Changes in version 0.3.4                        

  - Tests only; nothing in the package changed. One test expected
    /bin/sh back unchanged from the interpreter lookup, which
    canonicalises the directory: on a Linux with a merged /usr (current
    Ubuntu) that is /usr/bin/sh. It now uses a file of its own. The
    tests that need a POSIX shell, 0700 permissions or POSIX paths are
    skipped on Windows, which is not a supported platform. (Seen in the
    R-universe checks of 0.3.3.)

                        Changes in version 0.3.3                        

Bug fixes

  - With Python 3.13 or later, interrupting ss_link() could, rarely,
    leave the Python bridge hung: R stopped it only after
    spacescans.interrupt_grace (10 s), and a bridge whose R session had
    died stayed behind. A cancelled job could also be reported as failed
    instead of interrupted. Fixed in the Python package.

                        Changes in version 0.3.2                        

New features

  - workers now also covers the raster weights (NPS Soundscape Noise,
    VIIRS Nighttime Lights, TEMIS UV Exposure, ACAG), the distances of
    NHD Bluespace and TIGER Road Distance, and the HDF4 path of the
    TEMIS reader: Step 1 of every catalog source uses the workers. The
    result still does not depend on their number.

                        Changes in version 0.3.1                        

New features

  - ss_link(workers =) and ss_run(workers =): worker processes per
    pipeline step for the polygon-overlap weights (block groups, tracts,
    ZCTAs) and ACAG's NetCDF files. The default NULL is one process, as
    in 0.3.0. The result does not depend on the number of workers; each
    worker adds memory.

                        Changes in version 0.3.0                        

Breaking change

  - ss_link() returns one row per participant by default. The new
    argument output_level = c("participant", "episode") chooses the
    level; use output_level = "episode" for the behaviour of 0.2 (your
    data frame with the exposure columns added, one row per location
    episode).
  - output_level is the third argument of ss_link(): a call that passed
    data_dir by position must name it (data_dir = ).

New features

  - Participant level: pid (in the type of your id column) and one
    column per exposure, one row per participant in order of first
    appearance, including participants without any value. Each value is
    the mean of the participant's episode values weighted by
    contributing days, counted per variable (see ?ss_link). The result
    carries attr(, "output_level").
  - Linkage rates (attr(res, "coverage")) and the warning for a variable
    with a value for fewer than 90% of the rows count participants at
    participant level and episodes at episode level. ss_coverage() still
    estimates per location episode.
  - Registered datasets: ss_register_dataset() checks and prepares your
    own data once (polygon tables keyed by a Census geography, static,
    yearly or daily; GeoTIFF rasters, static, yearly, biweekly or daily;
    point, line or polygon features, linked as the distance to the
    nearest feature) and stores it with its metadata in a registry
    shared with the Python package. ss_datasets(), ss_dataset(),
    ss_update_dataset() and ss_remove_dataset() list, show, change and
    remove datasets; the option spacescans.datasets_dir (or the
    environment variable SPACESCANS_DATASETS_DIR) chooses the registry.
  - ss_link(custom = ) and ss_coverage(custom = ) take registered
    dataset names, ids, variable keys or records, alone or together with
    ss_custom_polygon() and ss_custom_raster() definitions, in the order
    given; variables may then be empty.
  - Jobs (ss_link(wait = FALSE), ss_job_attach()) work at both levels;
    job.json records the level and counts, never participant ids. A job
    directory written by 0.2 is read as episode level.
  - ss_link() refuses, with an ss_install() hint, a Python package that
    returns episodes when participants were asked for.
  - ss_install() works after an install from r-universe: it reads the
    repository from RemoteUrl and installs the Python package from the
    recorded commit (RemoteSha), or from the tag of the package version
    when no commit is recorded.

Package

  - Released under the MIT License (LICENSE).
  - Title: "Link Time-Referenced Geolocation Data to the Spatial and
    Contextual Exposome".

Documentation

  - The vignette and the README use the terms location episode,
    participant and time-referenced geolocation data, explain both
    output levels, and show two example scripts installed with the
    package: inst/examples/minimal-example.R (three synthetic
    participants and the NPS noise model, the catalog source whose files
    are all on the project mirror) and
    inst/examples/registered-dataset.R (a registered features dataset,
    no catalog data needed). The live tests run both.
  - ?ss_link and the vignette state two rules of the participant level:
    overlapping location episodes each contribute their own days, and
    variables derived after aggregation (the FARA flags, the ACAG _nbm
    columns) are derived from the participant's means.