Graph Pattern Matching Examples

This guide explains how to issue a pattern-matching query against a graph, and work with the results of that query.

The Datasets

This guide uses a dataset that models relationships between politicians, athletes, celebrities, and companies.

Read the Graphs

First, create a session by launching PGX in local mode:

session = pypgx.get_session(session_name="my-session")

Next, load the graphs into memory. For example, the first dataset can be loaded as follows:

connections_graph = session.read_graph_with_properties(
    "examples/graphs/connections.edge_list.json"
)

Submit Queries

You can submit a graph pattern matching query in PGQL, an SQL-like declarative language that allows you to express a pattern that consists of vertices and edges and constraints on the properties of the vertices and edges.

To submit a query to PGX, you can use the query_pgql() method of PgxGraph (which is the type of object you get when you load a graph via the session).

session.query_pgql(query)

Enemy of My Enemy is My Friend

Here, you will find a graph pattern inspired by the famous ancient proverb The enemy of my enemy is my friend. Specifically, you will find two entities which are connected by two edges of the feuds edge label. Vertices represent people or clans or countries. A pair of vertices which are feuding with each other will have an edge with the feuds edge label.

Such a query is written in PGQL as follows:

SELECT x.name, z.name
  FROM MATCH (x) -[e1:feuds]-> (y)
    , MATCH (y) -[e2:feuds]-> (z)
WHERE x <> z
ORDER BY x.name, z.name

Note that in the query, we order the results by x.name and then z.name.

Submit the query to PGX:

result_set = connections_graph.query_pgql(
    """
    SELECT x.name, z.name
    FROM MATCH (x) -[e1:feuds]-> (y), MATCH (y) -[e2:feuds]-> (z)
    WHERE x <> z ORDER BY x.name, z.name
    """
)

PgqlResultSet manages a result set of a query. A result set contains multiple results (such a query may match many sub-graphs). Each result consists of a list of result elements. The order of result elements follows the order of variables in the SELECT clause of a query.

Iterating over a query results means iterating over a pgql_result_elements dictionary.

You can get the pgql_result_elements dictionary as follows:

result_elements = result_set.pgql_result_elements

Top 10 Most Collaborative People

Another interesting query is finding the top 10 most collaborative people in the graph in a decreasing order of the number of collaborators. Such a query exploits various features of PGQL which include grouping, aggregating, ordering, and limiting the graph patterns found in the MATCH clause. The following query string expresses a user’s inquiry in PGQL.

result_set = connections_graph.query_pgql(
    """
    SELECT x.name, COUNT(*) AS num_collaborators
    FROM MATCH (x) -[:collaborates]-> ()
    GROUP BY x
    ORDER BY num_collaborators DESC, x.name
    LIMIT 10
    """
)

The above query does the following:

1. Find all collaboration relationship patterns from the graph by matching the collaborates edge label. 1. Group the found patterns by its source vertex. 1. Apply the count aggregation to each group to find the number of collaborators. 1. Order the groups by the number of collaborators in a decreasing order. 1. Take only the first 10 results.

print() method shows the name and the number of collaborators of the top 10 collaborative people in the graph.

result_set.print()

You can see the following in the console.

x.name	num_collaborators
Barack Obama	10
Charlie Rose	4
Dieudonne Nzapalainga	3
NBC	3
Nicolas Guerekoyame Gbangou	3
Omar Kobine Layama	3
Pope Francis	3
Angela Merkel	2
Beyonce	2
Eric Holder	2

Transitive Connectivity

Another interesting query is one that tests for reachability between vertices. What we are interested in is whether every person in the graph is transitively connected to every other person.

First, we find out how many persons there are in the graph by submitting the following PGQL query:

SELECT COUNT(*) AS numPersons
FROM MATCH (n:person)

The result is as follows:

numPersons
62

For each persone, we count the number of persons that can be reached by following zero or more edges. This query can be expressed in PGQL as follows:

PATH connects_to AS () <- ()
SELECT n.name, COUNT(*) AS reachabilityCount, COUNT(*) = 62 AS reachesAllPersons
FROM MATCH (n:person) -/:connects_to*/-> (m:person)
GROUP BY n
ORDER BY COUNT(*) DESC, n.name
LIMIT 20

In the above query, we express connectivity between two neighboring persons using a path pattern connects_to. We use a Kleene star (*) to express that the path pattern may repeatedly match zero or more times as we want to determine _transitive_ connectivity.

The query uses GROUP BY to make a group for each of the source devices n and then counts the number of reachable destination devices m. The first 20 results are as follows:

n.name	reachabilityCount	reachesAllPersons
Michelle Bachelet	45	false
Nicolas Maduro	45	false
Dieudonne Nzapalainga	43	false
Nicolas Guerekoyame Gbangou	43	false
Omar Kobine Layama	43	false
Pope Francis	43	false
Abdel Fattah eL-Sisi	39	false
Abdullah Gul	39	false
Jenji Kohan	39	false
Robin Wright	39	false
Janet Yellen	38	false
Jason Collins	38	false
Mary Barra	38	false
Serena Williams	38	false
Alfonso Cuaron	37	false
Angela Merkel	37	false
Barack Obama	37	false
Benedict Cumberbatch	37	false
Beyonce	37	false
Carl Icahn	37	false

Since we sorted by increasing reachabilityCount and since even the first person in the result transitively does not connect to every person in the graph (reachesAllPersons = false), we now know that all the persons in the graph are not fully reachable from each other.