[ENH] Add cummin() and cummax() functions for cumulative min/max calculation

[samukweku]

WIP for #3279

[oleksiyskononenko]

Thanks @samukweku, I guess this could be easily templated to support both cummax() and cummin() at the same time. Look at how this is implemented for rowmax()/rowmin(), for example.

[oleksiyskononenko]

Look at this example. In the current implementation cummax() becomes zero once a missing value is encountered:

>>> from datatable import dt
>>> DT = dt.Frame([-1, None, 1])
>>> DT[:, dt.f.C0.cummax()]
   |    C0
   | int64
-- + -----
 0 |    -1
 1 |     0
 2 |     1
[3 rows x 1 column]

Once we fix the issues I've mentioned in the comments above the output will be

   |    C0
   | int64
-- + -----
 0 |    -1
 1 |    -1
 2 |     1
[3 rows x 1 column]

and that will be more consistent with the regular max() function.

[oleksiyskononenko]

Btw, I've just pushed a fix for the sphinx build configuration. You should merge main onto your branch to make your build green.

[samukweku]

Thanks @samukweku, I guess this could be easily templated to support both cummax() and cummin() at the same time. Look at how this is implemented for rowmax()/rowmin(), for example.

thanks for the hint ... I'll give it a shot

[oleksiyskononenko]

@samukweku the building error is not related to the changes on this PR, just some glitch with the jenkins filesystem:

[2022-06-04T00:58:55.741Z]     def test_fread_from_url2():
[2022-06-04T00:58:55.741Z]         import pathlib
[2022-06-04T00:58:55.741Z]         root = os.path.join(os.path.dirname(__file__), "..", "..")
[2022-06-04T00:58:55.741Z]         path = os.path.abspath(os.path.join(root, "LICENSE"))
[2022-06-04T00:58:55.741Z]         url = pathlib.Path(path).as_uri()
[2022-06-04T00:58:55.741Z]         d0 = dt.fread(url, sep="\n")
[2022-06-04T00:58:55.741Z]         frame_integrity_check(d0)
[2022-06-04T00:58:55.741Z] >       assert d0.shape == (372, 1)
[2022-06-04T00:58:55.741Z] E       assert (0, 0) == (372, 1)
[2022-06-04T00:58:55.741Z] E         At index 0 diff: 0 != 372
[2022-06-04T00:58:55.741Z] E         Full diff:
[2022-06-04T00:58:55.741Z] E         - (372, 1)
[2022-06-04T00:58:55.741Z] E         + (0, 0)

So just continue what you're doing, if the error persists, I will look into it.

[oleksiyskononenko]

@samukweku Let me know if you have any questions on the templated implementation of cummin()/cummax() functions.

[samukweku]

thanks @oleksiyskononenko , I understand the template implementation. I'm sure I'll come back to it several times to study it more.

[samukweku]

@oleksiyskononenko kindly have a look, when you can; I have updated the docs for cumin and added tests for f.cummin