gardel/lightrag
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #1003 from danielaskdd/add-graph-search-mode

Browse Source 
Feat: Added minimum degree filter for graph queries
This commit is contained in:
zrguo
2025-03-07 12:37:44 +08:00
committed by GitHub
parent ea898aeda0 7e088ca528
commit 3286a0d57f
17 changed files with 256 additions and 135 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
lightrag/api/routers/graph_routes.py
View File

@@ -3,7 +3,6 @@ This module contains all graph-related routes for the LightRAG API.
"""
from typing import Optional
from fastapi import APIRouter, Depends
from ..utils_api import get_api_key_dependency, get_auth_dependency
@@ -25,23 +24,33 @@ def create_graph_routes(rag, api_key: Optional[str] = None):
return await rag.get_graph_labels()
@router.get("/graphs", dependencies=[Depends(optional_api_key)])
async def get_knowledge_graph(label: str, max_depth: int = 3):
async def get_knowledge_graph(
label: str, max_depth: int = 3, min_degree: int = 0, inclusive: bool = False
):
"""
Retrieve a connected subgraph of nodes where the label includes the specified label.
Maximum number of nodes is constrained by the environment variable `MAX_GRAPH_NODES` (default: 1000).
When reducing the number of nodes, the prioritization criteria are as follows:
1. Label matching nodes take precedence
2. Followed by nodes directly connected to the matching nodes
3. Finally, the degree of the nodes
1. min_degree does not affect nodes directly connected to the matching nodes
2. Label matching nodes take precedence
3. Followed by nodes directly connected to the matching nodes
4. Finally, the degree of the nodes
Maximum number of nodes is limited to env MAX_GRAPH_NODES(default: 1000)
Args:
label (str): Label to get knowledge graph for
max_depth (int, optional): Maximum depth of graph. Defaults to 3.
inclusive_search (bool, optional): If True, search for nodes that include the label. Defaults to False.
min_degree (int, optional): Minimum degree of nodes. Defaults to 0.
Returns:
Dict[str, List[str]]: Knowledge graph for label
"""
return await rag.get_knowledge_graph(node_label=label, max_depth=max_depth)
return await rag.get_knowledge_graph(
node_label=label,
max_depth=max_depth,
inclusive=inclusive,
min_degree=min_degree,
)
return router

1
lightrag/api/webui/assets/index-CH-3l4_Z.css Normal file
View File

File diff suppressed because one or more lines are too long

160
lightrag/api/webui/assets/index-DbuMPJAD.js → lightrag/api/webui/assets/index-CJz72b6Q.js
View File

File diff suppressed because one or more lines are too long

1
lightrag/api/webui/assets/index-rP-YlyR1.css
View File

File diff suppressed because one or more lines are too long

4
lightrag/api/webui/index.html
View File

@@ -5,8 +5,8 @@
<link rel="icon" type="image/svg+xml" href="./logo.png" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Lightrag</title>
<script type="module" crossorigin src="./assets/index-DbuMPJAD.js"></script>
<link rel="stylesheet" crossorigin href="./assets/index-rP-YlyR1.css">
<script type="module" crossorigin src="./assets/index-CJz72b6Q.js"></script>
<link rel="stylesheet" crossorigin href="./assets/index-CH-3l4_Z.css">
</head>
<body>
<div id="root"></div>

2
lightrag/base.py
View File

@@ -204,7 +204,7 @@ class BaseGraphStorage(StorageNameSpace, ABC):
@abstractmethod
async def get_knowledge_graph(
self, node_label: str, max_depth: int = 5
self, node_label: str, max_depth: int = 3
) -> KnowledgeGraph:
"""Retrieve a subgraph of the knowledge graph starting from a given node."""

