Release: 0.2.0

• Breaking: crimepy keeps its data in amos's workspace (amos 0.6.0), not
  in a "crimepy_data" folder of the home folder. The data folder is now the
  "crime" folder (options.DATA_SUBFOLDER) of the data folder of the
  workspace that holds the "code" folder (such as "projects/data/crime" for
  code in "projects/code"; see amos's Clerk), so NIBRS is in
  "data/crime/nibrs", beside the "studies" folder of amos projects' results.
  CRIMEPY_DATA, options.DATA_FOLDER (now None unless
  CRIMEPY_DATA is set), and a folder argument still name another. Move
  an existing "crimepy_data" folder's "nibrs" folder there (or set
  CRIMEPY_DATA) to keep its downloads and tables. Requires amos 0.6.0.

Release: 0.1.0

• Breaking: crimepy is built on amos datasets.
  crimepy.nibrs.load, load_tables, read, and tables.build (and its
  builders) return amos.Datasets, whose data is the table, and take
  label, task, and groups for modeling. crimepy.nibrs.load_frame
  returns a plain data frame, as load did. crimepy.combine.stack and
  merge return datasets, and crimepy.analyze.count returns a new dataset
  of counts, whose label is the counts.
• Every analysis is an amos technique: population_rates, crime_rates,
  percent_change, crosstab, and coverage join count_crimes,
  crime_trend, and clearance_rates. The functions of crimepy.analyze
  apply them, take datasets or data frames, and keep their tables in a
  dataset's tables. crime_rates loads the agencies of the data's years
  and states when it is given none.
• The functions of crimepy.plot take datasets, which keep their figures in
  figures, and crimepy.nibrs.decode decodes a dataset in place.
• Downloads of master files that break (as they do when the CDE's 15-minute
  link expires during a slow download) resume where they stopped, with a new
  link.
• crimepy requires amos 0.4.1.
• The README has a BibTeX entry for citing crimepy, and CITATION.cff has
  the date of the release.