Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Implement from_numpyro #811

Merged
merged 11 commits into from
Sep 12, 2019
Merged

Implement from_numpyro #811

merged 11 commits into from
Sep 12, 2019

Conversation

fehiepsi
Copy link
Member

@fehiepsi fehiepsi commented Sep 9, 2019

This PR addresses #810.

I marked some TODOs for future PRs (when NumPyro MCMC object stores model arguments/keywords for the purpose of extracting log_likelihoods, observed_data). Currently, users need to use numpyro.handlers, which is not handy, to obtain these informations.

~Currently, the names of sample stats in NumPyro are:

  • potential_energy
  • num_steps (which can be converted to tree depth by using np.log2 function)
  • accept_prob
  • mean_accept_prob (mean acceptance probability until current iteration during warmup adaptation or sampling)
  • diverging
  • adapt_state.step_size
  • adapt_state.inverse_mass_matrix

Do I need to rename those fields in sample_stats_to_xarray function to match arviz conventions?~

Resolution: rename fields to match PyMC3 stat names.

@ahartikainen
Copy link
Contributor

Maybe you could rename the variables that are common between stan and pymc3 so when we get more diagnostic plots we can have good defaults.

Also, you might add original names to attrs?

Maybe also have both tree depth and num_steps?

@fehiepsi
Copy link
Member Author

fehiepsi commented Sep 9, 2019

rename the variables that are common between stan and pymc3

Good idea! Thanks @ahartikainen , I'll make corresponding changes.

@canyon289
Copy link
Member

Wow this is great. I accidentally created a mild conflict when I merged in another docs pr but it should be a straightforward fix :) Let me know if you need any help

@canyon289
Copy link
Member

One question I do have, Are you planning on fixing the todos in this pr or in a later pr?

@fehiepsi
Copy link
Member Author

Hi @canyon289 , currently, supporting these TODOs (extracting log_likelihood and observed_rvs) is not possible with NumPyro MCMC object (because we don't store model's args, kwargs). Though it is an easy fix, we have to wait for the next NumPyro release to solve these TODOs in arviz. I think I should mark this PR as WIP and wait for that release. :)

@fehiepsi fehiepsi changed the title Implement from_numpyro [WIP] Implement from_numpyro Sep 10, 2019
@canyon289
Copy link
Member

canyon289 commented Sep 10, 2019

If you want we could make Github issues to track the todos and merge this in since there's good functionality included already! It also reduces your chance of further merge conflicts.

Let me know what you think :)

@fehiepsi
Copy link
Member Author

make Github issues to track the todos

Hi @canyon289 , that sounds a good plan to me because these TODOs are just for additional features (the API will not change). It has taken us a few months just for the last release but we hope that from now on, we will have more frequent release cycles. ^^

@fehiepsi fehiepsi changed the title [WIP] Implement from_numpyro Implement from_numpyro Sep 12, 2019
@canyon289
Copy link
Member

Awesome!! Merging

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

3 participants