msmathcomp / hyperbolic-tsne Goto Github PK

View Code? Open in Web Editor NEW

8.0 8.0 5.0 37.49 MB

Experiments and content for the hyperbolic barnes-hut tSNE paper.

License: MIT License

Python 58.85% Cython 15.41% Jupyter Notebook 25.55% Dockerfile 0.19%

hyperbolic-tsne's People

Contributors

Stargazers

Watchers

Forkers

milan7843 xdead27 chgarth sne4kers

hyperbolic-tsne's Issues

Parameter `DOF` in tsne.pyx

The parameter dof in tsne.pyx does not seem to actually be used in anything. Confirm that and if true, remove the parameter.

Format and add documentation tsne_utils.pyx

Remove functions/code not necessary for the replication

Remove abstraction layers

Trim the abstraction layers for cost function, solver, optimizer, etc. to only include those needed for hyperbolic embeddings.

Setup instructions

Which are the requirements + creation of the environment
Installation of HyperbolicTSNE (optional depending on how we setup the project)
Walk through running an experiment
Further explanation of the structure of the repo and how to obtain the results reported in the paper

Format and add documentation util.py

Format and add documentation quality_evaluation_.py

Format and add documentation tsne.pyx

Format and add documentation initializations_.py

Format and add documentation hd_mat_.py

Remove Euclidean support

Trim the repository down to only hold hyperbolic methods.

Add link to data (or reference to where it can be found)

Format and add documentation optimizer_.py

Format and add documentation solver_.py

Update setup instructions with new examples in root of project directory

Refactor/add functions' documentation

Attribute the repository correctly

We originally ripped this off UMAP, right, should we attribute that?

Format and add documentation data_loaders.py

Refactor experiments code

Each experiment has a script that generates data and a script that analyses the data and produce results (e.g. tables and figures)

Format and add documentation visualization.py

Format and add documentation cost_functions_.py

User-friendly data loader

If the data has not been downloaded by the user, give a quick error as to that the data is missing and where to download it / place it after download.

Format and add documentation hyperbolic_tsne_.py

PyPI package or module with sys paths?

Do we want users to install the HyperbolicTSNE package before running the experiments? Or is it better to reduce the setup overhead and just set the sys paths so that the interpreter knows where to find the functions?