dpu-kmeans 0.1.6

A package for the k-means algorithm on DPU.

dpu_kmeans

CI	status
pip builds

A project built with pybind11 and scikit-build, running the KMeans algorithm on in-memory processors with the UPMEM SDK.

Installation

• install the UPMEM SDK
• pip install dpu-kmeans

Usage

import numpy as np
from dpu_kmeans import DIMM_data, KMeans

X = np.array([[1, 2], [1, 4], [1, 0],[10, 2], [10, 4], [10, 0]])
dimm_data = DIMM_data(X)

kmeans = KMeans(2)

centroids, iterations, time = kmeans.fit(dimm_data)
print(centroids)

Output:

[[ 0.9998627  2.       ]
 [10.000137   2.       ]]

Alternatively you can import data from a CSV file:

dimm_data = DIMM_data("/path/to/data")

Development

• clone this repository
• install the UPMEM SDK
• install the build requirements in pyproject.toml
• cd dpu_kmeans
• pre-commit install
• pip install -e .
• python setup.py clean

OR

• clone this repository
• open folder in VS Code
• start in Dev Container

to debug: python setup.py develop --build-type Debug

Note: The dev container is for development only and uses the PIM simulator.

Testing

• clone this repository
• install the UPMEM SDK
• install nox
• cd dpu_trees
• nox

OR

• clone this repository
• open folder in VS Code
• start in Dev Container
• nox

Note: nox, python setup.py and pip might fail if you executed pip install -e . previously, delete the _skbuild cache or run python setup.py clean to solve.

Test call

import dpu_kmeans
dpu_kmeans.test_checksum()

Expected return: 0x007f8000