dicee.models.graph_model

Shared graph-conditioned model support, extracted from the DICE ULTRA implementation.

Architecture and graph construction adapted from https://github.com/DeepGraphLearning/ULTRA at commit 427966ad8ed60420eef034063d44f3153addff90. Reference fixtures and the grouped sampling and adversarial loss implementations also follow this upstream work. Graphs are runtime context, excluded from the transferable state dictionary.

MIT License

Copyright (c) 2023 MilaGraph

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Classes

GraphKGE	Shared graph lifecycle and entity-scoring interfaces for graph foundation models.
RelationGraphKGE	Shared relation prediction API with (head, tail) query pairs.

Module Contents

class dicee.models.graph_model.GraphKGE(args)

Bases: dicee.models.base_model.BaseKGE

Shared graph lifecycle and entity-scoring interfaces for graph foundation models.

name = 'GraphKGE'

config_prefix = 'graph'

graph_filename = 'graph.pt'

checkpoint_hint = 'all keys and shapes must match'

deterministic_inference = False

query_batch_size

num_direct_relations = 0

clear_inference_cache()

Drop derived representations when graph, weights or device change.

set_inference_backend(backend)

inference_token()

train(mode=True)

Set the module in training mode.

This has an effect only on certain modules. See the documentation of particular modules for details of their behaviors in training/evaluation mode, i.e., whether they are affected, e.g. Dropout, BatchNorm, etc.

Parameters:

mode (bool) – whether to set training mode (True) or evaluation mode (False). Default: True.

Returns:

self

Return type:

Module

init_entity_embeddings(embedding_dim=None)

Create (or re-create) the entity embedding table.

This is the single place that honours defer_large_embeddings: when entity rows are sharded across FSDP ranks the table must stay None until the trainer allocates the sharded adapter. Subclasses that need an entity table of a non-default width must call this method instead of assigning self.entity_embeddings directly, so the deferral is never silently undone.

Parameters:

embedding_dim (Optional[int]) – Width of the table. Defaults to embedding_dim.

init_relation_embeddings(embedding_dim=None)

Create (or re-create) the relation embedding table.

Relation tables are never deferred - they are small enough to be replicated on every rank.

Parameters:

embedding_dim (Optional[int]) – Width of the table. Defaults to embedding_dim.

abstractmethod get_embeddings()

Return the entity and relation embedding matrices as numpy arrays.

Returns:

• entity_embeddings (numpy.ndarray) – Shape (num_entities, embedding_dim).

• relation_embeddings (numpy.ndarray) – Shape (num_relations, embedding_dim).

set_graph(triples, num_entities=None, num_relations=None, inverse_relations=None)

Attach training facts; inverse_relations maps direct DICE IDs to inverse IDs.

Without a mapping every supplied relation is treated as a direct relation. Repeated facts and explicitly supplied inverse facts are deduplicated.

save_graph(path)

load_graph(path)

load_pretrained(path)

forward_grouped(triples, head_prediction=None)

forward_triples(x)

Score a batch of (head, relation, tail) index triples.

Parameters:

x (torch.LongTensor) – Shape (batch_size, 3) integer tensor where each row is [head_idx, relation_idx, tail_idx].

Returns:

Shape (batch_size,) triple scores.

Return type:

torch.FloatTensor

forward_k_vs_sample(x, target_entity_idx)

Score a (head, relation) batch against a sampled subset of entities.

Used by KvsSample and 1vsSample datasets. Sub-classes that support sample-based labelling must override this method.

Returns:

Shape (batch_size, k) score matrix where k is the number of sampled target entities.

Return type:

torch.FloatTensor

forward_k_vs_all(x)

Score a (head, relation) batch against every entity.

Sub-classes must override this method. The default implementation raises ValueError to make missing overrides obvious at runtime.

Returns:

Shape (batch_size, num_entities) score matrix.

Return type:

torch.FloatTensor

forward_k_vs_all_heads(x, target_entity_idx=None)

Score head candidates for DICE pairs (relation, tail).

forward(x, y_idx=None)

Route the forward pass to the appropriate scoring method.

Inspects the shape and type of x to decide which low-level scorer to call:

• Tuple (x, y_idx) → forward_k_vs_sample()

• (batch, 3) tensor → forward_triples()

• (batch, 2) tensor → forward_k_vs_all()

• BPE triple tensor → forward_byte_pair_encoded_triple()

• BPE pair tensor → forward_byte_pair_encoded_k_vs_all()

Parameters:

• x (torch.LongTensor or Tuple[torch.LongTensor, torch.LongTensor]) – Either a plain index tensor or a (triple_idx, target_idx) tuple for sample-based labelling.

• y_idx (torch.LongTensor, optional) – Target entity indices used by forward_k_vs_sample(). Ignored when x is a plain tensor.

Returns:

Score tensor whose shape depends on the selected scorer.

Return type:

torch.FloatTensor

class dicee.models.graph_model.RelationGraphKGE(args)

Bases: GraphKGE

Shared relation prediction API with (head, tail) query pairs.

forward_grouped(triples, head_prediction=None)

forward_triples(x)

Score a batch of (head, relation, tail) index triples.

Parameters:

x (torch.LongTensor) – Shape (batch_size, 3) integer tensor where each row is [head_idx, relation_idx, tail_idx].

Returns:

Shape (batch_size,) triple scores.

Return type:

torch.FloatTensor

forward_k_vs_sample(x, target_entity_idx)

Score relation candidates [K] or [B,K] for (head, tail) pairs.

forward_k_vs_all(x)

Score a (head, relation) batch against every entity.

Sub-classes must override this method. The default implementation raises ValueError to make missing overrides obvious at runtime.

Returns:

Shape (batch_size, num_entities) score matrix.

Return type:

torch.FloatTensor

forward_k_vs_all_relations

forward_k_vs_all_heads(x, target_entity_idx=None)

Score head candidates for DICE pairs (relation, tail).