CHANGELOG.md

Changelog

[0.10.1] - work in progress

Added

• readers provide source information.

Changed

Removed

[0.10.0] - 11-04-2026

Added

• [] pattern notation for specifying sequences of arguments.
• EdgeType and ArgRole enums.
• safety cap for match (_MAX_ARGROLE_ITEMS=10) against pathological edge arities.
• caching of computed Hyperedge/Atom properties.
• parse_to_jsonl method on Parser.
• unified parameter interface for parsers.
• method Parser.accepted_params.
• maximum depth protection for parsers.
• repl api for parsers.

Changed

• multiple patterns functions are now Hyperedge/Atom methods: is_wildcard, is_pattern, is_fun_pattern, is_variable, contains_variable, variable_name.
• hyperbase.py now delegating to smaller modules with well-defined concerns: builders.py, correctness.py, transforms.py, patterns.checks.py and patterns.matcher.py.
• replaced itertools.permutations with constraint-propagated backtracking in argrole matcher.
• parse_text renamed to parse; old iterator-based parse removed.
• read_source renamed to parse_source; read_source_to_jsonl renamed to parse_source_to_jsonl.
• renamed sentensize to get_sentences.
• hedge now uses an explicit stack instead of recursion (so that pathologically nested edge strings cannot exhaust Python's call stack).
• renamed parsers.correctness to parsers.badness.

Removed

• __add__ operator overloading in Hyperedge/Atom.

[0.9.0] - 05-04-2026

Added

• readers (txt, url, wikipedia).
• cli interface with repl, parsers, readers.
• hyperedge.Hyperedge.match function (calls parsers.match_pattern).
• ParseResults.
• HyperEdge.text.
• load_edges function.

Changed

• added get_parser to main functions (at hyperbase root).
• improved documentation.
• hedge now accepts ParseResults and can recursively add Hyperedge.text strings.
• renamed function hyperedge.str2atom to str_to_atom.
• Hyperedge is not a frozen dataclasse, instead of inheriting from tuple.
• hedge now throws exceptions instead of returning None values.
• Hyperedge.contains always recursive, removed deep argument.
• simplified Hyperedge.simplify.
• Hyperedge.insert_argrole now private, renamed to Hyperedge._insert_argrole.
• Renamed Hyperedge.insert_edge_with_argrole to add_argument, pos optional (default to add at the end).
• Renamed Hyperedge.edges_with_argrole to arguments_with_role.
• Renamed Hyperedge.normalized to normalise.

Removed

• function patterns.edge_matches_pattern.
• deprecated and obsolete methods from Hyperedge: is_atom, to_str, roots, insert_first_argument, connect, sequence, contains_atom_type, main_concepts, replace_main_concept, has_argroles.

[0.8.0] - 26-03-2026 - hyperbase is the successor of graphbrain

Added

• parser plugin foundation.
• more comprehensive Hyperedge.check_correctness.
• check parse correctness.
• type checking: full code coverage.

Changed

• renamed library to hyperbase.
• trimmed down library to the essentials: hyperedge, patterns and parser foundations.
• converted documentation to Material for MkDocs.

Removed

• hypergraph module, hypergraph database (memory module).
• alphabeta parser implementation.
• old scripts, examples, processors.
• ad hoc utils.
• cli.
• notebook utils.
• obsolete constants.

[0.7.0] - 05-03-2026

Added

• patterns.is_wildcard().
• Base class hypergraph.memory.keyvalue.KeyValue for key-value hypergraph databases, removing redundant code between LevelDB and SQLite.
• Tests for LevelDB (only the SQLite Hypergraph implementation was being directly tested).
• Hyperedge.is_unordered_pattern().
• Functional pattern: any.
• Parser-level inference of coreferences between proper nouns.
• Parser.parse_and_add() now adds token information to the main edge of a sentence: the attribute 'tokens' contains the list of tokens and 'tok_pos' a structure with the token position for each atom in the edge.
• Post-processing step improves parser.
• Hypergraph.get_attributes().

Changed

• Entire project is now in pure Python
• Python >=3.10 now required.
• Hypergraph.search(), .match() and .count() now working with functional patterns and argument role matching.
• Fixed bug on match_pattern(), variables were not correctly matched when argument roles were specified past the first position in an atoms functional pattern.
• Simplified default systems connectors (breaks hypergraph DB backwards compatibility).
• max_text argument in parser.parse_and_add().
• Matches from patterns with repeated variables are collected in lists.

Removed

• graphbrain.logic obsolete module.
• LevelDB backend

[0.6.1] - 31-10-2022

Changed

• Hyperedge.replace_argroles() .insert_argrole() and .add_argument() 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.

[0.6.0] - 27-10-2022

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.normalise().
• 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().

[0.5.0] - 28-07-2021

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.add_argument().
• 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.

[0.3.2] - 10-02-2020

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).

[0.3.1] - 03-02-2020

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.

[0.3.0] - 28-09-2019

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).
• Improved notebooks visualizations (show(), blocks(), vblocks()).

Removed

• graphbrain.funs module.

[0.2.2] - 13-09-2019

Added

• txt_parser agent.
• MANIFEST.in to include VERSION file in distribution.

Changed

• Fixing 'pip install graphbrain' on Linux/Windows.

[0.2.1] - 04-09-2019

Added

• claim_actors and corefs_dets agents.
• meaning.concepts module.

Changed

• Fixed example.
• hypergraph.sum_degree() and .sum_deep_degree().
• Parser improvements.
• Agent corefs_atoms is now corefs_unidecode.
• Improved docs.

Removed

• Obsolete 'work-in-progress' code.
• hg2json command.

[0.2.0] - 04-08-2019

Added

• Primary entities and deep degrees.
• Hyperedges have their own class, deriving from tuple.
• Atoms have a special class, deriving from Hyperedge.
• Restructuring of code files tree.
• Created agent system + first agents.

Changed

• Parsers now have own package.

Removed

• Old experimental code.

[0.1.0] - 14-06-2019

Added

• First release.