class theanets.losses.Hinge(target, weight=1.0, weighted=False, output_name='out')

Hinge loss function for classifiers.


The hinge loss as implemented here computes the maximum difference between the prediction \(q(x=k)\) for a class \(k\) and the prediction \(q(x=t)\) for the correct class \(t\):

\[\mathcal{L}(x, t) = \max(0, \max_k q(x=k) - q(x=t))\]

This loss is zero whenever the prediction for the correct class is the largest over classes, and increases linearly when the prediction for an incorrect class is the largest.

__init__(target, weight=1.0, weighted=False, output_name='out')


__init__(target[, weight, weighted, output_name])
accuracy(outputs) Build a Theano expression for computing the accuracy of graph output.
log() Log some diagnostic info about this loss.


variables A list of Theano variables used in this loss.