Possible: Rename dimensionality to size about core.matrix

[mikera]

Do you mean dimensionality or dimensions?

For a 3x3 matrix, dimensionality is 2 (it's a 2D matrix) and dimensions are [3 3]

Apart from that I like the word size in principle but the only main I was cautious about size was ambiguity: it might get confused with the total number of items in the matrix. That it what it means in NumPy at least.....

Any examples of what other matrix libraries use?

[ejackson]

Sorry, yes, I mean dimensions. For this R uses dim, MATLAB uses size, mathematica uses Dimensions. So no consensus there :)

For the total number of elements matlab provides numel (i think its a bit ugly). How about count, or card ?

[mjwillson]

numpy uses "ndim" for the number of array dimensions, and "shape" for the tuple of sizes along those dimensions.

It also uses the term "axis" a lot in its API for a specific dimension of an ndarray

[mikera]

I quite like "shape" as an alternative to "dimensions"

ecount might be good as a reasonably concise term for the total number of elements - it would be consistent with ereduce, emap etc, for the other element-by-element operations.

ndim feels unidiomatic to me in Clojure for some reason. Though I recognise I have a general bias against use of abbreviations....

[ejackson]

Yes, shape is nice, sounds like quorum to me :)

ecount I think is a problem. The e_verb_ convention is to apply verb to each element of the matrix, thus wouldn't ecount imply: apply the function count to each element ? The count we want is not an elementwise op, but a collectionwise op. Admittedly, this might be inferring too much from the convention.

I still think count or card are better.

[mikera]

OK sounds agreed dimensions -> shape

OTOH I don't think e*verb* does currently mean apply *verb* to each element at present - ereduce for example isn't applying reduce to each element, it is applying reduce to a hypothetical sequence of all elements. Likewise for emap. e+ works as shorthand for emap + in this sense.

ecount would be consistent with that, the sense that (count seq) returns the count of all elements in a seq.

Without having a specific formal definition in mind, I was interpreting e_something_ as "version of something that works over the array seen as a collection of elements". Rather like areduce for Java arrays in clojure.core. I imagine clojure.core would have acount as well were it not for the fact that count already works fine on Java arrays.

This is good discussion though - let's keep pushing this, naming is so important for any API.

[mikera]

Anyone want to do a patch for dimensions -> shape?

Happy to do it myself, but I'm conscious that other people may want to get involved with hacking the API and this is probably a good one to start with.....

[ejackson]

e_verb_. I see your point. I need to think about this a bit.

[mikera]

Seems like we can close this one.