src/ includes source files for clustering and clustering metrics metods
devel/ includes work in progress modules
test/ include jupyter notebooks with various tests
validation/ includes notes about clustering validation
Algorithms included were selected because:
- they were not present at the time in scikit-learn or extras (but perhaps they are now) e. g. Density Peaks or SNN
- self teaching
all clustering algorithms are imported by myclusters.py
all validation ones are imported by mymetrics.py
mdutils includes various helper functions for Molecular Dynamics
trajectories including, e. g. USR distance
Implementation is pure Python/numpy and slow. I did not even bother with unraveling triangular matrices (not often at least).
The following conventions in source and notebook holds:
- X when given or defined, is ALWAYS the feature or coordinate matrix, [npoints x nfeatures]
- D, the distance matrix is always expected to be a nxn square symmetric matrix with pair distances between the data ALL data set elements: 0 1 2 3 0 - d00 d01 ... 1 - d10 d11 ... 2 - ... 3 - ...
- W are the weights of data points ([npoints x 1] or None)
- clusters is always expected to be a list of all elements the data set where the elements are identified with the label of the corresponding CLUSTER (0 to n) or cluster centroids IF AVAILABLE; If centroids are available set(clusters) ALWAYS gives the centroids labels cluster labels are ALWAYS positive integers a label of -1 ALWAYS indentifies noise.