0.1.6

[0.1.6] - 2024-01-07

Breaking changes

• The standalone preprocess_ts function now defaults to not removing unreferenced
  individuals, populations, or sites, aiming to change the tree sequence tables as
  little as possible.

• get_dates (previously undocumented) has been removed, as posteriors can be
  obtained using return_posterior. The normalize terminology previously used
  in get_dates is changed to standardize to better reflect the fact that the
  maximum (not sum) is one, and exposed via the outside_standardize parameter.

• The Ne argument to date has been deprecated (although it is
  still in the API for backward compatibility). The equivalent argument
  population_size should be used instead.

• The CLI -verbosity flag no longer takes a number, but uses
  action="count", so -v turns verbosity to INFO level,
  whereas -vv turns verbosity to DEBUG level.

• The return_posteriors=True option with method="inside_outside"
  previously returned a dict that included keys start_time and end_time,
  giving the impression that the posterior for node age is discretized over
  time slices in this algorithm. In actuality, the posterior is discretized
  atomically over time points, so start_time and end_time have been
  replaced by a single key time.

• The return_posteriors=True option with method="maximization" is no
  longer accepted (previously simply returned None)

• Python 3.7 is no longer supported.

Features

• A new continuous-time method, "variational_gamma" has been introduced, which
  uses an iterative expectation propagation approach. Tests show this increases
  accuracy, especially at older times. A Laplace approximation and damping are
  used to ensure numerical stability. After testing, the node priors used in this
  method are based on a global mixture prior, which can be refined during iteration.
  Future releases may switch to using this as the default method.

• Priors may be calculated using a piecewise-constant effective population trajectory,
  which is implemented in the demography.PopulationSizeHistory class. The
  population_size argument to date accepts either a single scalar effective
  population size, or a PopulationSizeHistory instance.

• Added support and wheels for Python 3.11

• The .date() function is now a wrapper for the individual dating methods
  (accessible using tsdate.core.dating_methods), which can be called independently.
  (e.g. tsdate.inside_outside(...)). This makes it easier to document method-specific
  options. The API docs have been revised accordingly. Provenance is now saved with the
  name of the method used as the celled command, rather than "command": "date".

• Major re-write of documentation (now at
  https://tskit.dev/tsdate/docs/), to use the
  standard tskit jupyterbook framework.

Bugfixes

• The returned posteriors when return_posteriors=True now return actual
  probabilities (scaled so that they sum to one) rather than standardized
  "probabilities" whose maximum value is one.

• The population size is saved in provenance metadata (as a dictionary if
  it is a PopulationSizeHistory instance)

• preprocess_ts always records provenance as being from the preprocess_ts
  command, even if no gaps are removed. The command now has a (rarely used)
  delete_intervals parameter, which is normally filled out and saved in provenance
  (as it was before). If no gap deletion is done, the param is saved as []