class pystruct.models.MultiClassClf(n_features=None, n_classes=None, class_weight=None, rescale_C=False)[source]

Formulate linear multiclass SVM in C-S style in CRF framework.

Inputs x are simply feature arrays, labels y are 0 to n_classes.


n_features : int

Number of features of inputs x. If None, it is inferred from data.

n_classes : int, default=None

Number of classes in dataset. If None, it is inferred from data.

class_weight : None, or array-like

Class weights. If an array-like is passed, it must have length n_classes. None means equal class weights.

rescale_C : bool, default=False

Whether the class-weights should be used to rescale C (liblinear-style) or just rescale the loss.


No bias / intercept is learned. It is recommended to add a constant one feature to the data.

It is also highly recommended to use n_jobs=1 in the learner when using this model. Trying to parallelize the trivial inference will slow the infernce down a lot!


batch_inference(X, w[, relaxed])
batch_joint_feature(X, Y[, Y_true])
batch_loss(Y, Y_hat)
batch_loss_augmented_inference(X, Y, w[, ...])
continuous_loss(y, y_hat)
inference(x, w[, relaxed, return_energy]) Inference for x using parameters w.
initialize(X, Y)
joint_feature(x, y[, y_true]) Compute joint feature vector of x and y.
loss(y, y_hat)
loss_augmented_inference(x, y, w[, relaxed, ...]) Loss-augmented inference for x and y using parameters w.
__init__(n_features=None, n_classes=None, class_weight=None, rescale_C=False)[source]
inference(x, w, relaxed=None, return_energy=False)[source]

Inference for x using parameters w.

Finds armin_y, joint_feature(x, y)), i.e. best possible prediction.

For an unstructured multi-class model (this model), this can easily done by enumerating all possible y.


x : ndarray, shape (n_features,)

Input sample features.

w : ndarray, shape=(size_joint_feature,)

Parameters of the SVM.

relaxed : ignored


y_pred : int

Predicted class label.

joint_feature(x, y, y_true=None)[source]

Compute joint feature vector of x and y.

Feature representation joint_feature, such that the energy of the configuration (x, y) and a weight vector w is given by, joint_feature(x, y)).


x : nd-array, shape=(n_features,)

Input sample features.

y : int

Class label. Between 0 and n_classes.

y_true : int

True class label. Needed if rescale_C==True.


p : ndarray, shape (size_joint_feature,)

Feature vector associated with state (x, y).

loss_augmented_inference(x, y, w, relaxed=None, return_energy=False)[source]

Loss-augmented inference for x and y using parameters w.

Minimizes over y_hat:, y_hat), w) + loss(y, y_hat)


x : ndarray, shape (n_features,)

Unary evidence / input to augment.

y : int

Ground truth labeling relative to which the loss will be measured.

w : ndarray, shape (size_joint_feature,)

Weights that will be used for inference.


y_hat : int

Label with highest sum of loss and score.