Graphbrain Versions Save

Changed

• Hyperedge.replace_argroles() .insert_argrole() and .insert_edge_with_argrole() now works with functional patterns such as var.
• Fixed bug when matching patterns containing atoms functional pattern where no atom has argroles.

Removed

• interactive_case_generator() from graphbrain.notebook.

Added

• Hyperedge.atom and .not_atom properties.
• Hyperedge.mtype() and .connector_mtype() methods.
• Hyperedge.t, .mt, .ct and .cmt type shortcut properties.
• Functional patterns: var, atoms, lemma.
• Moved all patterns functions to graphbrain.patterns.
• Hypergraph.text().
• Parser.parse_and_add().
• Parser.sentences().
• Hyperedge.normalized().
• New package graphbrain.readers, for text readers from various sources.
• New package graphbrain.processors, for miscellaneous processors.
• Processor class.

Changed

• Coreference resolution now using the new spaCy experimental model.
• Now using spaCy transformer GPU models by default, can fallback to CPU model.
• Hyperedge.is_atom() deprecated.
• Improved Hyperedge.main_concepts().
• Hyperedge.atom() renamed to .inner_atom().
• Sequences keep track of their size, no need to specify position when adding edge to sequence.
• Improved wikipedia reader.
• Improved coreference resolution.
• LevelDB included by default again.
• Old package graphbrain.meaning renamed to graphbrain.utils.
• Hyperedge.argroles() now also works at relation/concept level.

Removed

• graphbrain.patterns.normalize_edge().
• graphbrain.stats obsolete package.
• graphbrain.cognition obsolete package.
• graphbrain.op obsolete module.
• Hyperedge .predicate() and .predicate_atom().

Added

• SQLite3 hypergraph database backend.
• Hypergraph.add_with_attributes().
• import and export commands.
• Hypergraph context manager for batch writes (with hopen(hg_locator) as hg ...).

Changed

• Main hypergraph database backend is now SQLite3.
• LevelDB backend becomes optional. (disabled by default)
• Neuralcoref becomes optional. (disabled by default)

Removed

• Hypergraph.atom_count().
• Hypergraph.edge_count().
• Hypergraph.primary_atom_count().
• Hypergraph.primary_edge_count().
• info command.
• corefs_unidecode agent.

[0.4.3] - 22-04-2021

Changed

• Fixed AlphaBeta bug related to temporary atoms being removed too soon from atom2tokens.
• Hypergraph.add_sequence() converts sequence name directly to atom.
• Parser level coreference resolution (neuralcoref) disabled by default, requires dedicated build.

[0.4.2] - 12-04-2021

Changed

• Solving wheel compilation issue.

[0.4.1] - 07-04-2021

Changed

• Solving issue with inclusion of auxiliary data file in non-binary distributions.

[0.4.0] - 07-04-2021

Added

• Agents system.
• Conjunctions resolution agent.
• Number agent (singular/plural relations) and related meaning.number module.
• Wikipedia agent.
• Coreference resolution in the parser through local fork of neuralcoref.
• atom_gender(), atom_number(), atom_person(), atom_animacy() in parsers.
• Coreference inference: gender, number and animacy.
• Coreference inference:'X is Y' relationships.
• dir_parser agent (looks for text files in directory tree).
• Function meaning.concepts.has_common_or_proper_concept(edge).
• Function Hypergraph.add() now has optional 'count' parameter.
• Function Hypergraph.sequences().
• Function Hyperedge.replace_argroles().
• Function Hyperedge.insert_argrole().
• Function Hyperedge.insert_edge_with_argrole().
• Function Hyperedge.all_atoms().
• Function Hyperedge.size().
• Function Hyperedge.predicate().
• Function Hyperedge.predicate_atom().
• Function Hyperedge.simplify().
• Function Hyperedge.connector_atom().
• Utility functions to show colored edges in the terminal.

Changed

• Special characters in atoms are now percent-encoded.
• parse() now returns a dictionary that includes inferred edges.
• parse() now returns a dictionary of edges to text.
• New parser framework, improved English parser + scripts to help train and test new parsers.
• Hyperedge main type character codes become upercase (apple/C instead of apple/c).
• Conjunction hyperedge type, character code: "J" (e.g.: and/J).
• Modifier hyperedge type can now be applied to anything.
• Searching patterns can now include types, argument roles and other atom parts.
• Hypergraph.search_count() renamed to Hypergraph.count().
• corefs_names agent is simplified.
• corefs_onto agent is simplified.
• Tests now run with pytest instead of the deprecated './setup.py test'.
• Hyperedge.replace_atom() optional unique argument.

Removed

• Meta-modifier hyperedge type.
• Auxiliary, subpredicate and dependency hyperedge types.
• Obsolete Hyperedge.nest() method.

Added

• simplify_role() on Atom objects produces an atom with only its simple type as role.

Changed

• Lemmas are now based on atoms with simplified roles.
• Improved actors agent (more accurate identification of actors, English only for now).

Added

• German parser (experimental and incomplete).
• Documentation.
• Hyperedge sequences.

Changed

• Improved hyperedge visualization in notebooks.
• Agents receive language and sequence.
• txt_parser agent creates a sequence.
• Refactored parsers code.
• Catching and logging exceptions during parse.
• Improved conflict agent.

Added

• Tests.
• Documentation.

Changed

• Graphbrain is now beta (main APIs considered stable).
• LevelDB edge attributes encoded in JSON.
• Renamed hypergraph() to hgraph() and moved function to jnit.
• Fixed notebook examples.
• Moved Hypergraph.lemma_degrees() to meaning.lemmas.
• Argroles location is normalized (first position in both predicates and builders).
• In the parsers package, create_parser() was moved to init and the factory module was removed.
• is_proper_concept() renamed to has_proper_concept() (meaning.concepts module)
• find_concept() renamed to strip_concept() (meaning.concepts module)

Removed

• graphbrain.funs module.