367
- PR merge metrics
- No merged PRs in 30d
Description
Is it possible to use Gumbel softmax trick with a beam search decoder in the existing implementation of Texar. I want to differentiate through the generated beam search samples into the generator network.
Contributor guide
No contributing guide indexed for this repository
Research direction
Start by locating Texar's existing beam search decoder and generator interfaces, then review how generated samples are represented during decoding. Determine whether Gumbel-softmax gradients can pass through beam-search samples; done requires a defined implementation path and validation of differentiation into the generator network.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, tensorflow
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 20/100