**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atomically over time points, so `start_time` and `end_time` have been
replaced by a single key `time`.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/](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 `[]`