69
lightrag/kg/networkx_impl.py
View File

@@ -232,19 +232,26 @@ class NetworkXStorage(BaseGraphStorage):
return sorted(list(labels))
async def get_knowledge_graph(
self, node_label: str, max_depth: int = 5
self,
node_label: str,
max_depth: int = 3,
min_degree: int = 0,
inclusive: bool = False,
) -> KnowledgeGraph:
"""
Retrieve a connected subgraph of nodes where the label includes the specified `node_label`.
Maximum number of nodes is constrained by the environment variable `MAX_GRAPH_NODES` (default: 1000).
When reducing the number of nodes, the prioritization criteria are as follows:
1. Label matching nodes take precedence
2. Followed by nodes directly connected to the matching nodes
3. Finally, the degree of the nodes
1. min_degree does not affect nodes directly connected to the matching nodes
2. Label matching nodes take precedence
3. Followed by nodes directly connected to the matching nodes
4. Finally, the degree of the nodes
Args:
node_label: Label of the starting node
max_depth: Maximum depth of the subgraph
min_degree: Minimum degree of nodes to include. Defaults to 0
inclusive: Do an inclusive search if true
Returns:
KnowledgeGraph object containing nodes and edges
@@ -255,6 +262,10 @@ class NetworkXStorage(BaseGraphStorage):
graph = await self._get_graph()
# Initialize sets for start nodes and direct connected nodes
start_nodes = set()
direct_connected_nodes = set()
# Handle special case for "*" label
if node_label == "*":
# For "*", return the entire graph including all nodes and edges
@@ -262,11 +273,16 @@ class NetworkXStorage(BaseGraphStorage):
graph.copy()
) # Create a copy to avoid modifying the original graph
else:
# Find nodes with matching node id (partial match)
# Find nodes with matching node id based on search_mode
nodes_to_explore = []
for n, attr in graph.nodes(data=True):
if node_label in str(n): # Use partial matching
nodes_to_explore.append(n)
node_str = str(n)
if not inclusive:
if node_label == node_str: # Use exact matching
nodes_to_explore.append(n)
else: # inclusive mode
if node_label in node_str: # Use partial matching
nodes_to_explore.append(n)
if not nodes_to_explore:
logger.warning(f"No nodes found with label {node_label}")
@@ -277,26 +293,37 @@ class NetworkXStorage(BaseGraphStorage):
for start_node in nodes_to_explore:
node_subgraph = nx.ego_graph(graph, start_node, radius=max_depth)
combined_subgraph = nx.compose(combined_subgraph, node_subgraph)
# Get start nodes and direct connected nodes
if nodes_to_explore:
start_nodes = set(nodes_to_explore)
# Get nodes directly connected to all start nodes
for start_node in start_nodes:
direct_connected_nodes.update(
combined_subgraph.neighbors(start_node)
)
# Remove start nodes from directly connected nodes (avoid duplicates)
direct_connected_nodes -= start_nodes
subgraph = combined_subgraph
# Filter nodes based on min_degree, but keep start nodes and direct connected nodes
if min_degree > 0:
nodes_to_keep = [
node
for node, degree in subgraph.degree()
if node in start_nodes
or node in direct_connected_nodes
or degree >= min_degree
]
subgraph = subgraph.subgraph(nodes_to_keep)
# Check if number of nodes exceeds max_graph_nodes
if len(subgraph.nodes()) > MAX_GRAPH_NODES:
origin_nodes = len(subgraph.nodes())
node_degrees = dict(subgraph.degree())
start_nodes = set()
direct_connected_nodes = set()
if node_label != "*" and nodes_to_explore:
start_nodes = set(nodes_to_explore)
# Get nodes directly connected to all start nodes
for start_node in start_nodes:
direct_connected_nodes.update(subgraph.neighbors(start_node))
# Remove start nodes from directly connected nodes (avoid duplicates)
direct_connected_nodes -= start_nodes
def priority_key(node_item):
node, degree = node_item
# Priority order: start(2) > directly connected(1) > other nodes(0)
@@ -356,7 +383,7 @@ class NetworkXStorage(BaseGraphStorage):
result.edges.append(
KnowledgeGraphEdge(
id=edge_id,
type="RELATED",
type="DIRECTED",
source=str(source),
target=str(target),
properties=edge_data,

36
lightrag/lightrag.py
View File

@@ -504,11 +504,39 @@ class LightRAG:
return text
async def get_knowledge_graph(
self, node_label: str, max_depth: int
self,
node_label: str,
max_depth: int = 3,
min_degree: int = 0,
inclusive: bool = False,
) -> KnowledgeGraph:
return await self.chunk_entity_relation_graph.get_knowledge_graph(
node_label=node_label, max_depth=max_depth
)
"""Get knowledge graph for a given label
Args:
node_label (str): Label to get knowledge graph for
max_depth (int): Maximum depth of graph
min_degree (int, optional): Minimum degree of nodes to include. Defaults to 0.
inclusive (bool, optional): Whether to use inclusive search mode. Defaults to False.
Returns:
KnowledgeGraph: Knowledge graph containing nodes and edges
"""
# get params supported by get_knowledge_graph of specified storage
import inspect
storage_params = inspect.signature(
self.chunk_entity_relation_graph.get_knowledge_graph
).parameters
kwargs = {"node_label": node_label, "max_depth": max_depth}
if "min_degree" in storage_params and min_degree > 0:
kwargs["min_degree"] = min_degree
if "inclusive" in storage_params:
kwargs["inclusive"] = inclusive
return await self.chunk_entity_relation_graph.get_knowledge_graph(**kwargs)
def _get_storage_class(self, storage_name: str) -> Callable[..., Any]:
import_path = STORAGES[storage_name]