Break isnull type operations out into a separate expression

[cpcloud]

Currently the isnan method uses np.isnan or math.isnan. We should add an additional isnull that checks for null values.

Most backends have something that corresponds to this operation.

cc @quasiben @chdoig @mrocklin

[mrocklin]

If this is commonly used in the following way

col[~col.isnan]

Then we might consider the following operation instead

col.nonnull

[cpcloud]

pandas actually has two names for this dropna and valid

just fyi

[cpcloud]

aesthetically not a fan of nonnull, but i could go for dropnull

[mrocklin]

Yeah, dropnull sounds better. We should shop around a few names to other folks at some point.

What is more convenient from a computational perspective, isnull or dropnull? This should probably drive the creation of the Expression. The syntax/names can be decided after the fact.

[cpcloud]

I think both should exist. Isnull gives you an indicator variable and drop null gives you values where the result of isnull is false.

[mrocklin]

In regards to syntax I agree. In regards to computation I think that this requires us to look at how computational backends handle things. For example. If everything is like numpy then we don't need to bother with dropnull; isnull is sufficient.