In neo4j-rstats/neo4r: A 'Neo4J' Driver

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

system("docker run --name neo4j --env NEO4J_AUTH=neo4j/password --publish=7474:7474 --publish=7687:7687 -d neo4j")
library(neo4r)
con <- neo4j_api$new(
  url = "http://localhost:7474", 
  user = "neo4j", 
  password = "password"
)
while (try(con$ping()) != 200){
  Sys.sleep(3)
}

Disclaimer: this package is still under active development. Read the NEWS.md to be informed of the last changes.

Read complementary documentation at https://neo4j-rstats.github.io/user-guide/

neo4r

The goal of {neo4r} is to provide a modern and flexible Neo4J driver for R.

It's modern in the sense that the results are returned as tibbles whenever possible, it relies on modern tools, and it is designed to work with pipes. Our goal is to provide a driver that can be easily integrated in a data analysis workflow, especially by providing an API working smoothly with other data analysis ({dplyr} or {purrr}) and graph packages ({igraph}, {ggraph}, {visNetwork}...).

It's flexible in the sense that it is rather unopinionated regarding the way it returns the results, by trying to stay as close as possible to the way Neo4J returns data. That way, you have the control over the way you will compute the results. At the same time, the result is not too complex, so that the "heavy lifting" of data wrangling is not left to the user.

The connexion object is also an easy to control R6 method, allowing you to update and query information from the API.

Server Connection

Please note that for now, the connection is only possible through http / https.

Installation

You can install {neo4r} from GitHub with:

# install.packages("remotes")
remotes::install_github("neo4j-rstats/neo4r")

or from CRAN :

install.packages("neo4r")

Create a connexion object

Start by creating a new connexion object with neo4j_api$new

library(neo4r)
con <- neo4j_api$new(
  url = "http://localhost:7474", 
  user = "neo4j", 
  password = "plop"
)

This connexion object is designed to interact with the Neo4J API.

It comes with some methods to retrieve information from it. ping(), for example, tests if the endpoint is available.

# Test the endpoint, that will not work :
con$ping()

Being an R6 object, con is flexible in the sense that you can change url, user and password at any time:

con$reset_user("neo4j")
con$reset_password("password") 
con$ping()

Other methods:

# Get Neo4J Version
con$get_version()
# List constaints (if any)
con$get_constraints()
# Get a vector of labels (if any)
con$get_labels()
# Get a vector of relationships (if any)
con$get_relationships()
# Get index 
con$get_index()

Call the API

You can either create a separate query or insert it inside the call_neo4j function.

The call_neo4j() function takes several arguments :

• query : the cypher query
• con : the connexion object
• type : "rows" or "graph": whether to return the results as a list of results in tibble, or as a graph object (with $nodes and $relationships)
• output : the output format (R or json)
• include_stats : whether or not to include the stats about the call
• meta : whether or not to include the meta arguments of the nodes when calling with "rows"

The movie graph

Starting at version 0.1.3, the play_movie() function returns the full cypher query to create the movie graph example from the Neo4J examples.

play_movies() %>%
  call_neo4j(con)

"rows" format

The user chooses whether or not to return a list of tibbles when calling the API. You get as many objects as specified in the RETURN cypher statement.

library(magrittr)

'MATCH (tom {name: "Tom Hanks"}) RETURN tom;' %>%
  call_neo4j(con)

'MATCH (cloudAtlas {title: "Cloud Atlas"}) RETURN cloudAtlas;' %>%
  call_neo4j(con)

"MATCH (people:Person)-[relatedTo]-(:Movie {title: 'Cloud Atlas'}) RETURN people.name, Type(relatedTo), relatedTo" %>%
  call_neo4j(con, type = 'row')

By default, results are returned as an R list of tibbles. For example here, RETURN tom will return a one element list, with object named tom. We think this is the more "truthful" way to implement the outputs regarding Neo4J calls.

When you want to return two nodes types, you'll get two results, in the form of two tibbles - the result is a two elements list with each element being labelled the way it has been specified in the Cypher query.

