TODO before v-gamma release

[hyanwong]

We are planning a tsdate release with variational-gamma set as the default method. Here's what we need to do before release:

Essential

• Decide on which variational_gamma parameters (e.g. expectation iterations) should be documented, if any. FWIW my preference would be to document as few as possible, in case we change them.
• Update the docs (e.g. population_size no longer needed by default)
• Test the CLI and change where necessary (e.g. CLI-use of variational_gamma specific params (e.g. max-shape, rescaling-intervals) #377)
• Add split_disjoint to preprocess_ts (Add split_disjoint_nodes to preprocess routine #373)
• Update changelog (this is a breaking change)

Nice to have

• brief explanation of variational_gamma approach in the docs
• brief explanation of rescaling approach in the docs
• Error out if no mutations in v_gamma variational_gamma method fails on tree seq with no mutations #390 (comment)
• Move population_size and priors out of .date() method #376
• Correctly return the posterior using variational_gamma (variational_gamma does not seem to return posteriors #388)
• Add accuracy tests for variational_gamma (Add variational method accuracy tests to test_accuracy.py #265)
• Perhaps Remove ignore_oldest_root option from CLI #262 ?
• Set permissive_json schema on the mutation table if it is empty and no schema is set.

For future releases

• Unphased singletons ()
• Site masking (Adjust edge spans according to exclusion mask #315)

Does this sound right @nspope ? I will try to tick these all off in the next week or two.

[nspope]

Yes, sounds good!

[hyanwong]

Do you think any specific variational_gamma parameters should be documented, @nspope ? Are there ones that new users might need to tweak in common circumstances?

[nspope]

max_iterations (the number of EP iterations) and rescaling_intervals (which may need to be tweaked if there's weirdness like missing singletons) are the only two that come to mind?

[hyanwong]

Thanks! Should we undocument max_shape then, do you think?

[hyanwong]

(I guess we can also default max_iterations to None, and reserve the right to change it to something adaptive in future.

[nspope]

Sure -- I think the defaults max_shape and min_step seem to be working fine.

[hyanwong]

min_step isn't made available in the signature anyway.

[hyanwong]

The docs are basically done now (although a bit rough-and-ready). Do you want to have a read through them @nspope and see if there is anything I've missed, or if you have suggested changes. It would be nice to get this out of the door.

[hyanwong]

brief explanation of rescaling approach in the docs

This could go here. If you want to provide a 2 or 3 sentence description @nspope, I'll bung it in.

[nspope]

Awesome thanks @hyanwong! I'll take a look in the next day or two.

[hyanwong]

Thanks. I'm assuming you would be happy to make a 0.2 release any time, and we can fix any doc issues later?

[nspope]

Sure, sounds good to me.