Skip to content

Latest commit

 

History

History
329 lines (235 loc) · 9.43 KB

Installing_a_PySAGES_Environment.md

File metadata and controls

329 lines (235 loc) · 9.43 KB
Raw
jupyter
jupytext kernelspec
formats main_language text_representation
ipynb,md
python
extension format_name format_version jupytext_version
.md
markdown
1.3
1.16.4
display_name name
Python 3
python3

Setting up the Environment

We set up a directory that we will use as an installation prefix. If you are not running in an interactive environment and you don't want to install packages in a custom path, the steps in this section are unnecessary.

import os
import sys

ver = sys.version_info

os.environ["PYSAGES_ENV"] = os.environ["PREFIX"] = "/env/pysages"
os.environ["PYTHON_SITE_INSTALL_DIR"] = f"lib/python{str(ver.major)}.{str(ver.minor)}/site-packages"
os.environ["PREFIX_USER_SITE"] = os.environ["PREFIX"] + "/" + os.environ["PYTHON_SITE_INSTALL_DIR"]

# The following lines are to prevent python packages being looked up from certain paths in this Colab instance
for path in ("/content", ""):
  if path in sys.path:
    sys.path.remove(path)
!mkdir -p $PREFIX_USER_SITE

We want to make the installation visible to the python system, but we will be installing packages into a custom location that is not checked by python by default. In this Colab, we achieve this in two steps.

First, we extend the environment variable PYTHONPATH. This helps newly started python environments to find packages.

import os
os.environ["PYTHONPATH"] = os.environ["PREFIX_USER_SITE"] + ":" + os.environ["PYTHONPATH"]

Because the notebook environment has already a running python we need to let it know about the new location. We achieve this by appending the sys.path with such location.

import os
import sys
sys.path.append(os.environ["PREFIX_USER_SITE"])

Adding HOOMD-blue Support

We first install some dependencies necessary to build HOOMD-blue. These may vary in number and names based on your environment and operating system.

!apt-get -qq install libeigen3-dev pybind11-dev > /dev/null

Building and Installing HOOMD-blue

The following clones the HOOMD-blue repo and sets the version to v4.7.0 (this is the newest version that builds with the system dependencies available in Ubuntu 22.04, which is the OS used in Colab as of the end of 2024).

# Get HOOMD-blue source code
rm -rf hoomd-blue
git clone -q https://github.com/glotzerlab/hoomd-blue.git
cd hoomd-blue
git checkout -q v4.7.0
git submodule update -q --init

___

We need to patch CMake/hoomd/HOOMDPythonSetup.cmake to being able to install in a custom site-packages path within this Colab instance. This is also done for hoomd conda builds (see for example here). In general you shouldn't need to do this.

cd hoomd-blue
wget -q -O- https://raw.githubusercontent.com/conda-forge/hoomd-feedstock/4eb9b8ecd47f6780fcdbcde90ad99c180b5e2f51/recipe/fix-python-site-dir.patch | patch -p1 -s

___

We first disable some HOOMD-blue components to save on installation time, and then, we compile and install the package.

This may take a while, so be mindful of not inadvertently re-running the cell.

cd hoomd-blue

# Compile and install
BUILD_PATH=/tmp/build/hoomd
rm -rf $BUILD_PATH
cmake -S . -B $BUILD_PATH \
    -DCMAKE_INSTALL_PREFIX=$PREFIX \
    -DBUILD_HPMC=OFF \
    -DBUILD_METAL=OFF \
    -DBUILD_MPCD=OFF \
    -DBUILD_TESTING=OFF \
    -DENABLE_GPU=ON \
    -DENABLE_TBB=ON \
    -DPLUGINS="" \
    -DPYTHON_SITE_INSTALL_DIR=$PYTHON_SITE_INSTALL_DIR/hoomd > /dev/null

cmake --build $BUILD_PATH --target install -j8 > /dev/null
# Or alternately to the line above
# cd $BUILD_PATH
# make install -j8 > /dev/null

Building and Installing the HOOMD-dlext Plugin

Now we can install the dlext plugin for HOOMD-blue. But, we need to get some dependecies first.

