Msprime

This is a major update to the underlying data structures in msprime to
generalise the information that can be modelled, and allow
for data from external sources to be efficiently processed. The
new Tables API enables efficient interchange of tree sequence data using
numpy arrays. Many updates have also been made to the tree sequence
API to make it more Pythonic and general. Most changes are backwards
compatible, however.

**Breaking changes**:

- The `SparseTree.mutations()` and `TreeSequence.mutations()` iterators no
longer support tuple-like access to values. For example, code like

> > for x, u, j in ts.mutations():
>
> : print("mutation at position", x, "node = ", u)
>
will no longer work. Code using the old `Mutation.position` and
`Mutation.index` will still work through deprecated aliases,
but new code should access these values through `Site.position`
and `Site.id`, respectively.
- The `TreeSequence.diffs()` method no longer works. Please use
the `TreeSequence.edge_diffs()` method instead.
- `TreeSequence.get_num_records()` no longer works. Any code using
this or the `records()` iterator should be rewritten to work with
the `edges()` iterator and num_edges instead.
- Files stored in the HDF5 format will need to upgraded using the
`msp upgrade` command.

**New features**:

- The API has been made more Pythonic by replacing (e.g.)
`tree.get_parent(u)` with `tree.parent(u)`, and
`tree.get_total_branch_length()` with `tree.total_branch_length`.
The old forms have been maintained as deprecated aliases. (64)
- Efficient interchange of tree sequence data using the new Tables
API. This consists of classes representing the various
tables (e.g. `NodeTable`) and some utility functions (such
as `load_tables`, `sort_tables`, etc).
- Support for a much more general class of tree sequence topologies.
For example, trees with multiple roots are fully supported.
- Substantially generalised mutation model. Mutations now occur at
specific sites, which can be associated with zero to many mutations.
Each site has an ancestral state (any character string) and
each mutation a derived state (any character string).
- Substantially updated documentation to rigorously define the
underlying data model and requirements for imported data.
- The `variants()` method now returns a list of alleles for each
site, and genotypes are indexes into this array. This is both
consistent with existing usage and works with the newly generalised
mutation model, which allows arbitrary strings of characters as
mutational states.
- Add the formal concept of a sample, and distinguished from 'leaves'.
Change `tracked_leaves`, etc. to `tracked_samples` (225).
Also rename `sample_size` to `num_samples` for consistency (227).
- The simplify() method returns subsets of a large tree sequence.
- TreeSequence.first() returns the first tree in sequence.
- Windows support. Msprime is now routinely tested on Windows as
part of the suite of continuous integration tests.
- Newick output is not supported for more general trees. (117)
- The `genotype_matrix` method allows efficient access to the
full genotype matrix. (306)
- The variants iterator no longer uses a single buffer for
genotype data, removing a common source of error (253).
- Unicode and ASCII output formats for `SparseTree.draw()`.
- `SparseTree.draw()` renders tree in the more conventional 'square
shoulders' format.
- `SparseTree.draw()` by default returns an SVG string, so it can
be easily displayed in a Jupyter notebook. (204)
- Preliminary support for a broad class of site-based statistics,
including Patterson's f-statistics, has been added, through
the {}`SiteStatCalculator`, and its branch length analog,
{}`BranchLengthStatCalculator`. The interface is still in development,
and is expected may change.

**Bug fixes**:

- Duplicate site no longer possible (159)
- Fix for incorrect population sizes in DemographyDebugger (66).

