Sometimes doubling results

[cbizon]

Query:

{
    "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "id": "NCBIGENE:1017",
                    "category": "biolink:Gene"
                },
                "n1": {
                    "category": "biolink:Pathway"
                }
            },
            "edges": {
                "e01": {
                    "subject": "n0",
                    "object": "n1"
                }
            }
        }
    }
}

See ranking-agent/aragorn#10

This query returns 116 results from BTE (always? mostly?). It sometimes returns 232 (2x116) from strider. This seems to happen semi-reproducibly if I call strider twice in rapid succession, but I have seen it on a first call as well. Possibly a BTE issue instead?

[patrickkwang]

Very difficult to reproduce and test, in part due to BTE rate limits. We also have not observed this with any other queries.

[cbizon]

Here's the same thing happening in cohd reproducibly:

query={
      "message": {
        "query_graph": {
            "nodes": {
                "n0": {
                    "name": "drug-induced liver injury",
                    "ids": [
                        "SNOMEDCT:197358007"
                    ]
                },
                "n1": {
                    "categories": [
                        "biolink:DiseaseOrPhenotypicFeature"
                    ],
                    "name": "Disease Or Phenotypic Feature"
                },
            },
            "edges": {
                "e0": {
                    "subject": "n0",
                    "object": "n1",
                    "predicates": [
                        "biolink:correlated_with"
                    ]
                }
            }
        }
    }
}

This one hop returns from a direct cohd call with 460 results in a second or two. It runs for 10 minutes (!) in strider, and returns 880 results. It turns out that this is a doubling because although cohd is returning 440 individual n1 results, some of them are repeated 2 or 3 times (with different kedges).

[cbizon]

Notebook here: https://github.com/ranking-agent/robogallery/blob/master/minihackathons/minihackathon%20workflow%20B%20step%201%20COHD.ipynb

[mmersmann]

Decisions:

1. Symmetric predicates should be responded to, regardless of how we ask.
2. Inverses: should only send canonical predicate direction. This means we should canonicalize query graphs.

Action: @patrickkwang to create a branch to address.

[mmersmann]

Recheck, given new Strider release, to see if this can be closed.

[vgardner-renci]

@cbizon @patrickkwang Can this issue be closed?

[cbizon]

I think so, but it's hard to tell because right now strider is returning 0 results for this query. That's because COHD has gone to TRAPI 1.2 and is therefore not in the kp-registry any more.

[patrickkwang]

Appears fixed on master.