Skip to content

Latest commit

 

History

History
25 lines (21 loc) · 1.65 KB

README.md

File metadata and controls

25 lines (21 loc) · 1.65 KB
Raw

LibLR

A basic Python Logistic Regression package with C++ backend

This light-weight project that references wepe's dive-into-ml-systems is intended for practicing implementing ML algorithms in C++ with linear algebra libraries, and calling dynamic library functions in Python using ctypes. In comparison to the tools used in this project, Google's TensorFlow framework relies heavily on linear algebra library Eigen in its C++ kernels, and uses SWIG to build the Python-C++ interface.

Dependencies

  • C++: Armadillo
  • Python: ctypes, numpy

Implementations

  • src/LogisticRegression.h and src/LogisticRegression.cc implements a simple Logistic Regression classifier with L2 regularization using Armadillo, a popular linear algebra library that provides easy-to-use Matlab-like APIs.
  • src/CWrapper.cc exports C-style functions that will be packaged into a dynamic library liblr.so
  • python/LibLR/LogisticClassifier.py defines a LogisticClassifier class which performs training and inference by calling the dynamic library functions using ctypes
  • src/main.cc tests the C++ implementations by training and evaluating on randomly (but carefully) generated datasets, while python/test.py tests the Python LogisticClassifier implementations in the same way

Steps

  • Install Armadillo, and properly set up g++ linking options in Makefile
  • Build dynamic library and move it into python/LibLR folder
make lib
  • Train and evaluate the Python LogisticClassifier
python2.7 python/test.py