**Deprecated**:

- The `records` iterator has been deprecated, and the underlying data
model has moved away from the concept of coalescence records. The
structure of a tree sequence is now defined in terms of a set of nodes
and edges, essentially a normalised version of coalescence records.
- Changed `population_id` to `population` in various DemographicEvent
classes for consistency. The old `population_id` argument is kept as a
deprecated alias.
- Changed `destination` to `dest` in MassMigrationEvent. The old
`destination` argument is retained as a deprecated alias.
- Changed `sample_size` to `num_samples` in TreeSequence and
SparseTree. The older versions are retained as deprecated aliases.
- Change `get_num_leaves` to `num_samples` in SparseTree. The
`get_num_leaves` method (and other related methods) that have
been retained for backwards compatibility are semantically incorrect,
in that they now return the number of **samples**. This should have
no effect on existing code, since samples and leaves were synonymous.
New code should use the documented `num_samples` form.
- Accessing the `position` attribute on a `Mutation` or
`Variant` object is now deprecated, as this is a property of a `Site`.
- Accessing the `index` attribute on a `Mutation` or `Variant` object
is now deprecated. Please use `variant.site.id` instead. In general,
objects with IDs (i.e., derived from tables) now have an `id` field.
- Various `get_` methods in TreeSequence and SparseTree have been
replaced by more Pythonic alternatives.

Major release providing new functionality and laying groundwork for
upcoming functionality.

**Breaking changes**:

- The HDF5 file format has been changed to allow for non-binary trees
and to improve performance. It is now both smaller and faster to
load. However, msprime cannot directly load tree sequence files
written by older versions. The `msp upgrade` utility has been
developed to provide an upgrade path for existing users, so that
files written by older versions of msprime can be converted to the
newer format and read by version 0.4.x of msprime.
- The tuples returned by the `mutations` method contains an element.
This will break code doing things like

> > for pos, node in ts.mutations():
>
> : print(pos, node)
>
For better forward compatibility, code should use named attributes
rather than positional access:

> > for mutation in ts.mutations():
>
> : print(mutation.position, mutation.node)
>
- Similarly, the undocumented `variants` method has some major changes:

1. The returned tuple has two new values, `node` and `index`
in the middle of the tuple (but see the point above about using
named attributes).
2. The returned genotypes are by default numpy arrays. To revert
to the old behaviour of returning Python bytes objects, use the
`as_bytes` argument to the `variants()` method.

**New features**:

- Historical samples. Using the `samples` argument to `simulate`
users can specify the location and time of all samples explicitly.
- HDF5 file upgrade utility `msp upgrade`
- Support for non-binary trees in the tree sequence, and relaxation
of the requirements on input tree sequences using the read_txt()
function.
- Integration with numpy, with zero-copy access to the low-level C API.
- Documented the variants() method that provides access to the sample
genotypes as either numpy arrays or Python bytes objects.
- New LdCalculator class that allows very fast calculation of r^2 values.
- Initial support for threading.
- The values returned mutations() method now also contain an `index`
attribute. This makes many operations simpler.
- New TreeSequence.get_time() method that returns the time a sample
was sampled at.

**Performance improvements**:

- File load times substantially reduced by pre-computing and storing
traversal indexes.
- O(1) implementation of TreeSequence.get_num_trees()
- Improved control of enabled tree features in TreeSequence.trees()
method using the `leaf_lists` and `leaf_counts` arguments.

**Bug fixes**:

- Fixed a precision problem with DemographyDebugger. 37
- Segfault on large haplotypes. 29

Feature release adding new import and export features to the API
and CLI.

- New `TreeSequence.write_records` and `TreeSequence.write_mutations`
methods to serialise a tree sequence in a human readable text format.
- New `msprime.load_txt()` method that parses the above formats, and
allows msprime to read in data from external sources.
- New `TreeSequence.write_vcf` method to write mutation information
in VCF format.
- Miscellaneous documentation fixes.

Feature release adding population related methods to the API.

- New `TreeSequence.get_population(sample_id)` method.
- New `TreeSequence.get_samples(population_id)` method.
- Added the optional `samples` argument to the
`TreeSequence.get_pairwise_diversity` method.
- Fixed a potential low-level buffer overrun problem.

Bugfix release affecting all users of the Python API. Version 0.2.0 contained a
confusing and inconsistent mix of times and rates being expressed in both
coalescent units and generations. This release changes _all_ times and rates
used when describing demographic models to generations, and also changes
all population sizes to be absolute. In the interest of consistency, the
units of the trees output by msprime are also changed to generations. This
is a major breaking change, and will require updates to all scripts using the
API.

