Why do the values ​​returned by yield affect each other

[T0L0ve]

General Question
写了个查询语句

go from "1" over withgenres YIELD dst(edge) as g,src(edge) as m 
|go from $-.g over withgenres reversely yield src(edge) as recomm , $-.m as m,$-.g as g 
|group by $-.m,$-.recomm YIELD $-.m as m,$-.recomm as recomm,tofloat(size(COLLECT($-.g))) as size 
|go from $-.m over withgenres YIELD dst(edge) as g,src(edge) as m,$-.recomm as recomm,$-.size as size
|group by $-.m,$-.recomm,$-.size YIELD $-.recomm as recomm,COLLECT($-.g) as set_a,$-.size as size
|go from $-.recomm over * YIELD dst(edge) as g,$-.recomm as recomm ,$-.size as size,$-.set_a as set_a
|group by $-.recomm ,$-.size,$-.set_a YIELD $-.recomm as recomm,$-.set_a as set_a,$-.size as size ,$-.size/size(toSet($-.set_a + COLLECT($-.g))) AS jaccard
|yield  $-.recomm as recomm,$-.jaccard as jaccard
|order by $-.jaccard DESC

结果是这样的

但我将yield 中的$-.set_a as set_a删除后就无法计算出jaccard值

这是删除后的结果

同理删除边上的size也影响了结果输出。

另外我想查询与某一部电影包含相同电影类型的电影，并且将结果按jaccard相似度排序，但我现在的查询语句需要花费0.45秒才能完成查询。如果是仅查询某一部电影包含相同电影类型的电影，也就是查询语句的前两句仅消耗0.08秒，所以我想知道该怎么优化我写的计算jaccard相似度的部分，或者是不是有什么函数能让结果直接按照相似度排序。

[xtcyclist]

This issue could not be reproduced in the latest version. Please try upgrading to the latest version and check again.

(root@nebula) [nba]> go from "Tim Duncan" over * yield properties($^).age as a, properties($$).age as b | group by $-.a, $-.b yield 1 as x, abs($-.a + $-.b) as c | yield $-.x, $-.c as c | order by $-.c
+------+----------+
| $-.x | c        |
+------+----------+
| 1    | 73       |
| 1    | 75       |
| 1    | 78       |
| 1    | 83       |
| 1    | __NULL__ |
+------+----------+
Got 5 rows (time spent 5415/5782 us)

You could write your own UDF to implement jaccard or check this: https://docs.nebula-graph.com.cn/3.4.0/graph-computing/algorithm-description/#jaccard.