New package architecture

[joshwlambert]

This pull request contains a large number of changes influencing the handling of epidemiological data, the API of the epidist class and its methods, as well as work on numerous helper functions.

Changes to the {epiparameter} library:

• The library has several new categories
  • both disease and pathogen in library Adding disease and pathogen genus to database #63
  • to account for the various ways in which parameters, summary statistics and uncertainty can be reported.
  • There are columns for methodological criteria used when inferring the distribution (e.g. censored, right-truncated).

Changes to the epidist class:

• The epidist class now does not directly read from the {epiparameter} library. Instead, it is a stand alone class that can be constructed by the user.
• To keep the useful functionality of the old epidist() function of reading directly from the library, the epidist_db() function has been added.
• The "encapsulated" functions that were part of the previous version of the epidist class have been replaced with class methods. These have new names that follow from the naming convention of the {distributional} package (which is depended on to handle distributions). Update epidist class #73
• The epidist class can handle continuous (with right-truncation Add right-truncation option for continuous distributions #82) and discretised distributions (using {distributional} or {distcrete}). Allow discrete distributions using {distcrete} #81, Add pdf, cdf, quantiles and random sampling methods for epidist class #83
• The epidist class now follows the structure of having a constructor, helper and validator function.

Addition of an epiparam class:

• In the old code the epiparameter library could be read in and viewed through several functions (epidist(), list_distributions() and pathogen_summary()), but there was no way to handle the library and manipulate it in custom ways. The epiparam class is a data structure for the library and is a subclass of a data.frame to provide printing and subsetting methods. Add epiparam class #74
• The epiparam class follows the structure of having a constructor, helper and validator function.

Addition of a vb_epidist class:

• In order to maintain all the functionality of the epidist class for vector-borne diseases in which there is an intrinsic and extrinsic delay (e.g. incubation period) the vb_epidist class provides a consistent interface by wrapping around the epidist class. Accomodate vector-borne diseases in database #50
• The vb_epidist class has the same probability distribution methods as the epidist class.
• The vb_epidist class follows the structure of having a constructor, helper and validator function.

Addition of JSON validation on library:

• There is a new github action workflow to validate the epiparameter data.
• There are helper functions in utils.R to help update the JSON data and JSON schema from the data dictionary yaml file and library csv file. Write data dictionary #64

Addition of conversion functions:

• Negative binomial
• Geometric

There are class conversion functions (as_epidist() and as_epiparam()). There is also a bind_epiparam() function that allows all classes defined in the {epiparameter} package to be bound to an existing epiparam object.

Helper functions for constructing objects have been added to aid in supplying information manually when constructing an epidist object. For example create_epidist_citation() and create_epidist_metadata().

Two exported functions, calc_dist_params() and create_prob_dist() are added which are mainly used internally but are exported in case users may find them helpful.

Added data protocol vignette that describes the process of building the {epiparameter} library. #68

The pathogen_summary() function has been removed as it's functionality was no longer deemed necessary.

Documentation and testing have been updated in line with the above-mentioned developments.

[pratikunterwegs]

Thanks @joshwlambert, looks good to me overall, and happy to approve. There are a few open conversations though, which seem to be optional suggestions, and I would recommend converting them to new issues to be addressed in a different PR.

Putting down a few things I noticed which might be good to convert to issues or keep in mind for future fixes:

1. The vignette chunks are set to eval = FALSE since the functions are out of date with the current PR - something to keep in mind.
2. There are over 1,300 tests which is potentially unwieldy - suggest eventually compressing tests - for example, a single key-value pair comparison (via a snapshot) for the column types in the parameter database, rather than a single expect for each column.
3. L.209 in epiparam_utils.R has a hardcoded number of columns.
4. Might be good to revisit the function get_percentiles() - haven't tested it but getting names and converting to decimal values, as well as determining the upper and lower percentile values seems a little hacky at the moment - am I right in thinking one value has to be > 50 and one < 50? An understandable choice, but not mentioned in the documentation. These lines in the tests suggests this is the case:

   epiparameter/tests/testthat/test-get_percentiles.R

   Lines 8 to 9 in 13c4f40

   	res <- get_percentiles(c(q_05 = 1, q_45 = 10))
   	expect_identical(res, NA)

   .