Skip to content

Latest commit

 

History

History
454 lines (358 loc) · 14.7 KB

CHANGELOG.md

File metadata and controls

454 lines (358 loc) · 14.7 KB

Change Log

Unreleased

(no unreleased changes yet)

v0.8.0

This release adds a handful of minor functions and expands the documentation.

Added

  • penman.layout.appears_inverted() (#37)
  • Tree branch sorting functions (#31)
    • penman.model.Model.original_order()
    • penman.model.Model.canonical_order()
    • penman.model.Model.random_order()
  • penman.layout.rearrange() (#31)
  • --rearrange=(canonical|random) command-line option

v0.7.2

Added

  • -q / --quiet command line option
  • -v / --verbose command line option
  • Loggers that print some diagnostic information at the DEBUG and INFO levels

Fixed

  • Remove superfluous POP layout markers when graphs have duplicated triples (#34)
  • Avoid KeyError on edge and attribute reification when graphs have duplicated triples (#35)

Changed

  • Model.reify() no longer inverts the incoming triple

v0.7.1

The v0.7.0 release was missing a declaration for the new penman.models sub-package, so it was not available for new installs. This release fixes that.

Fixed

  • The penman.models package is now declared in setup.py
  • The documentation now looks one level up when building (this is just "in case", as I think the missing package problem was the real culprit for faulty documentation builds.)

v0.7.0

This release comprises a major restructuring from previous versions. No longer is there a single penman.py module, but the penman package, which has the following modules:

  • __about__ - package meta-information
  • codec - high-level parsing and formatting
  • epigraph - epigraphical markers
  • exceptions - exception classes
  • graph - triple and graph classes
  • interface - load(), dump(), etc.
  • layout - interpretation and configuration of trees
  • lexer - low-level parsing
  • main - command-line interface
  • model - semantic model class
  • surface - surface alignment information
  • transform - tree and graph transformations
  • tree - tree class
  • types - static type checking definitions

In addition, there is a models sub-package for provided semantic models, although it currently only contains one: models.amr.

While some of the original API is preserved through imports in penman/__init__.py, there are a number of backward-incompatible changes in this release. Changes that affect the old API are listed below, but otherwise the new functionality is under the added modules.

Python Versions

  • Removed support for Python 2.7
  • Removed support for Python 3.3
  • Removed support for Python 3.4
  • Removed support for Python 3.5
  • Added support for Python 3.6
  • Added support for Python 3.7
  • Added support for Python 3.8

Added

  • penman.Graph.epidata
  • penman.Graph.metadata
  • penman.PENMANCodec.parse()
  • penman.PENMANCodec.parse_triples()
  • penman.PENMANCodec.iterparse()
  • penman.PENMANCodec.format()
  • penman.PENMANCodec.format_triples()
  • penman.codec
  • penman.epigraph
  • penman.epigraph.Epidatum
  • penman.exceptions
  • penman.exceptions.GraphError
  • penman.exceptions.LayoutError
  • penman.exceptions.SurfaceError
  • penman.exceptions.ModelError
  • penman.graph
  • penman.graph.CONCEPT_ROLE
  • penman.graph.Edge
  • penman.graph.Attribute
  • penman.layout
  • penman.layout.LayoutMarker
  • penman.layout.Push
  • penman.layout.POP
  • penman.layout.interpret()
  • penman.layout.configure()
  • penman.layout.reconfigure()
  • penman.layout.has_valid_layout()
  • penman.lexer
  • penman.lexer.PATTERNS
  • penman.lexer.PENMAN_RE
  • penman.lexer.TRIPLE_RE
  • penman.lexer.Token
  • penman.lexer.TokenIterator
  • penman.lexer.lex()
  • penman.model
  • penman.model.Model
  • penman.models
  • penman.models.amr
  • penman.models.amr.roles
  • penman.models.amr.normalizations
  • penman.models.amr.reifications
  • penman.models.amr.model
  • penman.surface (#19)
  • penman.surface.AlignmentMarker
  • penman.surface.Alignment
  • penman.surface.RoleAlignment
  • penman.surface.alignments()
  • penman.surface.role_alignments()
  • penman.transform
  • penman.transform.canonicalize_roles()
  • penman.transform.reify_edges() (#27)
  • penman.transform.reify_attributes()
  • penman.transform.indicate_branches()
  • penman.tree (#16)
  • penman.tree.Tree
  • penman.tree.is_atomic()

