v0.11.0

Release v0.11.0

Cached Graph

• We have updated the on-disk format of the graph to now be stable across versions.
• This new storage format is also updatable, allowing deltas to be inserted into an already existing file instead of having to resave the whole graph.
• You can now 'cache' a graph, attaching a file to it, and periodically save all changes which have been inserted. Great for checkpointing! See the example below:

from raphtory import Graph
g= Graph() #Create new graph

g.add_edge(1,1,2)
g.add_edge(3,4,5)
g.cache("example_graph.raph") #add some updates and write these to a file - keeping the file attached to the graph

g.add_node(2,3)
g.add_edge(10,2,3)
g.write_updates() #Add some new updates and write these out

GraphQL

• The GraphQL server has had a massive overhaul, allowing it to now create, update, and manage a working directory full of graphs.
• Graphs will now persist across server runs based on the new cache-graph file format described above.
• These new APIs also allow you to export subgraphs from one graph into another.

A basic example of this can be seen below:

from raphtory import graphql

server = graphql.GraphServer("graphs/")
server.run()
client = server.get_client() #run a local server and get the client 
#create a new graph in the 'example_graphs' namespace
client.query(""" 
        mutation {
            newGraph(path:"example_graphs/graph1",graphType:EVENT)
        }
""")
#add an edge into the new graph
client.query("""
query add_edge{ 
        updateGraph(path: "example_graphs/graph1") {
            addEdge(time:1,src:"node1",dst:"node2",properties: [
                {key: "new", value: "new props are  awesome"}
                {key: "number", value: 0.7}
                {key: "map", value: { a: "hi"}}
            ]){success}
        }
}""")

Pandas and Parquet loader changes:

• We have placed time as the second parameter within the graph.load_x_from_y to match g.add_node and g.add_edge.
• All parameters named props in the graph.load_x_from_y functions have been updated to properties.
• The Layer_in_df flag has been replaced by layer and layer_col which will now throw an error if both are set.
• The node_type_in_df flag has been replaced by node_type and node_type_col which will now throw an error if both are set.
• You can now set the node_type via load_node_props_from_x functions.
• We have removed the static load_from_pandas and load_from_parquet functions from the graph as these had WAY too many arguments and were all around very confusing. Instead you can now create a graph first and use load_edges_from_X etc.
• The Datareader behind the Pandas and Parquet loaders now chunks the data and loads these in in part, drastically reducing the memory overhead for large datasets on ingestion.

Minor changes

• Optimised the way node IDs are stored/referenced, greatly increasing both lookup and insertion speed. This does mean, however, that a graph can only have String XOR Integer ids now, not both at the same time.
• We have updated to rust 1.80.0

What's Changed

• Fix release by @fabianmurariu in #1693
• make the benchmark names consistent again by @ljeub-pometry in #1691
• Fix benchmark names by @ljeub-pometry in #1694
• Pre multi graph by @fabianmurariu in #1695
• Feature/more merge tests by @ljeub-pometry in #1702
• Features/cached graphs by @shivam-880 in #1656
• Feature/incremental protobuf by @ljeub-pometry in #1706
• update rust version for readthedocs by @ljeub-pometry in #1708
• Make sure the string interning happens on decoding protobuf files by @ljeub-pometry in #1711
• much cleaner and easier way to test all the features compile by @ljeub-pometry in #1710
• Post cache graphql fixes by @miratepuffin in #1709
• Fix/dfview by @shivam-880 in #1712
• Feature/cached graph by @ljeub-pometry in #1718
• Tidy/ingestor by @shivam-880 in #1714
• GraphQL mutation apis by @ljeub-pometry in #1719
• Update README.md by @miratepuffin in #1726
• Release v0.11.0 by @github-actions in #1727

Full Changelog: v0.10.0...v0.11.0