Density-weighted Uncertainty Sampling

Note

The generated animation can be found at the bottom of the page.

Google Colab Note: If the notebook fails to run after installing the needed packages, try to restart the runtime (Ctrl + M) under Runtime -> Restart session.

Notebook Dependencies

Uncomment the following cell to install all dependencies for this tutorial.

# !pip install scikit-activeml

---

import numpy as np
from matplotlib import pyplot as plt, animation
from sklearn.datasets import make_blobs

from skactiveml.utils import MISSING_LABEL, labeled_indices, unlabeled_indices
from skactiveml.visualization import plot_utilities, plot_decision_boundary

from sklearn.linear_model import LogisticRegression
from sklearn.mixture import GaussianMixture
from skactiveml.classifier import SklearnClassifier
from skactiveml.pool import UncertaintySampling

random_state = np.random.RandomState(0)

# Build a dataset.
X, y_true = make_blobs(n_samples=200, n_features=2,
                       centers=[[0, 1], [-3, .5], [-1, -1], [2, 1], [1, -.5]],
                       cluster_std=.7, random_state=random_state)
y_true = y_true % 2
y = np.full(shape=y_true.shape, fill_value=MISSING_LABEL)

# Initialise the classifier.
clf = SklearnClassifier(LogisticRegression(), classes=np.unique(y_true))
# Initialise the query strategy.
qs = UncertaintySampling(method='least_confident', random_state=random_state)
gmm = GaussianMixture(init_params='kmeans', n_components=5)
gmm.fit(X)
density = np.exp(gmm.score_samples(X))

# Preparation for plotting.
fig, ax = plt.subplots()
feature_bound = [[min(X[:, 0]), min(X[:, 1])], [max(X[:, 0]), max(X[:, 1])]]
artists = []

# The active learning cycle:
n_cycles = 20
for c in range(n_cycles):
    # Fit the classifier.
    clf.fit(X, y)

    # Get labeled instances.
    X_labeled = X[labeled_indices(y)]

    # Query the next instance/s.
    query_idx = qs.query(X=X, y=y, clf=clf, utility_weight=density)

    # Plot the labeled data.
    coll_old = list(ax.collections)
    title = ax.text(
        0.5, 1.05, f"Decision boundary after acquring {c} labels",
        size=plt.rcParams["axes.titlesize"], ha="center",
        transform=ax.transAxes
    )
    ax = plot_utilities(qs, X=X, y=y, clf=clf, utility_weight=density,
                        candidates=unlabeled_indices(y), replace_nan=None, res=25,
                        feature_bound=feature_bound, ax=ax)
    ax.scatter(X[:, 0], X[:, 1], c=y_true, cmap="coolwarm", marker=".",
               zorder=2)
    ax.scatter(X_labeled[:, 0], X_labeled[:, 1], c="grey", alpha=.8,
               marker=".", s=300)
    ax = plot_decision_boundary(clf, feature_bound, ax=ax)

    coll_new = list(ax.collections)
    coll_new.append(title)
    artists.append([x for x in coll_new if (x not in coll_old)])

    # Label the queried instances.
    y[query_idx] = y_true[query_idx]

ani = animation.ArtistAnimation(fig, artists, interval=1000, blit=True)

Once Loop Reflect

References:

The implementation of this strategy is based on Donmez et al.1 and Nguyen and Smeulders2.

1

Pinar Donmez, Jaime G Carbonell, and Paul N Bennett. Dual strategy active learning. In European Conference on Machine Learning, 116–127. Springer, 2007.

2

Hieu T Nguyen and Arnold Smeulders. Active learning using pre-clustering. In International Conference on Machine Learning, 79. 2004.