Save, Load and Add New Data into GraphRAG

[brianadit24]

Related to this notebook:
https://github.com/run-llama/llama_index/blob/main/docs/docs/examples/cookbooks/GraphRAG_v1.ipynb

Someone in here already try to save and load GraphRAG and also how about we add new data into GraphRAG?

[dosubot[bot]]

Hey @brianadit24! 👋 I'm here to assist you with any issues or questions you have regarding GraphRAG. Let's tackle this challenge together!

To save, load, and add new data into GraphRAG while working with the GraphRAG notebook from the llama_index project, you can use the SimplePropertyGraphStore class. Here are the steps:

1. Save the current state:
   Use the persist method to save the current state of the SimplePropertyGraphStore to a specified path.

   graph_store = SimplePropertyGraphStore()
   # Add your data to the graph_store here
   graph_store.persist(persist_path="path/to/save/graph_store.json")

2. Load a previously saved state:
   Use the from_persist_path class method to load the SimplePropertyGraphStore from a previously saved file.

   loaded_graph_store = SimplePropertyGraphStore.from_persist_path(persist_path="path/to/save/graph_store.json")

3. Add new data:
   Use the upsert_nodes and upsert_relations methods to add new nodes and relations to the SimplePropertyGraphStore.

   from llama_index.core.graph_stores.types import LabelledNode, Relation
   
   new_node = LabelledNode(id="new_node_id", properties={"name": "New Node"})
   new_relation = Relation(id="new_relation_id", start="start_node_id", end="end_node_id", properties={"type": "new_relation_type"})
   
   loaded_graph_store.upsert_nodes([new_node])
   loaded_graph_store.upsert_relations([new_relation])

Here is a complete example:

from llama_index.core.graph_stores.simple_labelled import SimplePropertyGraphStore
from llama_index.core.graph_stores.types import LabelledNode, Relation

# Initialize a new graph store
graph_store = SimplePropertyGraphStore()

# Add some data
node1 = LabelledNode(id="subject1", properties={"name": "Subject 1"})
node2 = LabelledNode(id="object1", properties={"name": "Object 1"})
relation1 = Relation(id="relation1", start="subject1", end="object1", properties={"type": "Relation 1"})

graph_store.upsert_nodes([node1, node2])
graph_store.upsert_relations([relation1])

# Save the current state
graph_store.persist(persist_path="path/to/save/graph_store.json")

# Load the previously saved state
loaded_graph_store = SimplePropertyGraphStore.from_persist_path(persist_path="path/to/save/graph_store.json")

# Add new data to the loaded graph store
new_node = LabelledNode(id="subject2", properties={"name": "Subject 2"})
new_relation = Relation(id="relation2", start="subject2", end="object1", properties={"type": "Relation 2"})

loaded_graph_store.upsert_nodes([new_node])
loaded_graph_store.upsert_relations([new_relation])

This approach allows you to manage the state of GraphRAG effectively ^[1][2].

To continue talking to Dosu, mention @dosu.