new Pareto-k diagnostics vignette

[avehtari]

Fixes #322

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 44fc43e is merged into master:

• ✔️as_draws_array: 146ms -> 145ms [-1.37%, +0.64%]
• 🚀as_draws_df: 71ms -> 68.3ms [-5.86%, -1.68%]
• ✔️as_draws_list: 170ms -> 170ms [-0.96%, +0.77%]
• ✔️as_draws_matrix: 28.2ms -> 28.3ms [-0.53%, +1.51%]
• ✔️as_draws_rvars: 78.9ms -> 79.3ms [-0.25%, +1.11%]
• ✔️summarise_draws_100_variables: 697ms -> 697ms [-0.18%, +0.2%]
• ✔️summarise_draws_10_variables: 111ms -> 111ms [-1%, +1.13%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[paul-buerkner]

Thanks! Do you think this vignette needs additional review? If yes, could @n-kall do it since you have been implementing these methods.

[n-kall]

Sure, I can look at this

[n-kall]

I haven't yet had a chance to run the vignette, but as the tests were passing (except on R 3.6) I have gone through and made some suggestions.

Most suggestions are removing the |> operator which is not available in R < 4.1.0. If we want to use the native pipe, I think would need to bump the required version to 4.1.0 (as far as I understand)

[n-kall]

I don't think we can use the |> operator unless we bump the required R version.

Instead this could be written

Suggested change

	dr<-array(data=replicate(4,as.numeric(arima.sim(n = N,
	list(ar = c(phi)),
	sd = sqrt((1-phi^2))))),
	dim=c(N,4,1)) |>
	as_draws_df() |>
	set_variables('xn')
	dr <- array(data = replicate(4, as.numeric(arima.sim(n = N,
	list(ar = c(phi)),
	sd = sqrt((1 - phi^2))))),
	dim = c(N, 4, 1))
	dr <- as_draws_df(dr)
	variables(dr) <- "xn"

[avehtari]

As posterior suggests dplyr, can I use %>%?

[paul-buerkner]

Yes, I think so

[n-kall]

Good point. dplyr is importing magrittr so should work. That will be a much better solution

[avehtari]

Switched to %>%

[n-kall]

Suggested change

	drt |> summarise_draws()
	summarise_draws(drt)

[n-kall]

Would it be better here to use default_convergence_measures()to focus on the convergence diagnostics at first?

[avehtari]

I like it as it is now

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 44fc43e is merged into master:

• ✔️as_draws_array: 137ms -> 137ms [-0.8%, +1.11%]
• ✔️as_draws_df: 65ms -> 64.1ms [-3.1%, +0.31%]
• 🚀as_draws_list: 163ms -> 161ms [-1.97%, -0.4%]
• ✔️as_draws_matrix: 27.6ms -> 27.7ms [-0.34%, +1.4%]
• ✔️as_draws_rvars: 80.2ms -> 81.2ms [-0.63%, +2.99%]
• ✔️summarise_draws_100_variables: 696ms -> 698ms [-0.09%, +0.51%]
• ✔️summarise_draws_10_variables: 110ms -> 110ms [-0.69%, +1.21%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[avehtari]

Thanks Noa.

I made the suggested, except for using %>% and not using default_convergence_measures().

I also fixed the two references in posterior.Rmd. Could have been a separate from this PR, but maybe this is ok

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 44fc43e is merged into master:

• ✔️as_draws_array: 142ms -> 143ms [-0.3%, +1.77%]
• 🚀as_draws_df: 65.3ms -> 63ms [-4.56%, -2.66%]
• ✔️as_draws_list: 158ms -> 157ms [-1.15%, +0.65%]
• ✔️as_draws_matrix: 27.3ms -> 27.4ms [-0.68%, +0.94%]
• ✔️as_draws_rvars: 79.8ms -> 79.5ms [-1.55%, +0.74%]
• ✔️summarise_draws_100_variables: 694ms -> 694ms [-0.2%, +0.2%]
• ✔️summarise_draws_10_variables: 110ms -> 110ms [-1.09%, +0.29%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[avehtari]

Thanks Noa, you have magical eyes for spotting errors!

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 44fc43e is merged into master:

• ✔️as_draws_array: 135ms -> 135ms [-1.32%, +1.17%]
• 🚀as_draws_df: 70.5ms -> 68.3ms [-5.29%, -0.96%]
• 🚀as_draws_list: 162ms -> 160ms [-2.3%, -0.06%]
• ✔️as_draws_matrix: 27ms -> 26.7ms [-2.23%, +0.61%]
• ✔️as_draws_rvars: 80.9ms -> 80.8ms [-1.34%, +1.08%]
• ✔️summarise_draws_100_variables: 675ms -> 678ms [-0.5%, +1.45%]
• ✔️summarise_draws_10_variables: 115ms -> 115ms [-1.11%, +1.21%]
  Further explanation regarding interpretation and methodology can be found in the documentation.

[n-kall]

Looks good! All the vignettes are failing to build with R 3.6 but I think that is due to the knitr / evaluate.
I think this could be merged