Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
if they chain like that then i think i know how to do the optimization.
but do they? look for example at dataset.groupByKey(...).mapValues(...)
Dataset[T].groupByKey[K] uses
Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
@cloud-fan that makes sense to me, but its definitely not a quick win to
create that optimization.
let me think about it some more
---
If your project is set up for it, you can reply to
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/13526
That's a good point, let's focus on `ds.groupBy(...).mapValues(...)` then.
One thought, in `mapValues`, we will project away the previous value
attributes, so the workflow should be:
```
Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
@cloud-fan i can try to optimize
```grouped.mapValues(...).mapValues(...)``` but its a bit of an anti-pattern
(there should be no need to do mapValues twice) so i dont think there is much
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/13526
To optimize `ds.groupBy(...).mapValues(...)`, yea it's not trivial as you
explained above. But for `grouped.mapValues(...).mapValues(...)`, I think it
should not be that hard, as it's a pattern
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/13526
2 chained `AppendColumns` will have 2 functions: T => U and U => W, so we
can combine them this way:
convert UnsafeRow to T
apply func to T to generate U
apply func to U to generate W
Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
@rxin i can give it a try (the optimizer rule)
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have
Github user rxin commented on the issue:
https://github.com/apache/spark/pull/13526
Alright merging in master. Thanks.
@koertkuipers would you be able to add the optimizer rule?
---
If your project is set up for it, you can reply to this email and have your
reply appear on
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Merged build finished. Test PASSed.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/67268/
Test PASSed.
---
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #67268 has
finished](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/67268/consoleFull)**
for PR 13526 at commit
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #67268 has
started](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/67268/consoleFull)**
for PR 13526 at commit
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Merged build finished. Test FAILed.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/67264/
Test FAILed.
---
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #67264 has
finished](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/67264/consoleFull)**
for PR 13526 at commit
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #67264 has
started](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/67264/consoleFull)**
for PR 13526 at commit
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/13526
retest this please
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/13526
it lacks an optimizer rule to collapse `AppendColumns`, but seems ok to
merge it first and add the rule in follow-up.
---
If your project is set up for it, you can reply to this email and have
Github user rxin commented on the issue:
https://github.com/apache/spark/pull/13526
cc @cloud-fan
This looks good to me -- but I don't remember why we didn't merge it
earlier already.
---
If your project is set up for it, you can reply to this email and have your
reply
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Merged build finished. Test PASSed.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/62388/
Test PASSed.
---
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #62388 has
finished](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/62388/consoleFull)**
for PR 13526 at commit
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #62388 has
started](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/62388/consoleFull)**
for PR 13526 at commit
Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
could we "rewind"/undo the append for the key and change it to a map that
inserts new values and key? so remove one append and replace it with another
operation?
---
If your project is set
Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
the tricky part with that is that (ds: Dataset[(K,
V)]).groupBy(_._1).mapValues(_._2) should return a
KeyValueGroupedDataset[K, V]
On Tue, Jun 7, 2016 at 8:22 PM, Wenchen Fan
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/13526
A possible approach maybe just keep the function given by `mapValues`, and
apply it before calling the function given by `mapGroups`. By doing this, we at
least won't make the performance worse,
Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
```
scala> val x = Seq(("a", 1), ("b", 2)).toDS
x: org.apache.spark.sql.Dataset[(String, Int)] = [_1: string, _2: int]
scala> x.groupByKey(_._1).mapValues(_._2).reduceGroups(_ +
Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
ok i will study the physical plans for both and try to understand why one
would be slower
---
If your project is set up for it, you can reply to this email and have your
reply appear on
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/13526
OK now I agree this is a useful API.
For performance, I would expect that
`ds.groupByKey(_._1).mapValues(_._2).mapGroups { case (k, vs) => (k, vs.sum) }`
should be at least as fast as
Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
can you explain a bit what is inefficient and would need an optimizer rule?
is it mapValues being called twice? once for the key and then for the new
values?
thanks!
---
If your
Github user koertkuipers commented on the issue:
https://github.com/apache/spark/pull/13526
see this conversation:
https://mail-archives.apache.org/mod_mbox/spark-user/201602.mbox/%3ccaaswr-7kqfmxd_cpr-_wdygafh+rarecm9olm5jkxfk14fc...@mail.gmail.com%3E
mapGroups is not a
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/60080/
Test PASSed.
---
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Merged build finished. Test PASSed.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/60082/
Test PASSed.
---
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Merged build finished. Test PASSed.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #60080 has
finished](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/60080/consoleFull)**
for PR 13526 at commit
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #60082 has
finished](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/60082/consoleFull)**
for PR 13526 at commit
Github user cloud-fan commented on the issue:
https://github.com/apache/spark/pull/13526
I doubt if this feature is really useful? I think users can easily call
`map` on the values during `mapGroups`, e.g.
```
val ds = Seq(("a", 10), ("a", 20), ("b", 1), ("b", 2), ("c",
Github user rxin commented on the issue:
https://github.com/apache/spark/pull/13526
cc @cloud-fan too
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #60082 has
started](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/60082/consoleFull)**
for PR 13526 at commit
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #60080 has
started](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/60080/consoleFull)**
for PR 13526 at commit
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Merged build finished. Test PASSed.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
Github user AmplabJenkins commented on the issue:
https://github.com/apache/spark/pull/13526
Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/60054/
Test PASSed.
---
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #60054 has
finished](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/60054/consoleFull)**
for PR 13526 at commit
Github user SparkQA commented on the issue:
https://github.com/apache/spark/pull/13526
**[Test build #60054 has
started](https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/60054/consoleFull)**
for PR 13526 at commit
45 matches
Mail list logo
