README.md

metadata

title: SGD Classifier
emoji: 😻
colorFrom: indigo
colorTo: green
sdk: static
app_file: index.html
pinned: false

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

SGD Classifier

SGDClassifier is a NumPy-only linear classifier trained with stochastic gradient descent. It supports hinge, log_loss, modified_huber, squared_hinge, perceptron, squared_error, huber, epsilon_insensitive, and squared_epsilon_insensitive losses.

Choose batch_selection="random" to independently sample each mini-batch, or batch_selection="permutation" to shuffle the dataset and consume all of its non-overlapping mini-batches before reshuffling.

from sgd_classifier import SGDClassifier

classifier = SGDClassifier(
    loss="log_loss",
    batch_size=32,
    batch_selection="permutation",
    max_epochs=20,
    random_state=42,
)
classifier.fit(X_train, y_train)
predictions = classifier.predict(X_test)

For manual or streaming training, one call performs exactly one SGD update and returns the classifier:

classifier = SGDClassifier(loss="hinge", learning_rate=0.01)
classifier.train_step(X_batch, y_batch, classes=[0, 1])

Interactive visualizer

Open index.html through a local web server to create a two-dimensional binary dataset by clicking on the plot. The model runs in the browser through Pyodide; each press of Take one SGD step selects one mini-batch, calls train_step, and redraws the decision regions and boundary.

python -m http.server 8000