Skip to content

Install hep-ml-lab with pip#

Prerequisites#

  • Python >= 3.9
  • ROOT v6.26/14 (other versions should be OK. Required by Delphes)
  • Madgraph5 v3.5.3 (other versions should be OK)
    • pythia8 v8.306 (installed by Madgraph5)
    • Delphes v3.5.0 (installed by Madgraph5)

The keras 3.0 requires the tensorflow 2.16 or higher, which is also the latest version. Before installing tensorflow, you need to install the following NVIDIA® softwares:

Check the offical page of tensorflow for more details.

Installation#

pip install hep-ml-lab

This will install the latest version of keras without any backends such as tensorflow, pytorch or jax. Since we have only tested hep-ml-lab on tensorflow, here we show how to install tensorflow as a backend. Then you can install tensorflow with the following command

pip install tensorflow[and-cuda]

Fix the issue with tensorflow 2.16#

There's an known issue with tensorflow 2.16: it couldn't recognize GPUs smoothly like before. So we have to link the related libraries manually. Use the following script to do so:

# set_nvidia.sh

# Attempt to locate the NVIDIA cudnn library file using Python.
NVIDIA_DIR=$(python -c "import nvidia.cudnn; print(nvidia.cudnn.__file__)" 2>/dev/null)

# Check if the NVIDIA directory variable is set (i.e., cudnn was found).
if [ ! -z "$NVIDIA_DIR" ]; then
    # Get the parent directory of the directory containing the __file__
    NVIDIA_DIR=$(dirname $(dirname $NVIDIA_DIR))

    # Iterate over all directories in the NVIDIA package directory.
    for dir in $NVIDIA_DIR/*; do
        # Check if the directory has a 'lib' subdirectory.
        if [ -d "$dir/lib" ]; then
            # Prepend the library path to LD_LIBRARY_PATH.
            export LD_LIBRARY_PATH="$dir/lib:$LD_LIBRARY_PATH"
        fi
    done
fi

Run it in the environment where you installed tensorflow:

source set_nvidia.sh

Now you can test if tensorflow recognizes your GPUs:

python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

If it is set up successfully, you will see the following output: 

[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

If you are in a conda environment and want to source it automatically, you can follow the steps below:

# Step1: Check the path of the conda environment
conda env list

# the path should look like
# /root/miniconda3/envs/py39
cd <path_to_the_environment>

# Step2: Create the folder if it doesn't exist
mkdir -p etc/conda/activate.d

# Step3: Copy the script to the folder
cp set_nvidia.sh etc/conda/activate.d/set_nvidia.sh

# Step4: Reopen the conda environment
conda activate <env_name>