[CLOSED] Decide on API required to support self-join workflows about ibis

[wesm]

Comment by wesm
Monday Jan 05, 2015 at 22:26 GMT

---

Looked at how sqlalchemy handles this:

http://docs.sqlalchemy.org/en/rel_0_8/orm/relationships.html#self-referential-query-strategies

punchline:

from sqlalchemy.orm import aliased

nodealias = aliased(Node)
SQLsession.query(Node).filter(Node.data=='subchild1').\
                join(nodealias, Node.parent).\
                filter(nodealias.data=="child2").\
                all()

This suggests that having a "view" API (name can change) is a reasonable solution

[wesm]

Comment by wesm
Tuesday Jan 06, 2015 at 17:40 GMT

---

Closing this. Here's an example of performing an aggregation on a self-joined table:

left = table
right = table.view()

metric = (left['a'] - right['b']).mean().name('metric')
joined = left.inner_join(right, [right['g'] == left['g']])
aggregated = joined.aggregate([metric], by=[left['g']])

The IR for the aggregation (console repr still a little messy, can't tell the parts of the aggregation apart yet) looks like:

ref_0
  TableView[table]
  a : int8
  b : int16
  c : int32
  d : int64
  e : float
  f : double
  g : string
  h : boolean

ref_1
SelfReference[table]
  Table: ref_0

Aggregation[table]
  InnerJoin[table]
    Table: ref_0
    Table: ref_1
    Equals[array(boolean)]
      Column[string] 'g' from table ref_1
      Column[string] 'g' from table ref_0
  Mean[double]
    Subtract[array(int32)]
      Column[int8] 'a' from table ref_0
      Column[int16] 'b' from table ref_1
  Column[string] 'g' from table ref_0