Skip to content

Latest commit

 

History

History
143 lines (100 loc) · 4.42 KB

INSTALL.md

File metadata and controls

143 lines (100 loc) · 4.42 KB
Raw

Setup Instructions on a GPU Machine

Important These steps should all be performed on a GPU machine.

0. Clone the SALIENT++ artifact repo

Before you with installing the SALIENT++ artifact dependencies, please make sure you have cloned this repo locally and changed into the cloned repo directory. Steps 5 & 6 below assume that your current working directory is the SALIENT++ cloned repo directory.

git clone [email protected]:MITIBMxGraph/SALIENTplus_artifact.git
cd SALIENTplus_artifact

1. Install Conda

Follow instructions on the Conda user guide. For example, to install Miniconda on an x86 Linux machine:

curl -Ls https://repo.anaconda.com/miniconda/Miniconda3-${CONDA_VER}-Linux-${conda_arch}.sh -o /tmp/Miniconda.sh &&\
bash Miniconda3-py38_4.10.3-Linux-x86_64.sh

SALIENT++ has been tested on Python 3.9.5.

It is highly recommended to create a new environment and do the subsequent steps therein. Otherwise, it is possible that you will run into unexpected issues with the installation.

For example, to create a new environment called salientplus:

conda create -n salientplus python=3.9.5 -y
conda activate salientplus

2. Install PyTorch

Follow instructions on the PyTorch homepage. For example, to install on a linux machine with CUDA 11.7:

conda install -y pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.7 -c pytorch -c nvidia

SALIENT++ should be compatible with newer versions of PyTorch (e.g., PyTorch 2), but we have not tested it. SALIENT++ was developed against PyTorch 1.10 and has been tested up to PyTorch 1.13.1.

3. Install OGB

conda install -y -c conda-forge ogb

SALIENT++ has been tested on OGB 1.3.5.

4. Install PyTorch-Geometric (PyG) and PyTorch Sparse

To get the latest version of PyG, follow the instructions on the PyG Github page.

conda install -y pyg -c pyg -c conda-forge
conda install -y pytorch-sparse -c pyg

SALIENT++ has been tested on PyTorch Geometric version 2.2.0.

5. Install SALIENT++'s fast_sampler

Important As with prior steps, you must install the fast_sampler module on a GPU machine.

Build and install the local fast_sampler module:

cd fast_sampler
python setup.py install
cd ..

To check that fast_sampler is properly installed, start python and run:

>>> import torch
>>> import fast_sampler
>>> help(fast_sampler)

You should see information of the package.

NOTE: Compilation requires a C++ compiler that supports C++17 (e.g., gcc >= 7).

6. Install METIS (optional, if using pre-generated partitions)

The following dependencies are required only if you plan to use our scripts for partitioning graph datasets (e.g., partitioners/run_4constraint_partition.py).

We recommend that you install METIS from source using our provided repository. We require that METIS is built with 64-bit types, which precludes the use of commonly distributed METIS libraries in existing packages.

git clone [email protected]:MITIBMxGraph/METIS-GKlib.git
cd METIS-GKlib
make config shared=1 cc=gcc prefix=$(realpath ../pkgs) i64=1 r64=1 gklib_path=GKlib/
make install

Next, install torch-metis, which provides a Python module named torch_metis with METIS bindings that accept PyTorch tensors.

cd ..
git clone [email protected]:MITIBMxGraph/torch-metis.git
cd torch-metis
python setup.py install
cd ..

NOTE: The torch_metis module normally requires some configuration of environment variables in order to function properly. In our attempt to streamline the process of exercising the SALIENT++ artifact, we made sure that the relevant scripts that use METIS set these variables internally. Please be advised that, outside of these artifact scripts, import torch_metis will not work without setting the necessary environment variables.

7. Install other dependencies

Install the following extra dependencies.

conda install -y -c conda-forge nvtx
conda install -y -c conda-forge matplotlib
conda install -y -c conda-forge prettytable

8. Post-installation: activating the SALIENT++ environment

To re-enable the SALIENT++ environment and exercise the artifact in a new terminal session, simply activate the salientplus conda environment:

conda activate salientplus