Configuration¶

All configuration is passed as arguments; Odin reads no global settings or environment variables of its own. This page collects every parameter in one place.

Engine construction¶

OdinEngine(...). See the OdinEngine API.

Parameter	Type	Default	Notes
`db`	`StandardDatabase`	required	Connected `python-arango` handle
`community_id`	`str`	`"global"`	Exploration scope
`cache_size`	`int`	`5000`	Graph-accessor LRU size (see Caching)
`auto_train`	`bool`	`True`	Train NPLL on first run if no model exists
`community_mode`	`str`	`"none"`	`"none"` = global, `"mapping"` = scoped

Retrieval¶

retrieve(...). See Tuning Retrieval.

Parameter	Type	Default	Effect
`seeds`	`list[str]`	required	Where exploration starts
`max_paths`	`int`	`50`	Output size
`hop_limit`	`int`	`3`	Maximum path length
`beam_width`	`int`	`64`	Paths kept per hop

Anchors¶

find_anchors(...)

Parameter	Type	Default
`seeds`	`list[str]`	required
`topn`	`int`	`20`

Schema inspection¶

SchemaInspector(...)

Parameter	Type	Default
`db`	`StandardDatabase`	required
`max_sample_docs`	`int`	`5`

inspect_arango_schema(db, output_file="schema.json").

Database connection (yours)¶

Odin does not manage the connection; you construct it with python-arango. In production, source these from the environment:

Variable (suggested)	Used for
`ARANGO_HOSTS`	ArangoDB URL(s)
`ARANGO_DB`	Database name
`ARANGO_USER`	Username
`ARANGO_PASSWORD`	Password (from a secrets manager)

import os
from arango import ArangoClient

db = ArangoClient(hosts=os.environ["ARANGO_HOSTS"]).db(
    os.environ["ARANGO_DB"],
    username=os.environ["ARANGO_USER"],
    password=os.environ["ARANGO_PASSWORD"],
)

See Connecting ArangoDB.

Requirements¶

Requirement	Version
Python	≥ 3.9
ArangoDB	≥ 3.10
PyTorch	≥ 2.0

Runtime dependencies (torch, python-arango, numpy, scipy, networkx, scikit-learn, gremlinpython) install automatically with pip install odin-engine.