embed

/semantica:embed

Generate and inspect graph embeddings. Usage: /semantica:embed <sub-command> [args]

$ARGUMENTS = sub-command + arguments.

---

compute [--labels <t1,t2>] [--rels <r1,r2>] [--dim N] [--walks N]

Generate Node2Vec embeddings for graph nodes.

from semantica.kg.node_embeddings import NodeEmbedder
from semantica.context import ContextGraph

graph = ContextGraph()
embedder = NodeEmbedder()

node_labels = labels_arg.split(",") if labels_arg else graph.get_all_node_types()
rel_types = rels_arg.split(",") if rels_arg else []

# All positional args required: graph_store, node_labels, relationship_types
embeddings = embedder.compute_embeddings(
    graph_store=graph,
    node_labels=node_labels,
    relationship_types=rel_types,
    embedding_dimension=int(dim_arg) if dim_arg else None,
    num_walks=int(walks_arg) if walks_arg else None,
)

# Store embeddings back on nodes
embedder.store_embeddings(
    graph_store=graph,
    embeddings=embeddings,
    property_name="node2vec_embedding",
)

Output:

Embeddings computed and stored.
  Nodes embedded:     N
  Embedding dim:      128
  Node types covered: [type1, type2, ...]
  
Sample (first 5 nodes):
  | Node | Type | Embedding dim | Stored |

---

similar <node_id> [--top N]

Find the most similar nodes to a given node in embedding space.

from semantica.kg.node_embeddings import NodeEmbedder
from semantica.context import ContextGraph, AgentContext

graph = ContextGraph()
embedder = NodeEmbedder()

# NodeEmbedder.find_similar_nodes uses the stored node2vec_embedding property
neighbors = embedder.find_similar_nodes(
    graph_store=graph,
    node_id=node_id,
    top_k=int(top_n) if top_n else 10,
    embedding_property="node2vec_embedding",
)

# Also use AgentContext for richer similarity with metadata
ctx = AgentContext(kg_algorithms=True)
entity_similar = ctx.find_similar_entities(
    entity_id=node_id,
    similarity_type="content",  # or "structural", "hybrid"
    top_k=int(top_n) if top_n else 10,
)