Skip to content

Latest commit

 

History

History
49 lines (31 loc) · 1.91 KB

README.md

File metadata and controls

49 lines (31 loc) · 1.91 KB
Raw
Python Requirements

See requirements.txt

[Optional]: pip install pynndescent to get first neighbours for large data

Usage:

typically you would run:

from finch import FINCH
c, num_clust, req_c = FINCH(data)

You can set options e.g., required number of cluster [optional] or distance etc,

c, num_clust, req_c = FINCH(data, initial_rank=None, req_clust=None, distance='cosine', ensure_early_exit=False, verbose=True)

Input:

  • data: numpy array (feature vectors in rows)
  • [OPTIONAL]
    • req_c: specify required number of cluster
    • distance: One of ['cityblock', 'cosine', 'euclidean', 'l1', 'l2', 'manhattan'] Recommended: 'cosine (default)' and 'euclidean (for 2D toy data)'
    • initial_rank: Nx1 vector of 1-neighbour indices
    • ensure_early_exit: (default: True) if set it may help for Unbalanced or large datasets, ensure purity of merges and helps early exit
    • use_ann_above_samples: Above this data size (number of samples) approximate nearest neighbors will be used to speed up neighbor discovery. set this for data where exact distances are not feasible to compute. [default = 70000]
    • verbos : for printing some output

Output:

  • c: N x P array, each column vector contains cluster labels for each partition P
  • num_clust: shows total number of cluster in each partition P
  • req_c: Labels of required clusters (Nx1). Only set if req_clust is not None.

Example: Cluster the STL-10 data (13000 images of 10 object classes. We provide the used 2048 CNN resnet features. We can load the data from /data/STL_10/data.mat. This has 13000 vectors stored as a matrix of size (13000,2048), each vector is 2048 dimensional.

See below the notebook for an example on clustering the STL-10 data, which depicts the usage of input params as well.