tigercontrol.models.optimizers.ONS

class tigercontrol.models.optimizers.ONS(pred=None, loss=<function mse>, learning_rate=1.0, hyperparameters={})

Online newton step algorithm.

__init__(pred=None, loss=<function mse>, learning_rate=1.0, hyperparameters={})

Initialize self. See help(type(self)) for accurate signature.

Methods

__init__([pred, loss, learning_rate, …])	Initialize self.
general_norm(x)
gradient(params, x, y[, loss])	Description: Updates parameters based on correct value, loss and learning rate.
norm_project(y, A, c)	Project y using norm A on the convex set bounded by c.
set_loss(new_loss)	Description: updates internal loss
set_predict(pred[, loss])	Description: Updates internally stored pred and loss functions :param pred: predict function, must take params and x as input :type pred: function :param loss: loss function.
update(params, x, y[, loss])	Description: Updates parameters based on correct value, loss and learning rate.