dpu_kmeans README

CI	status
pip builds
docs

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

Documentation

• API documentation

Installation

• install the UPMEM SDK
• pip install dpu-kmeans

If you get an error about RANLIB, create a symlink to llvm-ar:

sudo ln -s /usr/bin/llvm-ar /usr/bin/llvm-ranlib

Usage

import numpy as np
from dpu_kmeans import KMeans

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

kmeans = KMeans(2)
kmeans.fit(X)

print(kmeans.cluster_centers_)

Output:

[[ 0.99986267  2.        ]
 [10.00013733  2.        ]]

Development

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

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.

Templating

To use this project as a base for your own UPMEM DIMM project:

• click on "Use this template" in github
• create a new project from this one
• turn off Conda and Wheels workflows in github actions as they are not operational right now
• change folder src/dpu_kmeans to src/<your_project>
• change project name (all instances of dpu_kmeans) and info in:
  • README.md
  • setup.cfg
  • setup.py (cmake_install_dir="src/dpu_kmeans")
  • .gitignore (src/dpu_kmeans/dpu_program/)
  • CMakeLists.txt (project(dpu_kmeans VERSION ${VERSION}))
  • conda.recipe/meta.yaml (optional)
  • docs (optional)
• if you intend to use github actions to auto-publish to pypi, update the project secrets as described in Publishing package distribution releases using GitHub Actions CI/CD workflows

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