Tsinfer

**Features**

- When calling `sample_data.add_site()` the ancestral state does not need to be the
first allele (index 0): alternatively, an ancestral allele index can be given
(and if `MISSING_DATA`, the ancestral state will be imputed). ({pr}`718`,
{issue}`686` {user}`hyanwong`)

- The CLI interface now allows recombination rate (or rate maps) and mismatch ratios
to be specified ({pr}`731`, {issue}`435` {user}`hyanwong`)

- The calls to match-ancestors and match-samples via the CLI are now logged
in the provenance entries of the output tree sequence ({pr}`732` and `741`,
{issue}`730` {user}`hyanwong`)

- The CLI interface allows `--no-post-process` to be specified (for details of post-
processing, see "Breaking changes" below) ({pr}`727`, {issue}`721` {user}`hyanwong`)

- matching routines warn if no inference sites
({pr}`685`, {issue}`683` {user}`hyanwong`)

**Fixes**

- `sample_data.subset()` now accepts a sequence_length ({pr}`681`, {user}`hyanwong`)

- `verify` no longer raises error when comparing a genotype to missingness.
({pr}`716`, {issue}`625`, {user}`benjeffery`)

**Breaking changes**:

- The `simplify` parameter is now deprecated in favour of `post_process`, which
prior to simplification, removes the "virtual-root-like" ancestor (inserted by
tsinfer to aid the matching process) then splits the ultimate ancestor into separate
pieces. If splitting is not required, the `post_process` step can also be called as a
separate function with the parameter `split_ultimate=False` ({pr}`687`, {pr}`750`,
{issue}`673`, {user}`hyanwong`)

- Post-processing by default erases tree topology that exists before the first site
and one unit after the last site, to avoid extrapolating into regions with no data.
This can be disabled by calling `post_process` step as a separate function with the
parameter `erase_flanks=False` ({pr}`720`, {issue}`483`, {user}`hyanwong`)

- Inference now sets time_units on both ancestor and final tree sequences to
tskit.TIME_UNITS_UNCALIBRATED, stopping accidental use of branch length
calculations on the ts. ({pr}`680`, {user}`hyanwong`)

**Features**

- Added `ancestor(id)` to `AncestorData` class.
({pr}`570`, {issue}`569`, {user}`hyanwong`)

**Fixes**

- Mark zarr 2.11.0, 2.11.1 and 2.11.2 as incompatible due to `zarr-python`
bugs 965 and 967.
({issue}`643`, {pr}`657`, {user}`benjeffery`)

**Bugfixes**:

- Mutations at non-inference sites are now guaranteed to be fully parsimonious.
Previous versions required a mutation above the root when the input ancestral state
disagreed with the ancestral state produced by the parsimony algorithm. Now fixed by
using the new map_mutations code from tskit 0.3.7 ({pr}`557`, {user}`hyanwong`)

**New Features**:

**Breaking changes**:

- Oldest nodes in a standard inferred tree sequence are no longer set to frequencies ~2
and ~3 (i.e. 2 or 3 times as old as all the other nodes), but are spaced above the
others by the mean time between unique ancestor ages ({pr}`485`, {user}`hyanwong`)

- The `tsinfer.SampleData.from_tree_sequence()` function now defaults to setting
`use_sites_time` and `use_individuals_time` to `False` rather than `True`
({pr}`599`, {user}`hyanwong`)

Bugfix release

**Bugfixes**:

- Fix a bug in the core LS matching algorithm in which the rate of recombination
was being incorrectly computed ({issue}`493`, {pr}`514`, {user}`jeromekelleher`,
{user}`hyanwong`).

- `tsinfer.verify()` no longer requires that non-ancestral alleles in a SampleData
and Tree Sequence file are in the same order ({issue}`490`, {pr}`492`,
{user}`hyanwong`).

**New Features**:

- Inferred ancestral haplotypes may be truncated via
`AncestorData.truncate_ancestors()` to improve performance when inferring large
datasets ({issue}`276`, {pr}`467`, {user}`awohns`).

**Breaking changes**:

- tsinfer now requires Python 3.7

Major feature release, including some incompatible file format and API updates.

**New features**:

- Mismatch and recombination parameters can now be specified via the
recombination_rate and mismatch_ratio arguments in the Python API.

- Missing data can be accomodated in SampleData using the tskit.MISSING_DATA
value in input genotypes. Missing data will be imputed in the output
tree sequence.

- Metadata schemas for population, individual, site and tree sequence metadata
can now we be specified in the SampleData format. These will be included
in the final tree sequence and allow for automatic decoding of JSON metadata.

- Map non-inference sites onto the tree by using the tskit `map_mutations`
parsimony method. This allows us to support sites with > 2 alleles.

- Historical (non-contemporaneous) samples can now be accommodated in inference,
assuming that the true dates of ancestors have been set, by using the concept
of "proxy samples". This is done via the new function
`AncestorData.insert_proxy_samples()`, then setting the new
parameter `force_sample_times=True` when matching samples.

- The default tree sequence returned after inference when `simplify=True` retains
unary nodes (i.e. simplify is done with `keep_unary=True`.


**Breaking changes**:

- The ancestors tree sequence now contains the real alleles and not
0/1 values as before.

- Times for undated sites now use frequencies (0..1), not as counts (1..num_samples),
and are now stored as `tskit.UNKNOWN_TIME`, then calculated on the fly in the
variants() iterator.

- The SampleData file no longer accepts the `inference` argument to add_site.
This functionality has been replaced by the `exclude_positions` argument
to the `infer` and `generate_ancestors` functions.

- The SampleData format is now at version 5, and older versions cannot be read.
Users should rerun their data ingest pipelines.

- Users can specify variant ages, via `sample_data.add_sites(... , time=user_time)`.
If not `None`, this overrides the default time position of an ancestor, otherwise
ancestors are ordered in time by using the frequency of the derived variant (143).

- Change "age" to "time" to match tskit/msprime notation, and to avoid confusion
with the age since birth of an individual (149). Together with the 2 changes below,
this addition bumped the file format to 3.0.

- Add the ability to record user-specified times for individuals, and therefore
the samples contained in them (currently ignored during inference). Times are
added using `sample_data.add_individual(... , time=user_time)` (190).

- Change `tsinfer.UNKNOWN_ALLELE` to `tskit.MISSING_DATA` for marking unknown regions
of ancestral haplotypes (188) . This also involves changing the allele storage to a
signed int from `np.uint8` which matches the tskit v0.2 format for allele storage
(see https://github.com/tskit-dev/tskit/issues/144).

**Bugfixes**:

- Individuals and populations in the SampleData file are kept in the returned tree
sequence, even if they are not referenced by any sample. The individual and population
ids are therefore guaranteed to stay the same between the sample data file and the
inferred tree sequence. ({pr}`348`)

Bugfix release.

- Fix issue caused by upstream changes in numcodecs (136).