tensorflow / tensorflow/probability
Feature Request: Alternative Horseshoe Parameterization as tfp distribution
Nobody has claimed this yet.
- Dominant language
- Jupyter Notebook
- Stars
- 4.4k
- Forks
- 1.1k
- PR merge metrics
- No merged PRs in 30d
Description
The currently implemented version of the horseshoe distribution is not the parameterization that most ML papers use. This limits the ease of use of this as, for example, a prior in a tfp.layers.KLDivergenceAddLoss or in tfp.layers.DenseReparameterization. The regularized horseshoe would also be useful as an implemented distribution.
The alternative parameterization is shown here:
https://www.tensorflow.org/probability/api_docs/python/tfp/sts/SparseLinearRegression
And here:
http://proceedings.mlr.press/v80/ghosh18a/ghosh18a.pdf
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 comparing the existing horseshoe distribution with the alternative parameterization documented for tfp.sts.SparseLinearRegression and described in the linked paper. Review how the distribution is intended to work with tfp.layers.KLDivergenceAddLoss and tfp.layers.DenseReparameterization; done means the alternative and regularized horseshoe forms are available as usable TFP distributions.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100