This release also include some performance improvements and additional
functionality.

Mspms users are not affected, other than benefiting from performance
improvements.

Breaking changes:

- Time values are now rescaled into generations when a TreeSequence is
created, and so all times associated with tree nodes are measured in
generations. The time values in any existing HDF5 file will now be
interpreted as being in generations, so stored simulations must be
rerun. To minimise the chance of this happening silently, we have
incremented the file format major version number, so that attempts
to read older versions will fail.
- Growth rate values for the PopulationConfiguration class are now
per generation, and population sizes are absolute. These were in
coalescent units and relative to Ne previously.
- GrowthRateChangeEvents and SizeChangeEvents have been replaced with
a single class, PopulationParametersChange. This new class takes
an initial_size as the absolute population size, and growth_rate
per generation. Since the change in units was a breaking one,
potentially leading to subtle and confusing bugs, we decided that
the name refactoring would at least ensure that users would need
to be aware that the change had been made. This API should now
be stable, and will not be changed again without an excellent
reason.
- MigrationRateChangeEvent has been renamed to MigrationRateChange
and the migration rates are now per-generation.
- MassMigrationEvent has been renamed to MassMigration, and the
values of source and destination swapped, fixing the bug in
issue 14.
- The TreeSequence.records() method now returns an extra value,
potentially breaking client code.

Improvements:

- Added tutorial for demographic events.
- Added DemographyDebugger class to help view the changes in populations
over time.
- Added population tracking for coalescent events. We can now determine
the population associated with every tree node. The relevant information
has been added to the HDF5 file format.
- Improved performance for replication by reusing the same low-level
simulator instance. This leads to significant improvements for large
numbers of replicates of small simulations. Issue 8.
- Changed the TreeSequence.records() method to return named tuples.
- Added get_total_branch_length method. Issue 12.
- Fixed bug in reading Hapmap files. Issue 13.

Major update release, adding significant new functionality to the Python
API and several breaking changes. All code written for the 0.1.x API
will be affected, unfortunately.

Breaking changes:

- Sample IDs are now zero indexed. In previous versions of msprime, the
samples were numbered from 1 to n inclusive, which is not Pythonic.
This change has been made to make the API more usable, but will
cause issues for existing code.
- There is now an Ne parameter to simulate(), and recombination,
mutation and migration rates are now all per-generation. The
keyword arguments have been changed to recombination_rate
and mutation_rate, which should mean that silent errors will
be avoided. All rates in existing code will need to be
divided by 4 as a result of this. This change was made to make
working with recombination maps and per generation recombination
rates easier.
- Msprime now uses continuous values to represent coordinates, and
the num_loci parameter has been replaced with a new length parameter
to simulate(). Internally, a discrete recombination model is still
used, but by default the potential number of discrete sites is
very large and effectively continuous. True discrete recombination
models can still be specified by using the recombination_map
argument to simulate.
- The population_models argument to simulate() has been removed, and
replaced with the population_configuration and demographic_events
parameters. This was necessary to provide the full demographic
model.
- The HDF5 file format has been updated to accommodate the continuous
coordinates, along with other minor changes. As a consequence,
simulation results will be somewhat larger. Stored simulations will
need to be re-run and saved.
- Removed the random_seed key from the provenance JSON strings.
- Removed the simulate_tree() function, as it seemed to offer little
extra value.

New features:

- Simulation of variable recombination rates via arbitrary recombination
maps.
- Full support for population structure and demographic events.
- API support for replication via the num_replicates argument to simulate().
- Fully reworked random generation mechanisms, so that in the nominal
case a single instance of gsl_rng is used throughout the entire
simulation session.
- Addition of several miscellaneous methods to the TreeSequence API.
- Added NULL_NODE constant to make tree traversals more readable.