DRUHG - Dialectical Reflection Universal Hierarchical Grouping (друг).
Performs clustering based on densities and builds a minimum spanning tree.
Does not require parameters. (The parameter is space metric, e.x. euclidean)
The user can filter the size of the clusters with
size_range, for genuine result and genuine outliers set to [1,1].Parameter
fix_outliers allows to label outliers to their closest clusters via mstree edges.There are some optional tuning parameters but the actual algorithm requires none and is universal.
It works by applying the universal society rule: treat others how you want to be treated.
The core of the algorithm is to rank the subject's closest subjective similarities and amalgamate them accordingly.
Parameter
max_ranking controls precision vs productivity balance, after some value the precision and the result would not change.todo: Parameter
algorithm can be set to 'slow' to further enhance the precision.The dialectical distance reflects the opposite density.
Max( r/R d(r); d(R) ), where r and R are ranks from A to B and from B to A.
This orders outliers last and equal densities first.
It's great replacement for DBSCAN and global outliers detection.
Those ordered connections become trees. Two trees reflect of each other in their totality and can transfrom into cluster.
D N₂ K₁/(K₁+K₂) sum 1/dᵢ > N₁ - 1, where N is size of a tree, K is number of clusters in a tree.
This allows newly formed clusters to resist the reshaping.
import sklearn.datasets as datasets
import druhg
iris = datasets.load_iris()
XX = iris['data']
clusterer = druhg.DRUHG(max_ranking=50)
labels = clusterer.fit(XX).labels_It will build the tree and label the points. Now you can manipulate clusters by relabeling.
labels = dr.relabel(exclude=[7749, 100], size_range==[0.2, 2242], fix_outliers=1)
ari = adjusted_rand_score(iris['target'], labels)
print ('iris ari', ari)- Relabeling is cheap.
-
-
excludebreaks clusters by label number, -
size_rangerestricts cluster size by percent or by absolute number, -
fix_outlierscolors outliers by connectivity.
-
clusterer.plot(labels)It will draw mstree with druhg-edges.
clusterer.plot()It will provide interactive sliders for an exploration.
It can be slow on a highly structural data.
There is a parameters
max_ranking that can be used to decrease for a better performance.
PyPI install, presuming you have an up to date pip:
pip install druhgThe package tests can be run after installation using the command:
pytest -k "test_name"The tests may fail :-D
The druhg library supports Python 3.
We welcome contributions in any form! Assistance with documentation, particularly expanding tutorials, is always welcome. To contribute please fork the project make your changes and submit a pull request. We will do our best to work through any issues with you and get your code merged into the main branch.
The druhg package is 3-clause BSD licensed.
