PyNeuraLogic

Full Changelog: https://github.com/LukasZahradnik/PyNeuraLogic/compare/v0.7.18...v0.7.19

• Add option to disable raising exception on already started JVM
• Fix progress parameter for neuralize method
• Add support for grounding samples without queries

• Fixed settings propagation

• Variables and Constants now support typing
• Fixed #57
• Exposed all settings via brackets

• Add parameter for setting the JAR path in initialize
• Fix setting window title when drawing in matplotlib
• Add constant and variable getters
• Add normalization and self_loops parameters to gcnconv
• Update NeuraLogic backend

Full Changelog: https://github.com/LukasZahradnik/PyNeuraLogic/compare/v0.7.12...v0.7.13

• Add parameter for setting the JAR path in initialize
• Fix setting window title when drawing in matplotlib
• Add constant and variable getters
• Add normalization and self_loops parameters to gcnconv
• Update NeuraLogic backend

Full Changelog: https://github.com/LukasZahradnik/PyNeuraLogic/compare/v0.7.11...v0.7.12

• Added NeuralSample.get_atom to rettrieve substitutions for variables and value of the atom

Additions:

• neuralogic.initialize now exposes more global settings
• Datasets now contains Samples, which are tuples of examples and queries
• New GINEConv module
• New GENConv module
• GCNConv now supports normalization

Changes

• Template is now subscriptable (allows set, delete, get by index)
• Default rule aggregation exposed via Settings
• duplicate_groundings are now present in Metadata.__str__
• Global setter for the graphviz path
• Split dataset building into two steps - groundings can now be rendered
• Alternative way to define rule bodies using & (e.g, R.a <= R.b & R.c)
• Calling .draw() now display the image (if show=True and no path has been provided)

Changes:

• The Rule can be imported from neuralogic.core
• The NeuraLogic backend has been updated
• Added __repr__ methods for relations, metadata, rules, and templates
• Simplified accessing samples' target values (sample.target)
• Added a model parameter to the template.draw method to draw template with given weights

Additions:

• It is now possible to decode one-hot encoded features of edge attributes into a class-specific predicate (For Tensor Dataset)

Bug Fixes

• Fix CSV Dataset Terms Getter - CSV Dataset kept terms capitalized, resulting in changing their meaning from constants to variables.

Bug fixes

• Negation is now supported also for body literals referencing other rule's head
• Fixed serialization issues on some JVMs
• Fixed the direction of arrows in visualizations

Bug Fixes

• Fixed cycle breaking #44
• Wrong string representation of a negated relation

Changes

• Reintroduction of relation negation:
  • ~R.rel is a negation
  • -R.rel applies reverse transformation function

Bug fixes

• Fixed incorrect compression (removal) of transposition neurons.

Bug fixes

• Fixed missing skip connection in Transformer encoder/decoder

Additions:

• Transformer modules - Transformer, TransformerEncoder, and TransformerDecoder.
• Attention modules - Attention, MultiheadAttention.
• PositionalEncoding module.
• Softmax as a (splittable) aggregation function.
• Concatenation as an aggregation function.
• Reshape transformation function.
• Slice transformation function.
• Learning rate decay as an optimizer argument.
• Batching support.
• Progress bar for tracking dataset-building progress.
• Special arithmetic predicates for addition (add), subtraction (sub), and modulo (mod).

Changes:

• Concatenation can now be parametrized with a dimension.
• Refactored MLP module - the current version is more compact.
• Changed the behavior of Variable/V factory - only the first letter is capitalized.
• Performance improvements.