Most of the behavior for the map operator proposed in the Modin research papers is actually covered by the broadcast_* methods on dataframe/partition manager classes. We should refactor the codebase to use the logic for broadcast within map instead, which hopefully enables performance improvements.
map should also have an optional dtypes argument to hint the type of the resulting dataframe.
See also: stale algebra tracking issue #2924.
Most of the behavior for the
mapoperator proposed in the Modin research papers is actually covered by the broadcast_* methods on dataframe/partition manager classes. We should refactor the codebase to use the logic forbroadcastwithinmapinstead, which hopefully enables performance improvements.mapshould also have an optionaldtypesargument to hint the type of the resulting dataframe.See also: stale algebra tracking issue #2924.