Added Triangular type, At and Set methods, and tests

[btracey]

No description provided.

[sbinet]

naive comment: isn't "cleaner" to just spell it out
_ Matrix = (*Triangular)(nil)

admittedly more verbose, but no "pollution" of the global scope.

[btracey]

I'm following the pattern that exists there now. We can change it if there's a better way, but consistency is better.

[btracey]

Fixed comments and added a comment to NewTriangular about the underlying data. PTAL

[jonlawlor]

Is QR going to be changed to:

type QRFactor struct {
    QR    *Triangular
    rDiag []float64
}

[btracey]

I would vote it gets changed in some form. You can see my thoughts on Eigen at Issue #79 , and I have similar thoughts for Cholesky and QRFactor. I'm not a dictator though, we'll need to discuss how we want these to look. Getting the basic types in will make it easier to think about.

[btracey]

We should also consider if we want the Upper and Lower constants, or if we would rather have a boolean where true == Upper. There are only two possibilities, it would eliminate a panic possibility, and a boolean would match what we've chosen for Trans. However, it seemed to be that Upper and Lower read better, and it's not clear a priori which one should be true.

[jonlawlor]

You can type Upper and Lower to boolean and callers can just use true and false. http://play.golang.org/p/9FKN8h4SbT

[btracey]

Removed TriType and made the field a Boolean. We can revisit if we want to create a typed Boolean, but this is in line with the current Trans behavior.

[jonlawlor]

That bool change looks good to me; but I haven't reviewed all of the other changes in depth.

[btracey]

Pulled from master

[btracey]

Just made a new PR ( #94) that incorporates some changes from the symmetric discussion.