Matrix row/column sum/mean

[yuzeh]

Added some code for matrix {row, column} {sum, mean}.

M.sum(dim) returns the sum along dim (0 => sum rows, 1 => sum columns, default 0).
M.mean(dim) returns the mean along dim (same semantics as before).

If dim is not 0 or 1, null is returned.

The return value of M.sum and M.mean are Vector objects.

Unit tests are included.

[nihonjinrxs]

This looks great -- I'd provide some aliases, as stated in the line comments on your commit. @jcoglan, can we get this added to the mainline repo soon?

[jcoglan]

I'm not convinced these methods belong in the library since they have little to do with vector/matrix operations. Am I missing a use case? The implementations are also sub-optimal, making internal use of public APIs rather than accessing the elements array directly, and use Array.reduce() which does not exist on some JS platforms (I need to check what support for this is like these days).

[jcoglan]

Also, I think using numbers to indicate whether you wan rows or columns is a poor interface. The name of the method should indicate what it does rather than relying on magic numbers.

[yuzeh]

Sorry I didn't get to responding this sooner.

With respect to the two comments:

1. The interface/implementation can obviously be changed.
2. I wrote this pull request at a point when I was trying to do some machine learning in the browser. I've since given up on that.

One use case: In Markov models, we generally want a matrix M of transition probabilities, where entry M(i, j) is the probability of starting in state i and transitioning to state j in one timestep. Naturally, summing M(i, j) over all j should equal 1 for every i. Sometimes, then, the row sum is used to normalize probabilities.

Anyways, most of this comes from my background in MATLAB, and I have since given up on trying to do anything remotely matrix related in the browser. Let me know if you still want this pull req to go through. If you do, I'll polish it and resend it.