_ _ _ _

Hierarchical Manifold Approximation and Projection (HUMAP) is a technique based on UMAP for hierarchical dimensionality reduction. HUMAP allows to:

1. Focus on important information while reducing the visual burden when exploring huge datasets;
2. Drill-down the hierarchy according to information demand.

The details of the algorithm can be found in our paper on ArXiv. This repository also features a C++ UMAP implementation.

-----------Installation -----------

HUMAP was written in C++ for performance purposes, and provides an intuitive Python interface. It depends upon common machine learning libraries, such as scikit-learn and NumPy. It also needs the pybind11 due to the interface between C++ and Python.

Requirements:

• Python 3.6 or greater
• numpy
• scipy
• scikit-learn
• pybind11
• pynndescent (for reproducible results)
• Eigen (C++)

If you have these requirements installed, use PyPI:

pip install humap

Alternatively (and preferable), you can use conda to install:

conda install humap

If using pip:

HUMAP depends on Eigen. Thus, make it sure to place the headers in /usr/local/include if using Unix or C:\Eigen if using Windows.

Manual installation:

For manually installing HUMAP, download the project and proceed as follows:

python setup.py bdist_wheel

pip install dist/humap*.whl

The simplest usage of HUMAP is as it follows:

Fitting the hierarchy

import humap
from sklearn.datasets import fetch_openml


X, y = fetch_openml('mnist_784', version=1, return_X_y=True)

# build a hierarchy with three levels
hUmap = humap.HUMAP([0.2, 0.2])
hUmap.fit(X, y)

# embed level 2
embedding2 = hUmap.transform(2)

Refer to notebooks/ for complete examples.

C++ UMAP implementation

You can also fit a one-level HUMAP hierarchy, which essentially fits UMAP projection.

umap_reducer = humap.UMAP()
embedding = umap_reducer.fit_transform(X)

Please, use the following reference to cite HUMAP in your work:

@misc{marciliojr_humap2021,
  title={HUMAP: Hierarchical Uniform Manifold Approximation and Projection}, 
  author={Wilson E. Marcílio-Jr and Danilo M. Eler and Fernando V. Paulovich and Rafael M. Martins},
  year={2021},
  eprint={2106.07718},
  archivePrefix={arXiv},
  primaryClass={cs.LG}
}

HUMAP follows the 3-clause BSD license and it uses the open-source NNDescent implementations from EFANNA.

E-mail me (wilson_jr at outlook.com) if you like to contribute.