Github user jkbradley commented on the pull request:
https://github.com/apache/spark/pull/2595#issuecomment-58065527
The regressions are significant in some cases. I agree we may want to test
more, and maybe support 2 options eventually. Some results below. The first
set are mostly regressions; the second set are improvements.
EC2 cluster with 16 nodes, trees of depth 5, synthetic data from spark-perf:
examples | features | nTrees | feat/node | speedup
------- | --------- | ----------- | ------- | --------
20000 | 100 | 1 | 100 | 0.703125
20000 | 1500 | 1 | 1500 | 0.5658691857
2000000 | 100 | 1 | 100 | 1.081596993
2000000 | 1500 | 1 | 1500 | 0.6293609989
20000 | 100 | 5 | 10 | 0.6894015716
20000 | 1500 | 5 | 39 | 0.639494333
2000000 | 100 | 5 | 10 | 1.161488822
2000000 | 1500 | 5 | 39 | 0.9352459136
Same cluster, but using MNIST8m data from libsvm datasets (8M instances,
784 features):
nTrees | maxDepth | runtime | speedup
------- | --------- | ----------- | -------
1 | 5 | 89.03475246 | 1.004851675
1 | 9 | 35.82063408 | 1.21802967
5 | 5 | 155.8450781 | 1.265042585
5 | 9 | 74.30615172 | 4.541347304
---
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 if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
