FALCONModule

class mowl.nn.FALCONModule(nclasses, nentities, nrelations, heads_dict, tails_dict, embed_dim=128, anon_e=4, t_norm='product', max_measure='max', residuum='notCorD', loss_type='c', num_negs=4, device='cpu')

Bases: ALCModule

Implementation of the FALCON model [falcon2022], a fuzzy \(\mathcal{ALC}\) neural reasoner. Each class expression is mapped to a fuzzy set over a collection of (named and anonymous) entity embeddings, and axioms are scored through fuzzy logical operators.

Based on the original implementation at https://github.com/bio-ontology-research-group/FALCON

Methods Summary

forward(axiom, x, e_emb[, stage])	Computes the loss of an axiom given an entity-membership context.
forward_fs(cexpr, x, e_emb[, cur_index])	Recursively computes the fuzzy-set membership degree of a class expression over the entities e_emb.
get_cc_loss(fs)
sample_negatives(e, r, used_dict)

Methods Documentation

forward(axiom, x, e_emb, stage='train')

Computes the loss of an axiom given an entity-membership context.

Parameters:

• axiom – The (grouped) axiom or axiom pattern to score. Typically an instance of org.semanticweb.owlapi.model.OWLAxiom.

• x (torch.Tensor) – Tensor encoding the concrete classes, properties and individuals that fill the axiom pattern.

• e_emb (torch.Tensor) – Embeddings of the entities (including sampled anonymous entities) over which fuzzy memberships are evaluated.

forward_fs(cexpr, x, e_emb, cur_index=0)

Recursively computes the fuzzy-set membership degree of a class expression over the entities e_emb.

Parameters:

• cexpr – Class expression of type org.semanticweb.owlapi.model.OWLClassExpression.

• x (torch.Tensor) – Tensor encoding the named classes/properties of cexpr.

• e_emb (torch.Tensor) – Entity embeddings.

• cur_index (int, optional) – Current column offset into x while walking the expression tree. Defaults to 0.

get_cc_loss(fs)

sample_negatives(e, r, used_dict)