Tutorial on Bayesian workflow, with a SARS-CoV-2 running example

[fritzo]

open in colab

This also adds the helper utility

• pyro.ops.special.sparse_multinomial_likelihood

[martinjankowiak]

comments:

• typo: intellitently
• toc entry six missing closing ); alternatively change to colon
• MCMC is mentioned in overview but not elsewhere. remove?
• nit: "PANGO lineages of samples" +genomic
• "should lead to an approximately block diagonal covariance matrix" this is ambiguous. you mean "... in the approximate posterior... when using a mvn variational distribution" or something
• "While in principle you could be Bayesian about everything, it's best to focus probabilistic inference on the truly uncertain parts of your problem." out of context, remove?
• 5.5 / dataset["time_step_days"] should be inverted
• more comments/description of model? or explicitly defer/link to intro logistic growth tutorial?
• explain 0.01?
• nit: "is implausibly narrow" => "may be"
• "...is less sensitive to reparametrization." maybe +"w.r.t. interpreting the results of inference or something?" optimization performance is affected in both cases
• more text to make the way you move between log/exp space more clear? "most mutations have very little effect"
• typo: initialzing
• typo: wavely
• nit: HaarReparam +"(where appropriate)"?
• "Try combining multiple guides" => +"using"
• some sort of concluding sentence?

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[fritzo]

Thanks for the detailed review @martinjankowiak, I've tried to address all your review comments.

[martinjankowiak]

looks great!

• typo: its's
• typo: multiplicitave
• missing word: we'll need duplicate
• typo: space,,
• typo: langauges
• typo: initialzation

[fritzo]

done