'MATCH (tom:Person {name: "Tom Hanks"})-[:ACTED_IN]->(tomHanksMovies) RETURN tom,tomHanksMovies' %>%
  call_neo4j(con)

Results can also be returned in JSON, for example for writing to a file:

tmp <- tempfile(fileext = ".json")
'MATCH (people:Person) RETURN people.name LIMIT 1' %>%
  call_neo4j(con, output = "json") %>%
  write(tmp)
jsonlite::read_json(tmp)

If you turn the type argument to "graph", you'll get a graph result:

'MATCH (tom:Person {name: "Tom Hanks"})-[act:ACTED_IN]->(tomHanksMovies) RETURN act,tom,tomHanksMovies' %>%
  call_neo4j(con, type = "graph")

The result is returned as one node or relationship by row.

Due to the specific data format of Neo4J, there can be more than one label and property by node and relationship. That's why the results is returned, by design, as a list-dataframe.

We have designed several functions to unnest the output :

+unnest_nodes(), that can unnest a node dataframe :

res <- 'MATCH (tom:Person {name:"Tom Hanks"})-[a:ACTED_IN]->(m)<-[:ACTED_IN]-(coActors) RETURN m AS acted,coActors.name' %>%
  call_neo4j(con, type = "graph")
unnest_nodes(res$nodes)

Please, note that this function will return NA for the properties that aren't in a node.

Also, it is possible to unnest either the properties or the labels :

res %>%
  extract_nodes() %>%
  unnest_nodes(what = "properties")

res %>%
  extract_nodes() %>%
  unnest_nodes(what = "label")

• unnest_relationships()

There is only one nested column in the relationship table, thus the function is quite straightforward :

'MATCH (people:Person)-[relatedTo]-(:Movie {title: "Cloud Atlas"}) RETURN people.name, Type(relatedTo), relatedTo' %>%
  call_neo4j(con, type = "graph") %>%
  extract_relationships() %>%
  unnest_relationships()

Note that unnest_relationships() only does one level of unnesting.

• unnest_graph

This function takes a graph results, and does unnest_nodes and unnest_relationships.

'MATCH (people:Person)-[relatedTo]-(:Movie {title: "Cloud Atlas"}) RETURN people.name, Type(relatedTo), relatedTo' %>%
  call_neo4j(con, type = "graph") %>%
  unnest_graph()

Extraction

There are two convenient functions to extract nodes and relationships:

'MATCH (bacon:Person {name:"Kevin Bacon"})-[*1..4]-(hollywood) RETURN DISTINCT hollywood' %>%
  call_neo4j(con, type = "graph") %>% 
  extract_nodes()

'MATCH p=shortestPath(
  (bacon:Person {name:"Kevin Bacon"})-[*]-(meg:Person {name:"Meg Ryan"})
)
RETURN p' %>%
  call_neo4j(con, type = "graph") %>% 
  extract_relationships()

Convert for common graph packages

{igraph}

In order to be converted into a graph object:

• The nodes should be a dataframe with the first column being a series of unique ID, understood as "names" by igraph - these are the ID columns from Neo4J. Other columns are considered attributes.

• relationships need a start and an end, i.e. startNode and endNode in the Neo4J results.

Here how to create a graph object from a {neo4r} result:

G <- "MATCH a=(p:Person {name: 'Tom Hanks'})-[r:ACTED_IN]->(m:Movie) RETURN a;" %>% 
  call_neo4j(con, type = "graph") 

library(dplyr)
library(purrr)
# Create a dataframe with col 1 being the ID, 
# And columns 2 being the names
G$nodes <- G$nodes %>%
  unnest_nodes(what = "properties") %>% 
  # We're extracting the first label of each node, but 
  # this column can also be removed if not needed
  mutate(label = map_chr(label, 1))
head(G$nodes)

We then reorder the relationnship table:

