pearsonca / cabputils Goto Github PK

• Add linked TOC for major function groups
• update to include current functions

the idea behind overriding commandArgs is to leverage that as a concept - but maybe more appropriate to use a different name?

provide an argument for something like a parsing block, that turns sequence of arguments into assignments in the environment.

• add dependency on jsonlite
• add store method
• develop a config loading function.

optionally, for json vs rda [vs other?] have config loader do more than just make a list object. E.g. load the variables into environment, have default support, including actions associated with variables (e.g., casting, passing to options(...), ...)?

...and a CONTRIBUTING file to specify when to do so.

Need to add all the revised functions.

Create a ggplot scale that uses bands instead of grid lines. MVP:

• interface like scale_x_bands(...)
• breaks specifying the boundaries of the bands, labels centered in alternating bands, band styling set by a relevant theme(panel.grid.band = element_rect(...)) element
• ...

Want more info in error messages:

• in error message: show calling scope / checked function name (i.e. indicate helloworld in example errors)
• in error message: show value / variable suppled to checked function (i.e. 1, c("Alice","Bob"), and "")
• consolidate checker / checker_against backend - should be able to seek to location in sys.call for the gsub, then take some chunk after that as the other variables
• robustify get_pipe... to account for users not defining their functions with check_... names
• provide a check_(x, ...) function that will use other checkers

It is desirable to support having a regularized invocation of store() - i.e. to have a script that generates the same kind of thing (e.g. a data.table) which might get stored to different formats (e.g. csv or rds). Other examples: plot to various image file formats or even an rds for later composition with other plots.

Rather than having the user troubleshoot which arguments to pass by target file type, desire to have them provide all the arguments in ... for whatever storage targets they want to support, and have function internally strip out unnecessary arguments.

One potential issue is overlapping arguments with different meanings, but burn that bridge when I get to it.

Complement to stat_spaghetti, but produces a geom_ribbon-like set of aesthetics, instead of many lines. MVP:

• stat_river(...) interface
• produces central line + boundaries
• defaults to quantiling y by x, but want support for alternative transformations to determine central / bounding lines

Typical scenario comparison problem looks like:

• have a bunch of scenarios, grouping on different elements, with time series of outcomes (some times long format, i.e. two column of outcome and val; sometimes wide on outcome1, outcome2, ...), replicated
• a subset (often single element, but not necessarily) of those represent the reference case(s)
• we want to compute the matched differences of the outcomes
• in a variety of ways (averted, relative averted i.e. effectiveness, cumulative versions of previous, cumulative from a particular point in time)

Desired interface:

• need data.table + scenario key + replication key + reference scenario identifier + outcome identifier + desired calcs
• typical calculations should just be implemented
• want: support for some custom additional calculations
• want: variable time resolution output (i.e., if input is daily time series, support weekly / monthly / yearly / etc time series outputs)
• want smart imputing of input needs elements (e.g. from data.table key, date for date, check for a value column => long format)
• want data.table-like interface
• want quantile.data.table-ready output (as this is typical next step)

should store() ...

• have args = .args? that would be typical for my usage, but might be too-much-magic for other users. also there's an alternative if the commandArgs() ends up doing some assignment as well, with some expressed defaults preferences (e.g., might have step to create a tarfile)
• have obj be the first positional argument, for easier use with |>? again, downside seems to be potential for too-much-magic