[query] Weighted linear regression rows#10632
Conversation
|
sweet! |
patrick-schultz
left a comment
patrick-schultz
left a comment
There was a problem hiding this comment.
This looks good to me, though I didn't think very hard about the quality of the IR being generated. I do wonder if it might be worth trying to avoid the multiplications by 1 in the unweighted case, but that could be a future optimization.
| // )((accum, v) => accum.concat(v.toS).concat(" ")) | ||
| // .concat("] vs [ ") | ||
| // )((accum, v) => accum.concat(v.toS).concat(" ")) | ||
| // .concat(s"] and operation ${Pretty(body)}")) |
|
I can remove multiplications by 1 in a subsequent PR. I think not hard and probably not much of a difference. At the moment, the only person using this method will be using it to enable weighted linreg anyway, so not super worried. |
|
Something is wrong with the |
|
Reassigning to me since Patrick is OOO this week |
|
Great, thanks. Should be ready to be reviewed now, the performance is now where I want it to be. (18x faster than aggregators for Caitlin's use case) |
tpoterba
left a comment
tpoterba
left a comment
There was a problem hiding this comment.
Looks good! One question.
|
Also made it so that |
4 participants
No description provided.