[SPARK-6278][MLLIB] Mention the change of objective in linear regression

[mengxr]

As discussed in the RC3 vote thread, we should mention the change of objective in linear regression in the migration guide. @srowen

[mengxr]

Apache is out-of-sync again. This is the only change in this PR.

[srowen]

chose -> choose, and I suppose it could be made a little more crystal-clear by saying the step size needs to be multiplied by 2 ('scaled' somehow could mean divide or multiply, to me).

[mengxr]

done.

[SparkQA]

Test build #28465 has started for PR 4978 at commit f87ae71.

• This patch merges cleanly.

[SparkQA]

Test build #28465 has finished for PR 4978 at commit f87ae71.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):
  • class KMeansModel (val clusterCenters: Array[Vector]) extends Saveable with Serializable

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/28465/
Test PASSed.

[SparkQA]

Test build #28506 has started for PR 4978 at commit bfd6cff.

• This patch merges cleanly.

[SparkQA]

Test build #28506 has finished for PR 4978 at commit bfd6cff.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/28506/
Test FAILed.

[srowen]

Hm, it also occurred to me that if the step size doubles, then it affects the regularization parameter as well. Doesn't it have to be half as large as well in order to get the same result? I'm probably overlooking something about the formulation, but I didn't see the reg param updated in a96b727 and if the loss term was halved, leaving all else equal, the regularization term is relatively twice as large right?

[mengxr]

Right. The L2 regularization term didn't change. So to generate the exact result, we need to reduce the regularization constant by half and multiply the step size by 2.

[SparkQA]

Test build #28539 has started for PR 4978 at commit fb3bbe6.

• This patch merges cleanly.

[SparkQA]

Test build #28539 has finished for PR 4978 at commit fb3bbe6.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/28539/
Test PASSed.

[mengxr]

Merged into master and branch-1.3.