Removed

  • docopt dependency (#20)
  • penman.EncodeError
  • penman.AMRCodec
  • penman.Triple.inverted
  • penman.PENMANCodec.is_relation_inverted()
  • penman.PENMANCodec.invert_relation()
  • penman.PENMANCodec.handle_triple()
  • penman.PENMANCodec.triples_to_graph()
  • penman.original_order()
  • penman.out_first_order()
  • penman.alphanum_order()

Fixed

  • Graphs can no longer be encoded with attributes as the top (#15)
  • For AMR, both :mod and :domain are for non-inverted relations, although their inverses can be canonicalized to the other (#26)
  • Epigraphical layout markers allow the tree structure to be preserved without modifying the pure graph's triples (#25)

Changed

  • Restructured project as a package
  • Use Sphinx-generated documentation
  • Replaced "relation" with "role" when "role" is intended.
  • Replaced "node type" and "node label" with "concept"
  • Replaced "node identifier" with "variable"
  • Roles now include the colon (:ARG0, not ARG0), following convention
  • penman.PENMANCodec no longer takes the indent or relation_sort parameters
  • penman.PENMANCodec.encode() now takes indent and compact parameters
  • penman.PENMANCodec.iterdecode() works on streams (#21)
  • penman.PENMANCodec now reads comments with metadata (#23)
  • penman.PENMANCodec no longer accepts non-symbol variables (#13)
  • penman.dump() now writes iteratively to a stream (#22)
  • The following no longer take the cls parameter for a codec class, nor **kwargs to configure that class, but instead a model parameter for the semantic model:
    • penman.decode()
    • penman.encode()
    • penman.loads()
    • penman.dumps()
    • penman.load()
    • penman.dump()
  • The following now take the formatting parameters indent and compact:
    • penman.encode()
    • penman.dumps()
    • penman.dump()
  • penman.Graph.triples is now a member variable instead of a method
  • penman.Graph class is mutable (#32)
  • Concepts (node labels) in penman.Graph now have a special role known to the penman.graph module, which can help avoid some reentrancy issues (#29)

v0.6.2

Fixed

  • Value-cast patterns terminated with $ to invalid casts (#9)
  • Raise EncodeError when attempting to encode empty graphs (#14)
  • Redefine NODETYPE_RE for AMRCodec (#17)
  • Remove specific float and int parsing in variables/nodetypes for the default parser (the numeric values are still parses as atoms) (#17)

v0.6.1

Added

  • Some additional regular expressions on PENMANCodec to influence parsing behavior
  • CONTRIBUTING.md

Fixed

  • Allow numeric and string variables and node types

Changed

  • Grammar in README now more accurately reflect parsing behavior (and vice versa)

v0.6.0

Fixed

  • By default, always preserve relation order when given (fixes #6)
  • __version_info__ now gives integers for numeric parts

Added

  • PenmanError as a exception base class
  • EncodeError for errors during encoding; derived from PenmanError
  • Functions for ordering relations (these behave like sorted(), in that they take and return a list). Note that the type relation (given by codec.TYPE_REL), if present, always appears first, regardless of the sorting method.
    • original_order - the default; return the list as-is
    • out_first_order - sort all true orientations before inverted orientations; otherwise order is retained
    • alphanum_order - former default; sort relations alphabetically with embedded integers sorted numerically (like sort -V)
  • Added disclaimer to README.md saying that this module is not affiliated with ISI, the PENMAN project, or the AMR project
  • Triple objects can now have an inverted attribute that can be given as a fourth instantiation parameter. The object is still treated as a 3-tuple, so src, rel, tgt = triple still works (and triple[3] will not get the inversion status; use triple.inverted). Valid values are False (true orientation), True (inverted orientation), and None (no orientation specified).
  • --indent option for commandline usage
  • AMRCodec with a more restrictive grammar and special inversions

Changed

  • DecodeError now derives from PenmanError
  • PENMANCodec now takes a relation_sort parameter whose value is a function that sorts Triple objects (see Added above)
  • Updated the PEG definition in README.md, and accompanying prose.

v0.5.1

Fixed

  • Numeric conversion in PENMANCodec.handle_triple() now happens to both sources and targets, and handle_triple() is now run on the top, as well. Fixes #4
  • Graph.__str__() now initializes a default codec for serialization.
  • Incorrect script usage fixed in README.md and from penman.py --help
  • Updated documentation

v0.5.0

Fixed

  • Properly trim off ^ in triple-conjunction parsing.

