search

dattalab / keypoint-moseq

https://keypoint-moseq.readthedocs.io
Other
63 stars 25 forks source link

Installation on Linux x86_64 with NVIDIA GPU #125

Open dsanmiguel opened 5 months ago

dsanmiguel commented 5 months ago

When I followed the installation instructions I was not able to get a working installation of keypoint-moseq. Had various problems with compatibility especially with bokeh and running PCA in addition to other problems.

However, the following method was how I was able to get a working installation and with updated versions of everything. Not sure if the order of everything is relevant but this is the order in which I installed everything and as you can see I only had to specify versions for python and numpy.

conda create -n keypoint

conda activate keypoint

conda install mamba

mamba install python==3.10.* opencv pytables pip

mamba install jaxlib=*=*cuda* jax cuda-nvcc -c conda-forge -c nvidia

mamba install jupyterlab

mamba install numba chex jaxtyping tqdm optree

mamba install sphinx sphinx-rtd-theme autodocsumm myst-nb requests

mamba install seaborn cytoolz matplotlib tqdm ipykernel imageio[ffmpeg] statsmodels pyyaml holoviews[recommended] bokeh pandas networkx pynwb plotly ipython_genutils tabulate

/home/<yourusername>/anaconda3/envs/keypoint/bin/python3 -m pip install ndx_pose sleap_io vidio dynamax tensorflow tensorflow_probability

clone jax-moseq repository

uncomment all packages in jax_moseq setup.cfg and docs/requirements.txt except for jax_moseq, then in main directory

pip install -e .

clone keypoint-moseq repository

uncomment all packages in keypoint_moseq setup.cfg and docs/requirements.txt, then in main directory

pip install -e .

python3 -m ipykernel install --user --name keypoint --display-name "keypoint"

mamba install numpy==1.25