Cumsum prototype using dato distributed SFrame methods

dato_offset_cumsum.py

def offset_cumsum(sf, group_by, ordering, summed_column, starting_element=0, prefix='', suffix='_cumsum'):  
     import graphlab as gl
    from copy import copy
    from itertools import islice
    def _cumsum(l, initial, element=lambda (_, i): i, op=lambda acc, el: acc + el):
        prev = copy(initial)
        yield list(l[0] + (copy(initial),))
        if isinstance(initial, collections.Iterable):
            for i in islice(l, 1, None):
                prev = op(prev, list(element(i)))
                yield list(i + (copy(prev),))
        else:
            for i in islice(l, 1, None):
                prev = op(prev, element(i))
                yield list(i + (copy(prev),))

    new_column_name = prefix + summed_column + suffix
    cumsum_sf = sf.groupby(
        group_by,
        {summed_column: gl.aggregate.CONCAT(ordering, summed_column)}
    ).flat_map(
        [ordering, summed_column, new_column_name],
        lambda r: list(_cumsum(sorted(r[summed_column].items(), key=lambda k: k[0]), starting_element))
    ).sort(
        ordering
    )
    sf = sf.sort(ordering)
    sf[new_column_name] = cumsum_sf[new_column_name]
    return sf