Added

  • TOP_VAR and TOP_REL are added to PENMANCodec (these were module-level variables prior to v0.4.0); they are only used for triple-conjunctions. If left as None, they will not be printed or interpreted.
  • PENMANCodec.triples_to_graph to instantiate a Graph with awareness of codec configuration (e.g. TOP_VAR, TYPE_REL, etc.)

Removed

  • PENMANCodec.handle_value() - value conversion is now done in PENMANCodec.handle_triple(), which has access to relation name, etc.

Changed

  • Decoupled Graph from the codec by removing the codec parameter and its usage in the class methods. Graphs can be instantiated with awareness of the codec via the codec's triples_to_graph() method.
  • PENMANCodec returns all node type triples before other edge triples.
  • PENMANCodec reads/writes TOP nodes in triple conjunctions if TOP_VAR and TOP_REL are set (see above).
  • Default node types are no longer stored as triples. This means a graph like (a / a :ARG b) cannot say if b is a node or just a symbol. Let's see how far this gets us before re-adding such support.
  • PENMANCodec.handle_triple() now manages value conversion
  • PENMAN serialization now sorts relations like sort -V

v0.4.0

This release is a major rewrite of the serialization mechanism. Now a serialization codec manages the interface between graphs and PENMAN or triple conjunctions, and it can be instantiated with parameters or subclassed to customize behavior.

Added

  • PENMANCodec class for managing serialization to/from PENMAN and triple conjunctions
  • Graph.attributes() returns terminal relations

Removed

  • Module-level TOP and TOP_RELATION are no longer used for customizing serialization behavior (see PENMANCodec)
  • Graph.concepts()
  • Graph.constants()
  • Triple.is_inverted() (use PENMANCodec.is_relation_inverted())

Changed

  • Graph.to_triples() is now Graph.triples() and no longer has a normalize parameter, but can now be filtered by source, relation, and target, as with Graph.edges() and Graph.attributes()
  • Graph.edges() only returns triples between nonterminal nodes
  • Default relation for node types is instance, since instance-of seems to not be used, documented, or intuitive.
  • load(), loads(), dump(), and dumps() now take a cls parameter for a serialization codec, and any additional **kwargs are passed to its constructor.

v0.3.0

Added

  • TOP and TOP_RELATION as module-level variables
  • is_relation_inverted()
  • invert_relation()
  • Graph.top
  • Graph.variables()
  • Graph.concepts()
  • Graph.constants()
  • Graph.edges()

Changed

  • Graph.from_triples() can take an explicit top=... parameter
  • The indent parameter to Graph.to_penman() now takes a range of values:
    • indent=True is the default with adaptive indentation
    • indent=False and indent=None do not insert newlines and use a single space to delimit fields
    • indent=N, where N is an integer, indents N spaces after a newline
  • load(), loads(), dump(), and dumps() can now take a triples=... parameter (default: False); if True, read/write as triples

v0.2.0

Changed

  • PENMAN serialization uses slightly more sophisticated default weights for choosing when to invert edges, which prevents some ugly graphs.
  • Graph.to_triples() takes a normalize=True parameter (default False) to uninvert inverted edges.
  • Running the script with --triples now prints the logical conjunction of triples (e.g. instance-of(b, bark-01) ^ ARG0(b, d) ^ instance-of(d, dog)).

v0.1.0

First release with very basic functionality.

Added

  • Triple namedtuple for edge information (source, relation, target)
  • Graph stores graph data and provides methods for access
  • load()/loads() reads Penman files/strings
  • dump()/dumps() writes Penman files/strings