<p align="center"> <img src="image/spacemap-main.png" height="220"/> <img src="image/spacemap-crop.gif" height="220"/> </p>

SpaceMAP

SpaceMAP is a dimensionality reduction method utilizing the local and global intrinsic dimensions of the data to better alleviate the 'crowding problem' analytically.

Paper

https://icml.cc/virtual/2022/spotlight/18170

https://proceedings.mlr.press/v162/zu22a.html

Hyper-parameters

SpaceMAP has 4 main hyper-parameters: n-near/n-middle and d-local/d-global, which define the intrinsic dimensions and the hierarchical manifold approximation.

• n-near: number of neighbors in the near fields of each data point. (default: 20)
• n-middle: number of neighbors in the middle field of each data point. (default: 1% of the whole dataset)
• d-local: estimated intrinsic dimensions of the near fields of each data point. (default: Auto)
• d-global: estimated intrinsic dimension of the whole dataset. (default: Auto)

Installation

pip install spacemap-dr

Usage

from spacemap import SpaceMAP
import numpy as np

data = np.random.rand(1000, 100)
# Auto mode
spacemap = SpaceMAP()
# Manually change the parameters
spacemap = SpaceMAP(n_near_field=10, n_middle_field=100, d_local=0, d_global=0)

embedding = spacemap.fit_transform(data)

Example

Please refer to the example for details.

Citation

If you find SpaceMAP useful in your research, please consider citing:

@InProceedings{pmlr-v162-zu22a,
  title = 	 {{S}pace{MAP}: Visualizing High-Dimensional Data by Space Expansion},
  author =       {Zu, Xinrui and Tao, Qian},
  booktitle = 	 {Proceedings of the 39th International Conference on Machine Learning},
  pages = 	 {27707--27723},
  year = 	 {2022},
  editor = 	 {Chaudhuri, Kamalika and Jegelka, Stefanie and Song, Le and Szepesvari, Csaba and Niu, Gang and Sabato, Sivan},
  volume = 	 {162},
  series = 	 {Proceedings of Machine Learning Research},
  month = 	 {17--23 Jul},
  publisher =    {PMLR},
  pdf = 	 {https://proceedings.mlr.press/v162/zu22a/zu22a.pdf},
  url = 	 {https://proceedings.mlr.press/v162/zu22a.html},
}