G$relationships <- G$relationships %>%
  unnest_relationships() %>%
  select(startNode, endNode, type, everything()) %>%
  mutate(roles = unlist(roles))
head(G$relationships)

graph_object <- igraph::graph_from_data_frame(
  d = G$relationships, 
  directed = TRUE, 
  vertices = G$nodes
)
plot(graph_object)

This can also be used with {ggraph} :

library(ggraph)
graph_object %>%
  ggraph() + 
  geom_node_label(aes(label = label)) +
  geom_edge_link() + 
  theme_graph()

{visNetwork}

{visNetwork} expects the following format :

nodes

• "id" : id of the node, needed in edges information
• "label" : label of the node
• "group" : group of the node. Groups can be configure with visGroups
• "value" : size of the node
• "title" : tooltip of the node

edges

• "from" : node id of begin of the edge
• "to" : node id of end of the edge
• "label" : label of the edge
• "value" : size of the node
• "title" : tooltip of the node

(from ?visNetwork::visNetwork).

visNetwork is smart enough to transform a list column into several label, so we don't have to worry too much about this one.

Here's how to convert our {neo4r} result:

G <-"MATCH a=(p:Person {name: 'Tom Hanks'})-[r:ACTED_IN]->(m:Movie) RETURN a;" %>% 
  call_neo4j(con, type = "graph") 

# We'll just unnest the properties
G$nodes <- G$nodes %>%
  unnest_nodes(what = "properties")
head(G$nodes)  

# Turn the relationships :
G$relationships <- G$relationships %>%
  unnest_relationships() %>%
  select(from = startNode, to = endNode, label = type)
head(G$relationships)

visNetwork::visNetwork(G$nodes, G$relationships)

Sending data to the API

You can simply send queries has we have just seen, by writing the cypher query and call the api.

Transform elements to cypher queries

• vec_to_cypher() creates a list :

vec_to_cypher(iris[1, 1:3], "Species")

• and vec_to_cypher_with_var() creates a cypher call starting with a variable :

vec_to_cypher_with_var(iris[1, 1:3], "Species", a)

This can be combined inside a cypher call:

paste("MERGE", vec_to_cypher(iris[1, 1:3], "Species"))

Reading and sending a cypher file :

• read_cypher reads a cypher file and returns a tibble of all the calls:

read_cypher("data-raw/create.cypher")

• send_cypher reads a cypher file, and send it the the API. By default, the stats are returned.

send_cypher("data-raw/constraints.cypher", con)

Sending csv dataframe to Neo4J

The load_csv sends an csv from an url to the Neo4J browser.

The args are :

• on_load : the code to execute on load
• con : the connexion object
• url : the url of the csv to send
• header : whether or not the csv has a header
• periodic_commit : the volume for PERIODIC COMMIT
• as : the AS argument for LOAD CSV
• format : the format of the result
• include_stats : whether or not to include the stats
• meta : whether or not to return the meta information

Let's use Neo4J northwind-graph example for that.

# Create the query that will create the nodes and relationships
on_load_query <- 'CREATE (n:Product)
  SET n = row,
  n.unitPrice = toFloat(row.unitPrice),
  n.unitsInStock = toInteger(row.unitsInStock), n.unitsOnOrder = toInteger(row.unitsOnOrder),
  n.reorderLevel = toInteger(row.reorderLevel), n.discontinued = (row.discontinued <> "0");'
# Send the csv 
load_csv(url = "http://data.neo4j.com/northwind/products.csv", 
         con = con, header = TRUE, periodic_commit = 50, 
         as = "row", on_load = on_load_query)

Using the Connection Pane

{neo4r} comes with a Connection Pane interface for RStudio.

Once installed, you can go to the "Connections", and use the widget to connect to the Neo4J server:

Sandboxing in Docker

You can get an RStudio / Neo4J sandbox with Docker :

docker pull colinfay/neo4r-docker
docker run -e PASSWORD=plop -e ROOT=TRUE -d -p 8787:8787 neo4r

CoC

Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.

system("docker stop neo4j && sleep 2 && docker rm neo4j")

neo4j-rstats/neo4r documentation built on June 5, 2019, 4:36 p.m.