Linear baseline
Build a fast linear baseline for supervised classification.
Scikit-Learn · Class
Perceptron
Perceptron is a linear classifier backed by SGDClassifier. It uses perceptron loss with a constant learning rate and supports optional regularization, early stopping, and incremental fitting.
How it works
Perceptron is a linear classifier implemented through SGDClassifier with perceptron loss and a constant learning rate. It learns coefficients and, when enabled, an intercept by repeatedly passing over training data until the iteration limit or stopping criterion is reached.
Syntax
class sklearn.linear_model.Perceptron(*, penalty=None, alpha=0.0001, l1_ratio=0.15, fit_intercept=True, max_iter=1000, tol=0.001, shuffle=True, verbose=0, eta0=1.0, n_jobs=None, random_state=0, early_stopping=False, validation_fraction=0.1, n_iter_no_change=5, class_weight=None, warm_start=False)Parameters
| Name | Type | Description |
|---|---|---|
penalty | str or None, default=None | Regularization choice: None, l2, l1, or elasticnet. |
alpha | float, default=0.0001 | Multiplier for the regularization term. |
l1_ratio | float, default=0.15 | Elastic-net mixing value between 0 and 1. |
fit_intercept | bool, default=True | Whether to estimate an intercept. |
max_iter | int, default=1000 | Maximum passes over the training data for fit. |
tol | float or None, default=0.001 | Stopping threshold based on loss improvement. |
shuffle | bool, default=True | Whether to shuffle training data each epoch. |
verbose | int, default=0 | Verbosity level. |
eta0 | float, default=1.0 | Constant learning-rate multiplier. |
n_jobs | int or None, default=None | Number of CPU cores used for prediction, where applicable. |
random_state | int, RandomState instance, or None, default=0 | Controls pseudo-randomness. |
early_stopping | bool, default=False | Whether to use an internal validation split for early stopping. |
Returns
A fitted Perceptron estimator from fit; predictions or scores from its estimator methods.
Key points
- Perceptron is a linear classifier for classification tasks.
- Its implementation delegates to SGDClassifier.
- The class fixes perceptron loss and constant learning rate.
- max_iter specifies the maximum number of training-data passes.
- tol provides an optional stopping criterion.
- penalty selects the regularization method.
Common mistakes
- max_iter does not limit iterations performed by partial_fit.
- Disabling fit_intercept requires already centered input data.
- tol controls stopping based on insufficient loss improvement, not a fixed accuracy target.
Example: Train and predict with Perceptron
from sklearn.datasets import load_iris
from sklearn.linear_model import Perceptron
X, y = load_iris(return_X_y=True)
model = Perceptron(random_state=0)
model.fit(X, y)
print(model.predict(X[:3]))This trains a perceptron classifier on the Iris data and predicts the first three samples.
When to use it
Incremental learning
Train on data that arrives in batches.
Large feature spaces
Classify sparse, high-dimensional feature vectors.
Interview questions
QWhat is Perceptron?
Perceptron is a linear classification algorithm. Its implementation fixes SGDClassifier to perceptron loss and a constant learning rate.
Follow-up: How can it be expressed with SGDClassifier? Use SGDClassifier(loss="perceptron", eta0=1, learning_rate="constant", penalty=None).
QHow does Perceptron stop training?
max_iter is the maximum number of passes over the training data, also called epochs, when using fit. tol can stop training when the loss improvement is too small.
Follow-up: Does max_iter constrain partial_fit? No; the documentation says it affects fit, not partial_fit.
QHow are regularization settings configured?
The estimator supports regularization through penalty, with options such as l2, l1, and elasticnet. alpha scales the regularization term, while l1_ratio controls the L1/L2 mixture for elastic net.
Follow-up: What do the extreme l1_ratio values mean? Zero corresponds to L2 and one corresponds to L1.
Related
SGDClassifierLogisticRegressionRidgeClassifierPassiveAggressiveClassifier