Konstantin Tretyakov

So, apparently, the problem is not in the algorithm, but indeed in the memory consumption (the 64GB machine could handle the file). I tried to add a couple of changes...

No, I haven't tried neither Numba nor Cython because the initial aim was to have a super-simple pure-Python tool (also, numba did not even exist when this was first written,...

(I somehow missed your last comment before posting mine, now saw an email notification though). Nice, I'll check the ncl option and will probably replace my data structure with that...

Why not - it all boils down to taking the time to describe the model as an SK-expression. Given that SKLearn's boosting already works, it might not be too complicated....

I still did not get the time to look into this. Pull requests are still welcome ;)

Nice! `xgb_model.get_booster().get_dump(dump_format='json')` seems to make things really easy indeed. This should be pretty simple to implement in SKompiler as well.

The statement in setup.py `install_requires=["scikit-learn >= 0.22"],`, implies than an "explicitly supported version (i.e. last for which everything was known to be working) was 0.22.

True, given that the program creates folders with root permissions within cgroupfs, plain command-line options are probably unsafe. An /etc/isolate.conf, though, would be perfectly OK. Most importantly, it would make...

(.. and now I somewhy saw the later comment by lerks) If `/proc/cgroups/` is a universal way of figuring out the proper cgroup mount to use then looking the config...

The "config file solution" is simple and reliable enough to be implemented sooner rather than later (and it does not prevent adding autodetection later on). In this sense I agree...