A Python library implementing the Topological Landmark-Isomap algorithm presented in
Yan, Lin & Yaodong Zhao, Paul Rosen, Carlos Scheidegger & Bei Wang. (2018). Homology-Preserving Dimensionality Reduction via Manifold Landmarking and Tearing.
Clone this repo and move the base director wherever you wish. This library is not yet installable via pip or conda so you must import it like so:
import sys
sys.path.append('/path/to/TL-Isomap/src')
import tl_isomap
$ cd TL-Isomap
$ python
>>> import sys
>>> sys.path.append('src')
>>> import numpy as np
>>> import clustering as cl
>>> import filter_functions as ff
>>> import tl_isomap as tli
>>> data = np.load('test_data/spirals.npy')
>>> my_tli = tli.TLIsomap(data, ff.eccentricity_p(data, 2.0), cl.DBSCAN, num_bins = 10, dbscan_eps = 0.5, num_neighbors = 4, isomap_eps = 0.1)
>>> my_tli.run_mapper()
>>> my_tli.graph.edges
. . .
>>> embedding = my_tli.embed_k_dims(2) #an expensive computation
>>> embedding
. . .
The library offers the use of DBSCAN or SLC for clustering; different parameters must be passed to the constructor of the TLIsomap object depending on which clustering algorithm is to be used.