[SPARK-5986][MLLib] Add save/load for k-means

[yinxusen]

This PR adds save/load for K-means as described in SPARK-5986. Python version will be added in another PR.

[SparkQA]

Test build #28393 has started for PR 4951 at commit dce7055.

• This patch merges cleanly.

[SparkQA]

Test build #28394 has started for PR 4951 at commit b144216.

• This patch merges cleanly.

[SparkQA]

Test build #28393 has finished for PR 4951 at commit dce7055.

• 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/28393/
Test PASSed.

[SparkQA]

Test build #28394 has finished for PR 4951 at commit b144216.

• 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/28394/
Test PASSed.

[mengxr]

• We don't need wrapper.serialize for vectors.
• We need to store cluster indices. If the centers are saved to more than one partitions, we cannot easily load them back in the original order.

[SparkQA]

Test build #28408 has started for PR 4951 at commit cd390fd.

• This patch merges cleanly.

[yinxusen]

How about using Int here to represent indexes? I think there is no need to use Long.

[mengxr]

Yes, Int is sufficient. This class should be private. Please also check the scope of other classes.

[mengxr]

make this private. The class name could be changed to case class Cluster(id: Int, center: Vector). IndexedPoint is too general.

[SparkQA]

Test build #28408 has finished for PR 4951 at commit cd390fd.

• 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
  • case class IndexedPoint(id: Int, point: Vector)

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/SparkPullRequestBuilder/28408/
Test PASSed.

[mengxr]

Be specific about the imports. You only need Vector.

[SparkQA]

Test build #28453 has started for PR 4951 at commit 6dd74a0.

• This patch merges cleanly.

[SparkQA]

Test build #28453 has finished for PR 4951 at commit 6dd74a0.

• 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/28453/
Test PASSed.

[mengxr]

LGTM. Merged into master. Thanks!