!python -m pip install -q setuptools_scm > /dev/null

We then clone the hoomd-dlext repository and install the package via cmake as well. This cell is significantly faster than the HOOMD-blue installation.

# Get the plugin
rm -rf hoomd-dlext
git clone -q https://github.com/SSAGESLabs/hoomd-dlext.git
cd hoomd-dlext

# Build and install
BUILD_PATH=/tmp/build/hoomd-dlext
rm -rf $BUILD_PATH
cmake -S . -B $BUILD_PATH -DCMAKE_FIND_ROOT_PATH=$PREFIX &> /dev/null
cmake --build $BUILD_PATH --target install > /dev/null

This concludes the installation of the HOOMD-blue and its plugin for PySAGES. We quickly test the installation.

import hoomd
import hoomd.dlext

Adding OpenMM Support

Having previously set our the environment, we can now just simply install some required dependencies and build and install OpenMM.

Again, installing dependencies will be different depending on your operating system and python environment.

apt-get -qq install doxygen swig > /dev/null
python -m pip install -qq setuptools wheel Cython

Building and Installing OpenMM

The following clones the OpenMM repo and sets the version to v8.1.2 (the newest available when this notebook was last updated). Then, it configures and builds OpenMM.

This may take a while, so be mindful of not inadvertently re-running the cell.

# Get OpenMM source code
rm -rf openmm
git clone -q https://github.com/openmm/openmm.git
cd openmm
git checkout -q 8.1.2

# Compile and install
BUILD_PATH=/tmp/build/openmm
rm -rf $BUILD_PATH
cmake -S . -B $BUILD_PATH \
    -DCMAKE_INSTALL_PREFIX=$PREFIX \
    -DBUILD_TESTING=OFF \
    -DOPENMM_PYTHON_USER_INSTALL=ON \
    -Wno-dev > /dev/null

cmake --build $BUILD_PATH -j8 &> /dev/null
cmake --install $BUILD_PATH > /dev/null

The OpenMM python library needs to be build and installed separately. We also need to point for the library to be installed in our custom path.

export OPENMM_INCLUDE_PATH=$PREFIX/include
export OPENMM_LIB_PATH=$PREFIX/lib
BUILD_PATH=/tmp/build/openmm

# Install python package
cd $BUILD_PATH
make PythonInstall &> /dev/null

cd $BUILD_PATH/python
pip install --target $PREFIX_USER_SITE . &> /dev/null

Building and Installing the OpenMM-dlext Plugin

Similarly as shown for HOOMD-blue above, for OpenMM we need to build and install the corresponding openmm-dlext plugin.

# Get the plugin
rm -rf openmm-dlext
git clone -q https://github.com/SSAGESLabs/openmm-dlext.git
cd openmm-dlext

# Build and install
BUILD_PATH=/tmp/build/openmm-dlext
rm -rf $BUILD_PATH
cmake -S . -B $BUILD_PATH -Wno-dev > /dev/null
cmake --build $BUILD_PATH --target install &> /dev/null

If everything worked as expected, the following should run without issuing any errors.

import openmm
import openmm.dlext

Upload environment to Google Drive

These steps are not necessary to understand the setup of the environment. If you want to build your own environment, modify the lines such that it uploads to your own Google Drive.

We upload the data to a shared Google Drive so we can reuse our environment in other notebooks.

First, we mount our Google Drive file system to a local directory.

from google.colab import drive
drive.mount('/content/mnt')

We clean the cache of the code generated by python for our built packages such that the upload size is smaller.

python -m pip install -q pyclean > /dev/null
pyclean -q $PREFIX_USER_SITE

We then compress the environment into a zip file and copy it to a folder within Google Drive. Here we are choosing an existing Shared Drive, but if you were to do this you should choose a folder you have access to and write permissions.

%env PYSAGES_SHARED_ENV=/content/mnt/Shareddrives/pysages-env
cd $PYSAGES_ENV
zip -qr pysages-env.zip .
cp -f pysages-env.zip $PYSAGES_SHARED_ENV
rm -f pysages-env.zip