Skip to content

Latest commit

 

History

History
100 lines (82 loc) · 3.1 KB

manual_install.md

File metadata and controls

100 lines (82 loc) · 3.1 KB

! Note: Manual installation is only available for Linux operating systems.

Get the code

First, download and checkout the latest release

# from a directory of your choice
git clone https://github.com/weberlab-hhu/Helixer.git
cd Helixer
# git checkout dev # v0.2.0

System dependencies

Python 3.10

Python development libraries

Ubuntu (& co.)

sudo apt install python3-dev

Fedora (& co.)

sudo dnf install python3-devel

Virtualenv (optional)

We recommend installing all the python packages in a virtual environment: https://docs.python-guide.org/dev/virtualenvs/

For example, create and activate an environment called 'env':

python3 -m venv env
source env/bin/activate

The steps below assume you are working in the same environment.

Post processor

https://github.com/TonyBolger/HelixerPost

Setup according to included instructions and further add the compiled helixer_post_bin to your system PATH.

Most python dependencies of Helixer

We recommend to run pip install --upgrade pip1 and pip install wheel first before continuing with installing Helixer and it's requirements. So when in doubt, just run these commands first.

1 (required if instead of installing Helixer the output from pip is successfully installed UNKNOWN 0.0.0 or you get the error name, version not recognized (UNKNOWN 0.0.0)

# from the Helixer directory
pip install -r requirements.3.10.txt

Helixer itself

# from the Helixer directory
pip install .  # or `pip install -e .`, if you will be changing the code

Test Helixer

Helixer comes with test data and unit tests.

# switch to the Helixer code subdirectory
cd Helixer/helixer
# run the unit tests
pytest --verbose tests/test_helixer.py

GPU requirements (optional, but highly recommended for realistically sized datasets)

To run on a GPU (highly recommended for realistically sized datasets), tensorflow needs to be installed with its GPU requirements, see: https://www.tensorflow.org/install/gpu. It's recommended to install the cuda dependencies after the requirements and Helixer to prevent version mismatches.

# Linux instructions from the tensorflow GPU install
pip install tensorflow[and-cuda]  # inside a virtual environment, otherwise use: python3 -m pip install tensorflow[and-cuda]
# Verify the installation:
python3 -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

# sometimes the following error will pop up: Unable to register cuDNN factory... (and other factories)
# sometimes the following warning will pop up: tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
# usually those can be ignored and will not impair Helixer's performance

The following has been most recently tested.

system packages:

  • cuda-12-2
  • libcudnn8
  • libcudnn8-dev
  • nvidia-driver-555

A GPU with 11GB Memory (e.g. GTX 1080 Ti) can run the largest configurations described below, for smaller GPUs you might have to reduce the network or batch size.