Installation

Installing from scratch

The following command will create a new environment called easymode and install easymode and all dependencies into it, including Ais and Pom. tensorflow[and-cuda] brings its own CUDA and cuDNN libraries, so no CUDA toolkit or library-path setup is needed.

conda create -n easymode python=3.10 git -c conda-forge
conda activate easymode

pip install "tensorflow[and-cuda]==2.15.1"
pip install easymode

Older setup: TensorFlow 2.11

easymode also runs on TensorFlow 2.11, the version it was developed with. In that case the CUDA libraries come from conda:

# Create environment with CUDA support
conda create -n easymode python=3.10 cudatoolkit=11.2 cudnn=8.1 git -c conda-forge
conda activate easymode

# Install packages
pip install tensorflow==2.11.0 protobuf==3.19.0
pip install easymode

# Set up CUDA library paths (one-time setup)
mkdir -p $CONDA_PREFIX/etc/conda/activate.d $CONDA_PREFIX/etc/conda/deactivate.d

cat > $CONDA_PREFIX/etc/conda/activate.d/cuda_env.sh << 'EOF'
export OLD_LD_LIBRARY_PATH="$LD_LIBRARY_PATH"
export LD_LIBRARY_PATH="$CONDA_PREFIX/lib:$LD_LIBRARY_PATH"
EOF

cat > $CONDA_PREFIX/etc/conda/deactivate.d/cuda_env.sh << 'EOF'
export LD_LIBRARY_PATH="$OLD_LD_LIBRARY_PATH"
unset OLD_LD_LIBRARY_PATH
EOF

# Re-activate to apply CUDA paths
conda deactivate
conda activate easymode

To use --xla (in easymode segment and easymode train) in this environment, also add the CUDA compiler tools once:

conda install -c nvidia cuda-nvcc=11.3.58
mkdir -p $CONDA_PREFIX/lib/nvvm/libdevice
ln -sf $CONDA_PREFIX/lib/libdevice.10.bc $CONDA_PREFIX/lib/nvvm/libdevice/libdevice.10.bc
export XLA_FLAGS=--xla_gpu_cuda_data_dir=$CONDA_PREFIX/lib

Environment settings

AreTomo3

In case you want to use easymode reconstruct, you need to set the path to the AreTomo3 binary and define the required AreTomo3 environment initialization command. For example:

easymode set --aretomo3-path /public/EM/AreTomo/AreTomo3 --aretomo3-env "module load AreTomo/3.1.0"

Cache directory

When using easymode segment <feature> for the first time, the required model weights will be downloaded and saved to a local cache directory (~500 MB per model). The default location is ~/easymode/, but you can change it (for example if you want a central cache for all your users) as follows:

easymode set --cache-directory /public/easymode/cache/