tensorflow / tensorflow/probability
MCMC samples fill up GPU memory, how do I get model prediction only.
Nobody has claimed this yet.
- Dominant language
- Jupyter Notebook
- Stars
- 4.4k
- Forks
- 1.1k
- PR merge metrics
- No merged PRs in 30d
Description
I'm trying to fit a BNN using nuts, even though the model is not very large, the mcmc samples fill GPU memory really fast. I wonder f there is away to discard the samples and just compute the model predictions at each step of the NUTS sampler. This would drastically reduce the memory overhead needed to store the samples, which I do not really need.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
No files or tests are named. Start by locating the NUTS sampler's MCMC sample storage and prediction-related entry points, then determine how predictions could be retained without retaining samples. Done should mean model predictions remain available while GPU memory is not filled by stored MCMC samples.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- tensorflow
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 25/100