DenMune with a Custom Dimensionality Reducer (UMAP)

DenMune’s dim_reducer parameter is flexible, allowing you to pass not just pre-defined strings (‘tsne’, ‘pca’) but also any scikit-learn compatible estimator instance.

This example shows how to use UMAP (umap-learn) as the dimensionality reducer. This is particularly useful as UMAP is often faster than t-SNE and can be better at preserving global data structure.

Note: To run this example, you must have the umap-learn library installed. pip install umap-learn

import matplotlib.pyplot as plt
from sklearn.datasets import make_blobs

# Try to import UMAP. If it fails, skip the example.
try:
    from umap import UMAP
except ImportError:
    print("UMAP not found. Skipping this example.")
    # sphinx-gallery will not run the rest of the script if it exits with code 0
    import sys

    sys.exit(0)

from denmune_skl import DenMune

Generate high-dimensional data

We create a high-dimensional dataset that requires reduction to be clustered effectively by DenMune’s 2D-focused approach.

X_high_dim, y = make_blobs(
    n_samples=500, n_features=50, centers=5, cluster_std=2.5, random_state=42
)

Initialize DenMune with a UMAP instance

We create an instance of UMAP and pass it directly to the dim_reducer parameter of DenMune. DenMune will then use this instance for its dimensionality reduction step.

umap_reducer = UMAP(n_components=2, n_neighbors=15, min_dist=0.1, random_state=42)

model = DenMune(
    k_nearest=30,
    reduce_dims=True,
    target_dims=2,  # This will be ignored, but is a required parameter
    dim_reducer=umap_reducer,
    random_state=42,
)

labels = model.fit_predict(X_high_dim)

Visualize the results on the projected data

The clustering is performed on the 2D data projected by UMAP. We can access this projected data via the projected_X_ attribute.

X_projected = model.projected_X_

plt.figure(figsize=(10, 8))

# Plot projected data colored by DenMune labels
plt.subplot(1, 2, 1)
plt.scatter(X_projected[:, 0], X_projected[:, 1], c=labels, s=20, cmap="viridis")
plt.title(f"DenMune Labels (k={model.k_nearest})\n(Projected by UMAP)")
plt.xlabel("UMAP Component 1")
plt.ylabel("UMAP Component 2")

# Plot projected data colored by true labels for comparison
plt.subplot(1, 2, 2)
plt.scatter(X_projected[:, 0], X_projected[:, 1], c=y, s=20, cmap="viridis")
plt.title("True Labels\n(Projected by UMAP)")
plt.xlabel("UMAP Component 1")

plt.suptitle("DenMune with Custom UMAP Reducer")
plt.tight_layout(rect=[0, 0.03, 1, 0.95])
plt.show()