I’ve been experimenting with using map with a two dimensional key(tuple) for a map as originally suggested.
But it seems that operations will be much slower. I’ve just tried to reimplement a couple of them and bear in mind that my code sucks, but sucks as much doing lists then doing maps I guess and still lists seems to be faster than maps.
This is the code, map implementation:
https://github.com/JordiPolo/dataframe/blob/map_experiment/lib/data_frame_map.ex
list implementation(has many more functions, do not care about most):
https://github.com/JordiPolo/dataframe/blob/map_experiment/lib/data_frame_map.ex
The benchmark:
https://github.com/JordiPolo/dataframe/blob/map_experiment/lib/dataframe/benchmark.ex
Results:
transpose(list of lists) 314.23
transpose(maps) 123.85 - 2.54x slower
tail(maps) 3303.23
tail(list of lists) 1539.64 - 2.15x slower
rows(list of lists) 85.83
rows(maps) 3.86 - 22.21x slower
columns(list of lists) 198.51
columns(maps) 129.22 - 1.54x slower
I’m specially worried by the columns function as I know the list version was done with no love for performance whatsoever, still seems to outperform a map with tuples as keys.
I’ll try Tensor next, although I want to keep experimenting with map a little bit more.
