What is pattern comprehension and custom projection in neo4j cypher - neo4j

I was reading cypher refcard in which I came across following:
Pattern comprehensions may be used to do a custom projection from a match directly into a list:
MATCH (a)
RETURN [(a)-->(b) WHERE b.name = 'Bob' | b.age]
I prepared simple graph and tried similar looking queries on it. But it kept giving error Invalid input 'W': expected whitespace, comment, a relationship pattern on WHERE.
Q1. Whats the meaning of the above cypher, should it return all paths (a)-->(b) with b.name=Bob or return b.age?
Q2. I never saw path specification (a)-->(b) after RETURN. Obviously I am missing some basics here. Whats that?

NOTE: Pattern comprehension was only introduced in Neo4j 3.1, versions 3.0.x and below won't have this feature.
Answer to Q1: The meaning in this example is: "Given variable a (since it is in scope from earlier in the query) find an outgoing relation to some node and bind it to the variable b where node b's name property is 'Bob'. Populate a list with the age property of every b node.
The | in this context separates the pattern and where clause from the expression of what values to populate into the resulting list.
Not sure I'm following what you're asking about in Q2.
For your specific usage, why it's giving you an error, we need to be able to see what you're doing with it to figure out the problem. Can you add that to your description?
Though if I was to venture a guess, you might be using a pattern in the pattern comprehension that doesn't have any relationships, something like this:
return [(a:Person) | a.name] as names
Currently usages like this will fail when there is no relationship in the pattern, something I consider a bug, and filed as such to the issues list.
For more info, here's the pattern comprehension entry in the dev guide, and a longer writeup about pattern comprehension (and map projection).

Related

Back-reference neo4j

Can I use some back-reference sort of mechanism in neo4j? I'm not interested in what matches the query, just that it is the same thing in many places. Something like:
MATCH (a:Event {diagnosis1:11})
MATCH (b:Event {diagnosis1:15})
MATCH (c:Event {diagnosis1:5})
MATCH (a)-[rel:Next {PatientID:*}]->(b)
MATCH (b)-[rel1:Next {PatientID:\{1}]->(c)
The idea is that I just require the attribute IDs from both edges to be the same, without specifying it. The whole purpose of it would be not generating all possible matchings, to then filter them, but only hop in the specific places.
I've asked something similar in a more specific way here.
Edit: I know WHERE clauses can be used for that, but they filter the query AFTER matching the edges and nodes. I want to do that DURING the matching!
Use a WHERE clause with simple references, there's no need for back references:
MATCH (a)-[rel:Next]->(b)
MATCH (b)-[rel1:Next]->(c)
WHERE rel.PatientID = rel1.PatientID
Update
First of all, Cypher is a declarative query language: you express what you want, the runtime takes care of executing and optimizing it any way it can, so it's not that obvious that it would do it the way you think it will, or that using "back references" would magically solve the problem; it's just another way of writing the same thing.
So, your problem is that the match creates all the relationship pairs before filtering them. How about splitting the match in 2 phases using WITH?
MATCH (a:Event {diagnosis1:11})-[rel:Next]->(b:Event {diagnosis1:15})
WITH a, b, rel
MATCH (b)-[rel1:Next]->(c:Event {diagnosis1:5})
WHERE rel1.PatientID = rel.PatientID
That should only select the second relationships that match the first, but I'm not sure if it's an O(n^2) algorithm in Cypher's runtime.
Otherwise, if you drop to the Java API (which would mean either an extension or a procedure, depending on your version of Neo4j), you can probably implement in O(n) by
scanning all the relationships between a and b, indexing them by PatientID in some multimap (see Guava, or use a Map<K, Collection<V>>); this is O(n)
then doing the same for all the relationships between b and c, still O(n)
iterate on the keys of one multimap to get the values in both and match them, still O(n)

What does a comma in a Cypher query do?

