Building a differentiable unsupervised tree: DOUGLAS

This example shows how to use the Douglas tree for a dataset with few features.

The DOUGLAS model builds a differentiable tree by associating different constructed binnings of the data per feature to clusters. The thresholds are learnt by GEMINI maximisation.

from sklearn import datasets, metrics

from gemclus.tree import Douglas

Load the dataset

iris = datasets.load_iris()
X = iris["data"]
y = iris["target"]

Create the douglas tree and fit it

model = Douglas(n_clusters=3, gemini="mmd_ova", max_iter=100, n_cuts=1)
y_pred_linear = model.fit_predict(X)

print("Score of model is: ", model.score(X))
print("ARI of model is: ", metrics.adjusted_rand_score(y, y_pred_linear))

Score of model is:  1.4741506860502334
ARI of model is:  0.5681159420289855