very different outputs of apoc page-rank and new page-rank?

[jexp]

Like 2 orders of magnitude different score values?

CALL algo.pageRankStream('Person','KNOWS') yield node, score
> WITH * order by score DESC LIMIT 10
> RETURN id(node), node.firstName, node.lastName, score;
+-------------------------------------------------------------------+
| id(node) | node.firstName | node.lastName | score                 |
+-------------------------------------------------------------------+
| 2018160  | "Chen"         | "Yang"        | 0.0035972294696054335 |
| 2025347  | "Wei"          | "Wang"        | 0.00338002511591596   |
| 2016633  | "Rudolf"       | "Hoffmann"    | 0.003266771041832335  |
| 2024012  | "Rafael"       | "Suissa"      | 0.0032483767089390176 |
| 2024437  | "Zhang"        | "Hu"          | 0.002917721858609681  |
| 2020592  | "Paul"         | "Bach"        | 0.0026936801886415437 |
| 2017530  | "Yang"         | "Li"          | 0.002374719927258027  |
| 2017368  | "Karl"         | "Fischer"     | 0.002339354242148342  |
| 2020038  | "Wei"          | "Zhu"         | 0.002295584953345968  |
| 2022831  | "Lei"          | "Xu"          | 0.002257252684833366  |
+-------------------------------------------------------------------+
10 rows
172 ms
neo4j-sh (?)$ 
neo4j-sh (?)$ MATCH (p:Person) WITH collect(p) as nodes
> CALL apoc.algo.pageRankWithConfig(nodes,{iterations:20,types:'KNOWS'}) YIELD node, score
> WITH * order by score DESC LIMIT 10
> RETURN id(node), node.firstName, node.lastName, score;
+------------------------------------------------------+
| id(node) | node.firstName  | node.lastName | score   |
+------------------------------------------------------+
| 2019080  | "Anucha"        | "Charoenpura" | 0.48365 |
| 2016852  | "Bingbing"      | "Liu"         | 0.47604 |
| 2022828  | "Javier"        | "Gonzalez"    | 0.47339 |
| 2022593  | "Mohamed"       | "Achiou"      | 0.46651 |
| 2017530  | "Yang"          | "Li"          | 0.45797 |
| 2020196  | "Eve-Mary Thai" | "Hoang"       | 0.45622 |
| 2017641  | "Alexander"     | "Hleb"        | 0.43794 |
| 2016164  | "Avraham"       | "Alhouthi"    | 0.43247 |
| 2018520  | "Jun"           | "Zhang"       | 0.42199 |
| 2020288  | "Yang"          | "Lei"         | 0.41781 |
+------------------------------------------------------+
10 rows
32118 ms

[mknblch]

There are several variations of PageRank. We implemented PR = ( 1 - d ) / N + d * sum ( PR_v / outDegree_v ). We should check both implementations.

[jexp]

Afaik that's the same that the other page-rank implements (the textbook one). Yes, please check.

[mknblch]

I reviewed our pageRank implementation and noticed a possible bug which might lead to wrong results. We work on the pageRank array while we doing writes on it in one iteration. We should create an empty buffer for the new ranks while reading on the old ranks from the previous iteration.

[jexp]

Cool, please send the PR when you get it fixed.

[knutwalker]

possibly fixed in #174

[knutwalker]

@tomasonjo can you please verify whether this mismatch still happens?