Project is not in working condition
A go implementation of HDBSCAN based on the original paper and this excellent writeup by the author of the scikit-learn module for HDBSCAN.
The executable takes the following parameters:
Usage of ./hdbscan:
-algorithm string
Which algorithm to use. Choices are best, generic, prims_kdtree, prims_balltree, boruvka_kdtree, boruvka_balltree (default "best")
-alpha float
Alpha value (default 1)
-cluster_selection_method string
Method to select clusters from the condensed tree (default "eom")
-data_path string
Data CSV file path (default "data.csv")
-gen_min_span_tree
Wheter to generate the minimum spanning tree with regard to mutual reachability
-leaf_size int
Number of leaf node in the tree (when using a tree-based method) (default 40)
-metric string
Distance metric used for the clustering (default "euclidean")
-min_cluster_size int
Minimal cluster size (default 5)
-save_path string
Data CSV save path (default "save.csv")
As you might have figured out, these are directly taken from the Python implementation. For clarification on their respective usage, please see the module's documentation.
It is worth noting that most of these parameters are useless at the moment and are subject to change.
git clone git@github.com:Belval/hdbscan.gitgo build -o hdbscan -i ./src./hdbscan -h
This is a non-optimal implementation in its current form. The algorithm used to build the MST is O(n^2) and I am fairly sure some computations are wasted here and there.
Notably, the implementation uses float64 instead of the smaller float32. This is mostly due to golang's "default" being float64 and not because the precision is needed.
Ideally I would like to turn this into an actual golang module instead of a CLI executable in the future.