Pedro Ilídio

Regarding item 3, `scipy.stats.rankdata` consistently outperforms applying `numpy.argsort` twice, so we'd better switch to it. At this point, I think we can drop item 4's idea to avoid complicating the...

Based on `scipy`s implementation, I've rewritten the ranking step in the last commit to avoid sorting twice as before, using the already defined `sort` function of the `_splitter` module. A...

Thanks for your input, @betatim. I agree that precomputed ranks and a sparse version can wait for a future PR. I see that passing `X_ranks`, especially, would need deeper code...

I've modified the memory benchmarking script to use [filprofiler](https://github.com/pythonspeed/filprofiler), in order to track allocation instead of [resident memory](https://pythonspeed.com/articles/measuring-memory-python/). Results are more consistent, showing the expected 3-fold increase in memory usage...

> Looks like [ramnes/notion-sdk-py](https://github.com/ramnes/notion-sdk-py) is supported, although it has far less stars... And using the official API more directly, I suppose it's way simpler and easier to maintain and shouldn't...

It can be solved by disabling Dvorak, changing the first line of `.xinitrc` from ``` setxkbmap -layout br,dvorak -option "grp:alt_caps_toggle" ``` to ``` setxkbmap -layout br ```

Apparently this was solved in JuNest project using a custom version of `makepkg`: https://github.com/fsquillace/junest/issues/234, but I really don't know how all this stuff works. As I work without root in...