Installation

Stable release

Python

To install abess on Python, you can simply get the stable version with:

$ pip install abess

If you don't have pip installed, this Python installation guide can guide you through the process.

R

To install stable version into R environment, run the command:

install.packages("abess")

Latest version

This page gives instructions on how to build and install abess from the source code. If the instructions do not help for you, please feel free to ask questions by opening an issue.

First of all, clone our the latest github project to your device:

$ git clone https://github.com/abess-team/abess.git
$ cd abess

Next, there have different processing depend on the programming langulage you prefer.

Python

Before installing abess, some dependent libraries should be installed first, which may be a little different in different platforms:

  • Linux: $ sudo apt install bash (for Ubuntu, but other Linux systems are similar);

  • Windows: $ choco install git (using Chocolatey), or manually install the software and add them into PATH; Windows user will also need to download Microsoft C++ Build Tools, and then install the "Desktop development with C++" module inside.

  • MacOS: $ brew install bash (using Homebrew).

Then, you need to install pybind11 via pip install pybind11 (or other methods). After that, we can manually install abess by conducting command:

$ cd ./python
$ pip install .

or

$ cd ./python
$ python setup.py install --user

If it finishes with Finished processing dependencies for abess, the installation is successful.

Alternatively, if you would like to develop abess, install abess in editable mode (it is very convenient for development):

$ cd ./python
$ pip install -e .

or

$ cd ./python
$ python setup.py develop --user

Note that some may meet "Permission denied" problem like this issue when installing with pip install -e .. There are three solutions: 1. run the command as administrator; 2. feel free to use python setup.py develop --user instead; 3. try to edit setup.py like here (not recommend).

R

To install the development version, some dependencies need to be installed. Before installing abess, some dependencies should be installed first, which may be a little different in different platforms:

  • Linux: $ sudo apt install autoconf (for Ubuntu, other Linux systems are similar);

  • Windows: install Rtools.

  • MacOS: $ brew install autoconf.

Then, you need to install R library dependencies Rcpp and RcppEigen via conducting install.packages(c("Rcpp", "RcppEigen")) in R console.

After installing dependencies, run the following code in terminal/bash:

If it finishes with * DONE (abess), the installation is successful.

Dependencies

C++

Our core C++ code is based on some dependencies:

  • Eigen (version 3.3.4): a C++ template library for linear algebra: matrices, vectors, numerical solvers, and related algorithms.

  • Spectra (version 1.0.0): a header-only C++ library for large scale eigenvalue problems.

They would be automatically included while installing the abess packages.

OpenMP

To support OpenMP parallelism in Cpp, the dependence for OpenMP should be install. Actually, many compliers and tools have supported and you can check here.

What is more, if you receive a warning like “Unknown option ‘-fopenmp’” while installing abess, it means that OpenMP has not been enabled. Without OpenMP, abess only use a single CPU core, leading to suboptimal learning speed.

To enable OpenMP:

  • In Windows, Visual C++ or many other C++ compliers can support OpenMP API, but you may need to enable it manually in additional features (based on the complier you use).

  • In Linux, the dependence would be supported if GCC is installed (version 4.2+).

  • In MacOS, the dependence can be installed by:

    $ brew install llvm
    $ brew install libomp
    

Python

Some basic Python packages are required for abess. Actually, they can be found on abess/python/setup.py and automatically installed during the installation.

  • pybind11: seamless operability between C++11 and Python

  • numpy: the fundamental package for array computing with Python.

  • scipy: work with NumPy arrays, and provides many user-friendly and efficient numerical routines.

  • scikit-learn: a Python module for machine learning built on top of SciPy.

  • pandas: support data manipulation and input.

Furthermore, if you want to develop the Python packages, some additional packages should be installed:

R

The R version should be 3.1.0 and newer in order to support C++11. abess R package relies on limited R packages dependencies:

  • Rcpp: convert R Matrix/Vector object into C++.

  • RcppEigen: linear algebra in C++.

Furthermore, if you would to develop the R package, it would be better to additionally install: