[SPARK-5015] [mllib] Random seed for GMM + make test suite deterministic

[jkbradley]

Issues:

• From JIRA: GaussianMixtureEM uses randomness but does not take a random seed. It should take one as a parameter.
• This also makes the test suite flaky since initialization can fail due to stochasticity.

Fix:

• Add random seed
• Use it in test suite

CC: @mengxr @tgaloppo

[jkbradley]

Btw, in the docs, I did not say that the random seed is only used for initializing clusters since we may want to use it elsewhere in the future.

[mengxr]

LGTM.

[tgaloppo]

For the single cluster case, I don't think the seed can make a difference. With only one cluster, this just makes sure the right mean and covariance are computed correctly.

[jkbradley]

In general, you're correct. The issue is that, with some probability, the 5 data points chosen for initializing the cluster will be identical, causing initialization of the covariance matrix to fail.

I'd be fine with having this test use only 1 random seed instead of 5, but it's fast anyways.

[tgaloppo]

Got it!

[SparkQA]

Test build #25333 has finished for PR 3981 at commit f0df4fd.

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

[mengxr]

Merged into master.