Added an 'elementary' function. - #4573

Im not sure whether this function should be added, but it will definitely need test cases as well if it is.

Tests are still forthcoming.

I haven't looked at this in detail, but yes, please don't use np.matrix in
new code or docs.

For now the best way to write matrix multiplication is just as a call to
np.dot.
On 9 Apr 2014 22:24, "Matt Pagan" notifications@github.com wrote:

Tests are still forthcoming.

—
Reply to this email directly or view it on GitHubhttps://github.com//pull/4573#issuecomment-40018304
.

So, I've added test coverage, switched to using np.dot instead of np.matrix, dropped the string formatting, and reworded the documentation. What do numpy devs think of this patch? Any other thoughts?

hm elementary matrices may be useful for mathematical notation, but why would you want to use them computations?
E.g. if you want to swap two rows of a matrix, it is far more efficient to just swap them instead of creating an elementary matrix and the doing a very expensive matrix multiplication.

You might want to compose several linear transformations by multiplying their corresponding matrices together.

hm it would still be very inefficient. But its probably useful for code clarity on operations where performance does not matter and could become more useful for other cases when numpy gains a smart matrix-dot chaining function.

Other cases I can imagine it might be useful:

• have an API that takes an arbitrary matrix, want to pass in an elementary
  matrix
• teaching
  On 26 Jul 2014 13:51, "Julian Taylor" notifications@github.com wrote:

hm it would still be very inefficient. But its probably useful for code
clarity on operations where performance does not matter and could become
more useful for other cases when numpy gains a smart matrix-dot chaining
function.

—
Reply to this email directly or view it on GitHub
#4573 (comment).

In [11]: elementary(3,0,1, multiplier=3)
Out[11]: 
array([[ 1.,  0.,  0.],
       [ 3.,  1.,  0.],
       [ 0.,  0.,  1.]])

is the last 1 on the diagonal correct here? it isn't a multiplier anymore in this case.

Needs rebase.

I was looking for a numpy elementary function that returns row j of the identity matrix without actually constructing the identity matrix, and I found this PR which does a different thing.