Installation¶

You can install featomic in different ways depending on which language you plan to use it from.

Installing the Python module¶

Pre-compiled wheels¶

The easiest way to install featomic is to use pip.

pip install --upgrade pip
pip install featomic

Building from source¶

If you want to build the code from source, you’ll need a Rust compiler, which you can install using rustup or your OS package manager; and git.

# Make sure you are using the latest version of pip
pip install --upgrade pip

git clone https://github.com/metatensor/featomic
cd featomic
pip install .

# alternatively, the same thing in a single command
pip install git+https://github.com/metatensor/featomic

Installing the C/C++ library¶

This installs a C-compatible shared library that can also be called from C++, as well as CMake files that can be used with find_package(featomic).

git clone https://github.com/metatensor/featomic
cd featomic/featomic
mkdir build
cd build
cmake <CMAKE_OPTIONS_HERE> ..
make install

The build and installation can be configures with a few cmake options, using -D<OPTION>=<VALUE> on the cmake command line, or one of the cmake GUI (cmake-gui or ccmake). Here are the main configuration options:

Option	Description	Default
CMAKE_BUILD_TYPE	Type of build: debug or release	release
CMAKE_INSTALL_PREFIX	Prefix in which the library will be installed	`/usr/local`
INCLUDE_INSTALL_DIR	Path relative to `CMAKE_INSTALL_PREFIX` where the headers will be installed	`include`
LIB_INSTALL_DIR	Path relative to `CMAKE_INSTALL_PREFIX` where the shared library will be installed	`lib`
BUILD_SHARED_LIBS	Default to installing and using a shared library instead of a static one	ON
FEATOMIC_INSTALL_BOTH_STATIC_SHARED	Install both the shared and static version of the library	ON
FEATOMIC_FETCH_METATENSOR	Automatically fetch, build and install metatensor (a dependency of featomic)	OFF

Using the Rust library¶

Add the following to your project Cargo.toml

[dependencies]
featomic = {git = "https://github.com/metatensor/featomic"}

Installing the TorchScript bindings¶

For usage from Python¶

You can install the code with pip:

pip install --upgrade pip
pip install featomic[torch]

You can also build the code from source

pip install --upgrade pip

git clone https://github.com/metatensor/featomic
cd featomic/python/featomic_torch
pip install .

# alternatively, the same thing in a single command
pip install git+https://github.com/metatensor/featomic#subdirectory=python/featomic_torch

For usage from C++¶

git clone https://github.com/metatensor/featomic
cd featomic/featomic-torch
mkdir build && cd build
cmake ..
# configure cmake if needed
cmake --build . --target install

Compiling the TorchScript bindings requires you to manually install some of the dependencies:

the C++ part of PyTorch, which you can install on it’s own. You can also use the installation that comes with a Python installation by adding the output of the command below to CMAKE_PREFIX_PATH:
```
python -c "import torch; print(torch.utils.cmake_prefix_path)"
```
the C++ interface of featomic, which itself requires the C++ interface of metatensor;
the TorchScript interface of metatensor. We can download and build an appropriate version of it automatically by setting the cmake option -DFEATOMIC_TORCH_FETCH_METATENSOR_TORCH=ON

If any of these dependencies is not in a standard location, you should specify the installation directory when configuring cmake with CMAKE_PREFIX_PATH. Other useful configuration options are:

Option	Description	Default
CMAKE_BUILD_TYPE	Type of build: debug or release	release
CMAKE_INSTALL_PREFIX	Prefix in which the library will be installed	`/usr/local`
CMAKE_PREFIX_PATH	`;`-separated list of path where CMake will search for dependencies.
FEATOMIC_TORCH_FETCH_METATENSOR_TORCH	Should CMake automatically download and install metatensor-torch?	OFF