A co-worker coded something like this:
match (a)-[r]->(b), (c) set c.x=y
What does the comma do? Is it just shorthand for MATCH?
Since Cypher's ASCII-art syntax can only let you specify one linear chain of connections in a row, the comma is there, at least in part, to allow you to specify things that might branch off. For example:
MATCH (a)-->(b)<--(c), (b)-->(d)
That represents three nodes which are all connected to b (two incoming relationships, and one outgoing relationship.
The comma can also be useful for separating lines if your match gets too long, like so:
MATCH
(a)-->(b)<--(c),
(c)-->(d)
Obviously that's not a very long line, but that's equivalent to:
MATCH
(a)-->(b)<--(c)-->(d)
But in general, any MATCH statement is specifying a pattern that you want to search for. All of the parts of that MATCH form the pattern. In your case you're actually looking for two unconnected patterns ((a)-[r]->(b) and (c)) and so Neo4j will find every combination of each instance of both patterns, which could potentially be very expensive. In Neo4j 2.3 you'd also probably get a warning about this being a query which would give you a cartesian product.
If you specify multiple matches, however, you're asking to search for different patterns. So if you did:
MATCH (a)-[r]->(b)
MATCH (c)
Conceptually I think it's a bit different, but the result is the same. I know it's definitely different with OPTIONAL MATCH, though. If you did:
MATCH (a:Foo)
OPTIONAL MATCH (a)-->(b:Bar), (a)-->(c:Baz)
You would only find instances where there is a Foo node connected to nothing, or connected to both a Bar and a Baz node. Whereas if you do this:
MATCH (a:Foo)
OPTIONAL MATCH (a)-->(b:Bar)
OPTIONAL MATCH (a)-->(c:Baz)
You'll find every single Foo node, and you'll match zero or more connected Bar and Baz nodes independently.
EDIT:
In the comments Stefan Armbruster made a good point that commas can also be used to assign subpatterns to individual identifiers. Such as in:
MATCH path1=(a)-[:REL1]->(b), path2=(b)<-[:REL2*..10]-(c)
Thanks Stefan!
EDIT2: Also see Mats' answer below
Brian does a good job of explaining how the comma can be used to construct larger subgraph patterns, but there's also a subtle yet important difference between using the comma and a second MATCH clause.
Consider the simple graph of two nodes with one relationship between them. The query
MATCH ()-->() MATCH ()-->() RETURN 1
will return one row with the number 1. Replace the second MATCH with a comma, however, and no rows will be returned at all:
MATCH ()-->(), ()-->() RETURN 1
This is because of the notion of relationship uniqueness. Inside each MATCH clause, each relationship will be traversed only once. That means that for my second query, the one relationship in the graph will be matched by the first pattern, and the second pattern will not be able to match anything, leading to the whole pattern not matching. My first query will match the one relationship once in each of the clauses, and thus create a row for the result.
Read more about this in the Neo4j manual: http://neo4j.com/docs/stable/cypherdoc-uniqueness.html

Cypher: Ordering Nodes on Same Level by Property on Relationship

I am new to Neo4j and currently playing with this tree structure:
The numbers in the yellow boxes are a property named order on the relationship CHILD_OF.
My goal was
a) to manage the sorting order of nodes at the same level through this property rather than through directed relationships (like e.g. LEFT, RIGHT or IS_NEXT_SIBLING, etc.).
b) being able to use plain integers instead of complete paths for the order property (i.e. not maintaining sth. like 0001.0001.0002).
I can't however find the right hint on how or if it is possible to recursively query the graph so that it keeps returning the nodes depth-first but for the sorting at each level consider the order property on the relationship.
I expect that if it is possible it might include matching the complete path iterating over it with the collection utilities of Cypher, but I am not even close enough to post some good starting point.
Question
What I'd expect from answers to this question is not necessarily a solution, but a hint on whether this is a bad approach that would perform badly anyways. In terms of Cypher I am interested if there is a practical solution to this.
I have a general idea on how I would tackle it as a Neo4j server plugin with the Java traversal or core api (which doesn't mean that it would perform well, but that's another topic), so this question really targets the design and Cypher aspect.
This might work:
match path = (n:Root {id:9})-[:CHILD_OF*]->(m)
WITH path, extract(r in rels(path) | r.order) as orders
ORDER BY orders
if it complains about sorting arrays then computing a number where each digit (or two digits) are your order and order by that number
match path = (n:Root {id:9})-[:CHILD_OF*]->(m)
WITH path, reduce(a=1, r in rels(path) | a*10+r.order) as orders
ORDER BY orders

Cypher query optimisation - Utilising known properties of nodes

Setup:
Neo4j and Cypher version 2.2.0.
I'm querying Neo4j as an in-memory instance in Eclipse created TestGraphDatabaseFactory().newImpermanentDatabase();.
I'm using this approach as it seems faster than the embedded version and I assume it has the same functionality.
My graph database is randomly generated programmatically with varying numbers of nodes.
Background:
I generate cypher queries automatically. These queries are used to try and identify a single 'target' node. I can limit the possible matches of the queries by using known 'node' properties. I only use a 'name' property in this case. If there is a known name for a node, I can use it to find the node id and use this in the start clause. As well as known names, I also know (for some nodes) if there are names known not to belong to a node. I specify this in the where clause.
The sorts of queries that I am running look like this...
START
nvari = node(5)
MATCH
(target:C5)-[:IN_LOCATION]->(nvara:LOCATION),
(nvara:LOCATION)-[:CONNECTED]->(nvarb:LOCATION),
(nvara:LOCATION)-[:CONNECTED]->(nvarc:LOCATION),
(nvard:LOCATION)-[:CONNECTED]->(nvarc:LOCATION),
(nvard:LOCATION)-[:CONNECTED]->(nvare:LOCATION),
(nvare:LOCATION)-[:CONNECTED]->(nvarf:LOCATION),
(nvarg:LOCATION)-[:CONNECTED]->(nvarf:LOCATION),
(nvarg:LOCATION)-[:CONNECTED]->(nvarh:LOCATION),
(nvari:C4)-[:IN_LOCATION]->(nvarg:LOCATION),
(nvarj:C2)-[:IN_LOCATION]->(nvarg:LOCATION),
(nvare:LOCATION)-[:CONNECTED]->(nvark:LOCATION),
(nvarm:C3)-[:IN_LOCATION]->(nvarg:LOCATION),
WHERE
NOT(nvarj.Name IN ['nf']) AND NOT(nvarm.Name IN ['nb','nj'])
RETURN DISTINCT target
Another way to think about this (if it helps), is that this is an isomorphism testing problem where we have some information about how nodes in a query and target graph correspond to each other based on restrictions on labels.
Question:
With regards to optimisation:
Would it help to include relation variables in the match clause? I took them out because the node variables are sufficient to distinguish between relationships but this might slow it down?
Should I restructure the match clause to have match/where couples including the where clauses from my previous example first? My expectation is that they can limit possible bindings early on. For example...
START
nvari = node(5)
MATCH
(nvarj:C2)-[:IN_LOCATION]->(nvarg:LOCATION)
WHERE NOT(nvarj.Name IN ['nf'])
MATCH
(nvarm:C3)-[:IN_LOCATION]->(nvarg:LOCATION)
WHERE NOT(nvarm.Name IN ['nb','nj'])
MATCH
(target:C5)-[:IN_LOCATION]->(nvara:LOCATION),
(nvara:LOCATION)-[:CONNECTED]->(nvarb:LOCATION),
(nvara:LOCATION)-[:CONNECTED]->(nvarc:LOCATION),
(nvard:LOCATION)-[:CONNECTED]->(nvarc:LOCATION),
(nvard:LOCATION)-[:CONNECTED]->(nvare:LOCATION),
(nvare:LOCATION)-[:CONNECTED]->(nvarf:LOCATION),
(nvarg:LOCATION)-[:CONNECTED]->(nvarf:LOCATION),
(nvarg:LOCATION)-[:CONNECTED]->(nvarh:LOCATION),
(nvare:LOCATION)-[:CONNECTED]->(nvark:LOCATION)
RETURN DISTINCT target
On the side:
(Less important but still an interest) If I make each relationship in a match clause an optional match except for relationships containing the target node, would cypher essentially be finding a maximum common sub-graph between the query and the graph data base with the constraint that the MCS contains the target node?
Thanks a lot in advance! I hope I have made my requirements clear but I appreciate that this is not a typical use-case for Neo4j.
I think querying with node properties is almost always preferable to using relationship properties (if you had a choice), as that opens up the possibility that indexing can help speed up the query.
As an aside, I would avoid using the IN operator if the collection of possible values only has a single element. For example, this snippet: NOT(nvarj.Name IN ['nf']), should be (nvarj.Name <> 'nf'). The current versions of Cypher might not use an index for the IN operator.
Restructuring a query to eliminate undesirable bindings earlier is exactly what you should be doing.
First of all, you would need to keep using MATCH for at least the first relationship in your query (which binds target), or else your result would contain a lot of null rows -- not very useful.
But, thinking clearly about this, if all the other relationships were placed in separate OPTIONAl MATCH clauses, you'd be essentially saying that you want a match even if none of the optional matches succeeded. Therefore, the logical equivalent would be:
MATCH (target:C5)-[:IN_LOCATION]->(nvara:LOCATION)
RETURN DISTINCT target
I don't think this is a useful result.

