Quickdraws relies on cuda-enabled pytorch for speed, and it is expected to work on most cuda-compatible Linux systems.
It is strongly recommended to either set up a python virtual environment, or a conda environment:
python -m venv venv
source venv/bin/activate
pip install --upgrade pip setuptools wheel
conda create -n quickdraws python=3.11 -y
conda activate quickdraws
pip install --upgrade pip setuptools wheel
It is necessary for anything bigger than toy examples to use either:
- on Linux, a cuda-enabled version of pytorch (now available by default from PyPI)
- on macOS, the latest pytorch, which can leverage the MPS backend
Use the pytorch configuration helper to find suitable installation instruction for your system, based on your preferred CUDA version.
If you want a specific CUDA version, you may need to start with something like pip install torch --index-url https://download.pytorch.org/whl/cu118.
The code snippet below will probably work for most systems, and should install quickdraws in approximately 10 minutes for cuda or much quicker on macOS:
pip install quickdraws
Once you install quickdraws, three executables should be available in your path:
convert-to-hdf5quickdraws-step-1quickdraws-step-2.
Clone the Git repository to access the example data and script demonstrating how these can be used:
git clone https://github.com/PalamaraLab/quickdraws.git
cd quickdraws
bash run_example.sh
To make changes to the quickdraws sourcecode, obtain the repository and install it using poetry. Assuming 8BCD you have poetry installed:
git clone https://github.com/PalamaraLab/quickdraws.git
cd quickdraws
poetry install
See https://palamaralab.github.io/software/quickdraws/manual/
See https://www.stats.ox.ac.uk/publication-data/sge/ppg/quickdraws/
For any technical issues please contact Hrushikesh Loya (loya@stats.ox.ac.uk)
Loya et al., "A scalable variational inference approach for increased mixed-model association power" under review
- Bug fix, introduced --chunksize argument in quickdraws-step-1 aswell
- Better memory usage and speed for step 0, introduced --chunksize argument
- Minor updates to documentation
- Remove reliance on specific CUDA torch for Linux
- Resolve numpy dependency conflict
- Minor updates to documentation
- Remove reliance on pre-release Torch for macOS
First public release to accompany the paper (see citation above).