legate-raft is a legate-enabled version of libraft
.
The project is still in an experimental state and mainly contains prototype versions
of naive bayes, knn, and kmeans.
Legate-raft is available as an anaconda package and can be installed using:
conda install -c legate -c rapidsai -c conda-forge legate-raft
To include development releases add the legate/label/experimental
channel.
The legate-raft
library depends on legate-dataframe
,
which in turn depends on legate
and cupynumeric
.
In most cases, you should not have to compile these from scratch. You can create a local development environment using conda:
conda env create -f conda/environments/all_cuda-125_arch-x86_64.yaml -n legate-raft-dev
conda activate legate-raft-dev
This will install all build and test requirements.
After this, you can build and install legate-raft
using:
./build.sh
note that the install is currently not an in-place build, so you need to rebuild to test Python changes.
The main exposed algorithm are currently:
- kNN via
from legate_raft.knn import run_knn
- kmeans via
from legate_raft.sklearn.cluster import KMeans
working similar to the sklearn version. - naive bayes via
from legate_raft.sklearn.naive_bayes import MultinomialNB
All of these and especially any further functionality should be considered to be in a prototype or exploratory stage.
Please see our our guide for contributing to legate-raft.