motionmapperpy : Modified Python 3.0 implementation of MotionMapper

This package is a GPU accelerated implementation of the MotionMapper pipeline for creating low dimensional density maps using tSNE or UMAP. Some methodologies may differ from the original implementation, please refer to the source code for a detailed look.

Package functions are:

• Subsampling training points by running mini-tSNEs on a group of datasets.
• Re-embedding new points on a learned tSNE map.
• Watershed segmentation and grouping.

Installation:

Using a conda environment or pip

1. (OPTIONAL) Create a new conda environment conda create -n mmenv python=3.6
2. Activate desired conda environemnt conda activate mmenv
3. Download the repository and unzip contents. Open terminal and navigate to unzipped folder containing setup.py.
4. Run

pip install -U h5py==2.1 
pip install numpy scikit-image hdf5storage
python setup.py install

Using the supplied Pixi environment

1. Download and install pixi curl -fsSL https://pixi.sh/install.sh | sh
2. Download the repo and unzip contents. Open terminal and navigate to unzipped folder containing pixi.toml.
3. Run

pixi install
pixi s

Additionally, install cupy (if GPU present on system) by following the instructions here.

Demo.

After installation, run "cd demo && python3 demo.py".

Issues:

Please post any code related issues at https://github.com/bermanlabemory/motionmapperpy/issues with a complete error trace where possible.