patrickm663 / hmd.jl Goto Github PK

Hey, great work on this.

I am currently using some HMD data for my work, and would really like to be able to store datasets directly into the packages (i.e. without asking end users for credentials). Taking a look at your code, I wander if you considered this option: did you saw somewhere a clause that does not allow you to redistribute the datas ?

I could for example think of a routine check in CI, ran say weekly or monthly, that would download and store the main zip file in some artefact dependency.

What do you think ?

Edit: From their user agreement, it looks like the datasets are CC BY 4.0, which means that we are free to redistribute them.

It seems they need a combine() call (due to duplicate Ages before being processed with unstack()). In addition, a flag so that users don't call :Cohort accidentally :)

Ideally, we don't want our passwords sitting in plain text. It should be possible to assign environment variables in one's shell (or as user input that just persists throughout) to keep things hidden.

This will also help from a testing perspective to improve coverage.

Some thoughts:

• Population pyramid plots
• Lee-Carter and co models using JuliaActuary packages
• Basic plots and analysis of life expectancy etc.
• Basic life insurance calculations using JuliaActuary packages

More ideas and any PRs welcome!

See JuliaData/CSV.jl#1124 Hopefully a patch is coming soon upstream! Otherwise, will look at building a reader and writer myself :)

When dealing with mortality data, it is typical to 'square' it into an $n\times m$ matrix, where entries $m(x,t)$ correspond to the mortality rate for a life aged $x$ in cohort/year $t$.

A function like transform() & transform!() could take in mortality date and apply this transformation. Where there are multiple representations of the mortality data (e.g. males, females, total), a Dict could be returned with DataFrame objects as values, so t["Males"] returns a DataFrame of transformed Male rates. Similarly, a custom struct could work.