NEW NOTEBOOK - counterfactual inference via calculating excess deaths during Covid-19 era

[drbenvincent]

Hi. This PR contributes a new notebook. The primary topic is calculating excess deaths in England and Wales during the Covid-19 era. The main things of interest are the concepts covered:

• causal inference
• linear regression for time-series data
• Bayesian workflow

Links to Google's Causal Uplift approach are drawn.

Very happy to receive any feedback 🙂

NOTE: it is likely that the checks will fail. I have been having a very hard time with the pre-commit checks. As far as I can work out, there is some auto spelling modification which is applied only to the myst version of the notebook which causes a conflict between the ipynb and myst files. I do not know how to get around this as I have very little experience with commit hooks. I was only able to even commit by bypassing the pre-commit checks with git commit --no-verify.

[drbenvincent]

I would also suggest separating out the binning notebook into a separate PR.

I'll look into this, I didn't realise I committed any changes to this. It shouldn't be there

[drbenvincent]

I would also suggest separating out the binning notebook into a separate PR.

Done. No idea why changes to the binning files got wrapped up in this

[drbenvincent]

Tests pass! It's a miracle!

Although it has autocorrected a URL from www.ons.gov.uk/ to www.owns.gov.uk Why are we getting autocorrection on URLs? That seems like madness.

[drbenvincent]

Right. This is not going to pass unless I purposefully include the wrong URL, which is not sensible. We need to do something about codespell in the precommit hooks, see #385.

[cluhmann]

Can you try using the full URL? That is, "http://" whatever? This seems so suggest that URLs should be excluded from spell-checking.

[drbenvincent]

Can you try using the full URL? That is, "http://" whatever? This seems so suggest that URLs should be excluded from spell-checking.

Odd, I'm already doing that:

[cluhmann]

Or perhaps that's an option that can be enabled on codespell? I didn't look closely.

[drbenvincent]

Looks like there are some args:

pymc-examples/.pre-commit-config.yaml

Lines 106 to 111 in 9de026c

	- repo: https://github.com/codespell-project/codespell
	rev: v2.1.0
	hooks:
	- id: codespell
	files: myst_nbs/
	args: ["--write-changes", "--ignore-words-list", "hist,fpr,fro,lik"]

[cluhmann]

So maybe the solution to the codespell part of #385 is to modify that pre-commit.

[drbenvincent]

Yep. Trying to figure out what the correct argument is

[drbenvincent]

Think I've got it. I'll create a separate PR to fix the codespell URI stuff

[drbenvincent]

Ok. I've filed a separate PR here #388 to fix this.

I initially tried to just edit the precommit file locally. This then does make the local pre-commit checks pass BUT it was not letting me push the changes while the .pre-commit-config.yaml file is unstaged.

So in order to make progress here, we first have to close #388. Then I can rebase or whatever and make progress with this notebook.

[cluhmann]

Yeah, if the only thing holding up this PR is some pre-commit stuff then I think that's probably fine. If we get this PR approved, it will still be blocked until the pre-commit gets fixed. But then this will be automatically unblocked once that happens. So now I guess I can take a look at the actual notebook!

[drbenvincent]

Yeah, if the only thing holding up this PR is some pre-commit stuff then I think that's probably fine. If we get this PR approved, it will still be blocked until the pre-commit gets fixed. But then this will be automatically unblocked once that happens. So now I guess I can take a look at the actual notebook!

Happy to get comments on the content. I believe @juanitorduz will have time to provide additional review next week - we're back to needing 2 approving reviewers.

[review-notebook-app]

View / edit / reply to this conversation on ReviewNB

juanitorduz commented on 2022-07-04T13:02:19Z
----------------------------------------------------------------

Line #22.        mu = pm.Deterministic(

As we are using a linear model I think it would be nice to scale the data ... maybe through a https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.StandardScaler.html ?

[juanitorduz]

@drbenvincent this is a very cool notebook! 🚀 Just left a small comment regarding scaling the data, this of course should not be a blocker to merge it.

It would be great to have a follow up notebook where we allow more complex models like a local linear trend and maybe auto-regressive terms? :)

[drbenvincent]

Thanks for everyone's input on this so far. Had some issues with my main branch so life will be simpler switching over to this new PR, #391
I'll close this PR momentarily and do the changes in the new PR

[drbenvincent]

It would be great to have a follow up notebook where we allow more complex models like a local linear trend and maybe auto-regressive terms? :)

Yes, I think they could be idea for the future. Perhaps for a more time-series focussed notebook.