Question/Feature request: Figuring out how often the model/gradient was evaluated by NUTS
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 9.8k
- Forks
- 2.3k
- Avg merge
- 21h 39m
- Merged PRs (30d)
- 5
Description
This question came up during office hours:
Which sampler stat should be used to answer how often a model was evaluated during MCMC?
@junpenglao @ColCarroll
For example, these variables are tracked by NUTS:
- process_time_diff
- lp
- max_energy_error
- perf_counter_diff
- diverging
- acceptance_rate
- step_size
- energy_error
- perf_counter_start
- n_steps
- tree_depth
- step_size_bar
- energy
If we're not currently tracking the information of how often the model was evaluated, we should definitely add that stat!
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
Start by inspecting the NUTS sampler-stat collection and the listed statistics, especially n_steps and tree_depth, to determine whether any already represent model evaluations. If the count is not tracked, define the statistic's meaning and expose it alongside the existing NUTS stats; done means users can identify how often the model was evaluated during MCMC.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100