Justin Basilico

It looks like if you have a NaN element in the code it can cause vectors to be declared equal that are not. Example code: ``` assertFalse( VectorFactory.getDenseDefault().copyValues(Double.NaN).equals( VectorFactory.getDenseDefault().copyValues(0.0))); ```

Add support for doing regression with an L1 penalty.

DefaultWeightedPair is missing an equals and hashCode implementations.

Have the ability to just take in the data, normalize, train, and bake the normalization into the coefficients.

Add direct support for multiclass logistic regression.

There is a binary versus for all pairs and a winner-take-all but not an adapter for one class versus another.

Adding back SummaryStatistics implementations after 4.0.0 release. Pull request created to collect more feedback.

This could be a bit nicer of an interface than having to cast back and forth to doubles.

Add incremental versions of Bernoulli and Binomial distributions.

In the MutableInteger, MutableDouble, and MutableLong classes, add methods that take primitives for basic operations like +, -, *, and / along with the self-modifying versions.