Implement first/last reductions

[cpcloud]

No description provided.

[jreback]

prob should also do tail for these back ends as well (I think that is an issue already)

[cpcloud]

that would go in a separate pr and would probably just be what @mrocklin said:

def tail(expr, n=10):
    return expr[-n:]

@jreback first blaze PR ? :)

[cpcloud]

another approach to first/last would be to can this PR and make arbitrary expressions work with groupbys

[cpcloud]

then we wouldn't really need a first/last expression, we could use expr[0] and expr[-1]

[mrocklin]

What is the benefit to this over Slice?

[cpcloud]

Slice doesn't work in reductions

[cpcloud]

Sorry in summarys

[cpcloud]

I'm going to play with this today and see if I can get slices to work in groupbys. If it's simple then I'll close this PR. I DO think that first and last are more readable names for these computations, but that's subjective and I don't feel too strongly about it.

[mrocklin]

They can be names / user-functions without being expressions. Probably
various checks need to be changed from isinstance(expr, Reduction) to
not iscollection(expr.dshape).

On Sat, Dec 20, 2014 at 8:07 AM, Phillip Cloud notifications@github.com
wrote:

I'm going to play with this today and see if I can get slices to work in
groupbys. If it's simple then I'll close this PR. I DO think that first and
last are more readable names for these computations, but that's subjective
and I don't feel too strongly about it.

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

[cpcloud]

@mrocklin whenever you get a chance.

@mrocklin can i ask that you rebase off of master after this PR instead of merging? i think if you merge master into your branch that many of the dead imports that i removed here will come back.

[cpcloud]

The only thing I haven't done here is allow first and last to work with reductions in pandas because we can only do agg at the moment and first and last are Slices. Even though pandas does have a first/last reduction on GroupBys we don't actually know that slice[0] is called first and slice[-1] is called last.

[cpcloud]

we should merge #832 (negative slicing with toolz.tail) before this PR

[mrocklin]

Need to make sure that this is the same as DataShape(10, 10, bool_) rather than DataShape(10, 10, DataShape(bool_))

[jreback]

q -> English like bah!

[cpcloud]

happy to remove first and last here, i simply wanted expr[0] and expr[-1] to work in the context of a groupby, the change from _dtype to schema fixes that

[mrocklin]

Ah, so we don't need first and last methods?

[cpcloud]

no those aren't strictly necessary

[cpcloud]

@mrocklin can this be merged?

[mrocklin]

Looks great. Merging.