Python 0.3.0

Major feature release

This release adds metadata schemas, set-like operations, mutation times, SVG drawing improvements and many others. This release also comes with wheels for windows, osx and linux.

❤️ Many thanks go to the tskit community and contributors for their awesome work on this release. ❤️

Breaking changes

• The default display order for tree visualisations has been changed to minlex (see below) to stabilise the node ordering and to make trees more readily comparable. The old behaviour is still available with order="tree".

• File system operations such as dump/load now raise an appropriate OSError instead of tskit.FileFormatError. Loading from an empty file now raises and EOFError.

• Bad tree topologies are detected earlier, so that it is no longer possible to create a TreeSequence object which contains a parent with contradictory children on an interval. Previously an error was thrown when some operation building the trees was attempted (@jeromekelleher, #709).

• The TableCollection object no longer implements the iterator protocol. Previously list(tables) returned a sequence of (table_name, table_instance) tuples. This has been replaced with the more intuitive and future-proof TableCollection.name_map and TreeSequence.tables_dict attributes, which perform the same function (@jeromekelleher, #500, #694).

• The arguments to TreeSequence.genotype_matrix, TreeSequence.haplotypes and TreeSequence.variants must now be keyword arguments, not positional. This is to support the change from impute_missing_data to isolated_as_missing in the arguments to these methods (@benjeffery, #716, #794).

New features

• New methods to perform set operations on TableCollections and TreeSequences. TableCollection.subset subsets and reorders table collections by nodes (@mufernando, @petrelharp, #663, #690). TableCollection.union forms the node-wise union of two table collections (@mufernando, @petrelharp, #381 #623).

• Mutations now have an optional double-precision floating-point time column. If not specified, this defaults to a particular NaN value (tskit.UNKNOWN_TIME) indicating that the time is unknown. For a tree sequence to be considered valid it must meet new criteria for mutation times, see mutation requirements. Also added function TableCollection.compute_mutation_times. Table sorting orders mutations by non-increasing time per-site, which is also a requirement for a valid tree sequence (@benjeffery, #672).

• Add support for trees with internal samples for the Kendall-Colijn tree distance metric. (@daniel-goldstein, #610)

• Add background shading to SVG tree sequences to reflect tree position along the sequence (@hyanwong, #563).

• Tables with a metadata column now have a metadata_schema that is used to validate and encode metadata that is passed to add_row and decode metadata on calls to table[j] and e.g. tree_sequence.node(j) See metadata (@benjeffery, #491, #542, #543, #601).

• The tree-sequence now has top-level metadata with a schema (@benjeffery, #666, #644, #642).

• Add classes to SVG drawings to allow easy adjustment and styling, and document the new tskit.Tree.draw_svg() and tskit.TreeSequence.draw_svg() methods. This also fixes #467 for duplicate SVG entity id s in Jupyter notebooks (@hyanwong, #555).

• Add a to_nexus function that outputs a tree sequence in Nexus format (@saunack, #550).

• Add extension of Kendall-Colijn tree distance metric for tree sequences computed by TreeSequence.kc_distance (@daniel-goldstein, #548).

• Add an optional node traversal order in tskit.Tree that uses the minimum lexicographic order of leaf nodes visited. This ordering ("minlex_postorder") adds more determinism because it constraints the order in which children of a node are visited (@brianzhang01, #411).

• Add an order argument to the tree visualisation functions which supports two node orderings: "tree" (the previous default) and "minlex" which stabilises the node ordering (making it easier to compare trees). The default node ordering is changed to "minlex" (@brianzhang01, @jeromekelleher, #389, #566).

• Add _repr_html_ to tables, so that jupyter notebooks render them as html tables (@benjeffery, #514).

• Remove support for kc_distance on trees with unary nodes (@daniel-goldstein, #508).

• Improve Kendall-Colijn tree distance algorithm to operate in O(n^2) time instead of O(n^2 * log(n)) where n is the number of samples (@daniel-goldstein, #490).

• Add a metadata column to the migrations table. Works similarly to existing metadata columns on other tables (@benjeffery, #505).

• Add a metadata column to the edges table. Works similarly to existing metadata columns on other tables (@benjeffery, #496).

• Allow sites with missing data to be output by the haplotypes method, by default replacing with -. Errors are no longer raised for missing data with isolated_as_missing=True; the error types returned for bad alleles (e.g. multiletter or non-ascii) have also changed from _tskit.LibraryError to TypeError, or ValueError if the missing data character clashes (@hyanwong, #426).

• Access the number of children of a node in a tree directly using tree.num_children(u) (@hyanwong, #436).

• User specified allele mapping for genotypes in variants and genotype_matrix (@jeromekelleher, #430).

• New root_threshold option for the Tree class, which allows us to efficiently iterate over 'real' roots when we have missing data (@jeromekelleher, #462).

• Add pickle support for TreeSequence (@terhorst, #473).

• Add tree.as_dict_of_dicts() function to enable use with networkx. See the tutorial (@winni2k, #457).

• Add tree_sequence.to_macs() function to convert tree sequence to MACS format (@winni2k, #727).

• Add a keep_input_roots option to simplify which, if enabled, adds edges from the MRCAs of samples in the simplified tree sequence back to the roots in the input tree sequence (@jeromekelleher, #775, #782).

Bugfixes

• #453 - Fix LibraryError when tree.newick() is called with large node time values (@jeromekelleher, #637).

Deprecated

• The sample_counts feature has been deprecated and is now ignored. Sample counts are now always computed.
• For TreeSequence.genotype_matrix, TreeSequence.haplotypes and TreeSequence.variants the impute_missing_data argument is deprecated and replaced with isolated_as_missing. Note that to get the same behaviour impute_missing_data=True should be replaced with isolated_as_missing=False (@benjeffery, #716, #794).