Cypher match path based on previous relationship

I am trying to impose restrictions on my path match pattern.
I would like to match the next relationship based on the type of the previous used relationship.
Here is an example of a simplified Database:
(A)-1-(B)-2-(C)-1-(E)-2-(F)
| |
3----(D)----3
Query 1:
start n=node(A), m=node(F)
match p=n-[r*]-m
return p
should result both paths
(A)-1-(B)-2-(C)-1-(E)-2-(F)
(A)-1-(B)-3-(D)-3-(E)-2-(F)
However, when running the query starting from node (F):
start m=node(F),n=node(A)
match p=m-[r*]-n
return p
The result should be only:
(F)-2-(E)-1-(C)-2-(B)-1-(A)
Path
(F)-2-(E)-3-(D)-3-(B)-1-(A)
should not be valid, since it violates these constrains:
Coming from a -1- type relationship you can proceed to either a
-2- or -3- relationship.
Coming from a -2- or -3- type relationship you can only proceed to
a -1- relationship.
These paths are valid:
()-1-()-2-()
()-1-()-3-()
()-2-()-1-()
()-3-()-1-()
These path are not valid:
()-3-()-2-()
()-2-()-3-()
First, upvote for the very detailed, specific, and well laid out question.
Unfortunately, I don't think it's possible to do what you want to do with Cypher, I think you need the Traversal API to do this. Cypher is a declarative language; that is, you tell it what you want, and it goes and gets it for you. Using the traversal API is an imperative approach to query; that is, you tell neo4j exactly how to traverse the graph.
Your query here imposes constraints about the order in which relationships get traversed, and what makes a valid path. Nothing wrong with that, but I believe that imposing constraints on the order of traversal implicitly means you're telling cypher which way to traverse, and you just can't do that with cypher because it's declarative. Another common example of the declarative vs. imperative thing is breadth-first vs. depth-first search. If you're looking for certain nodes, you can't tell cypher to traverse breadth-first vs. depth-first; you just tell it which nodes you want, and it goes and gets them.
Now, paths can be treated like collections in cypher via the relationships() function. And you can use the filter and reduce functions to work with individual relationships. But your query is harder in that you need code that says something like "If the first relationship in a path is a 1, then the next must be a 2 or a 3". This is exactly the sort of thing that you can do with Evaluators in the Traversal API. Check the interface and you can see how you could write your own method that would implement exactly the logic you're talking about via Evaluator#evaluate(Path path).
As a general note, because declarative query (cypher) hides traversal details from you, IMHO it's always better to use declarative query for ease, if you can specify what you want declaratively. But there are cases where you have to control the order of traversal, and for that you need traversal. (I have had cases where I need all nodes connected to something else, via breadth-first search only, to a maximum depth of 3, along complex relationship criteria -- I couldn't use cypher for that).
To give you a way forward, check the link I provided on the traversal framework. Perhaps you can describe your query as a TraversalDescription, and then hand it off to neo4j to run.

Resources