genome.optimizers

Optimizers for model estimation

Genome implements various optimizers for model estimation.

If you would like to add additional optimization, refer to the guide on Extending Genome.

class genome.optimizers.Adadelta(params, learning_rate=1.0, rho=0.95, consider_constants=None)

Bases: genome.optimizers.BaseOpt

update(cost, params, epsilon=1e-06)

class genome.optimizers.Adagrad(params, learning_rate=1.0, consider_constants=None)

Bases: genome.optimizers.BaseOpt

update(cost, params, epsilon=1e-06)

class genome.optimizers.Adam(params, learning_rate=1.0, beta_1=0.9, beta_2=0.999, consider_constants=None)

Bases: genome.optimizers.BaseOpt

update(cost, params, epsilon=1e-06)

class genome.optimizers.BaseOpt(params, learning_rate, consider_constants)

Bases: object

set_consider_constants(consider_constants)

set_learning_rate(lr)

class genome.optimizers.MomentumSGD(params, learning_rate=0.1, momentum=0.9, consider_constants=None, use_nesterov=False)

Bases: genome.optimizers.BaseOpt

set_momentum(momentum)

update(cost, params)

class genome.optimizers.RMSProp(params, learning_rate=0.1, rho=0.9, consider_constants=None)

Bases: genome.optimizers.BaseOpt

update(cost, params, epsilon=1e-06)

class genome.optimizers.SGD(params, learning_rate=0.1, consider_constants=None)

Bases: genome.optimizers.BaseOpt

Stocastic gradient descent - An iterative method for optimizing an objective function

Computes the gradient of the cost function w.r.t. the parameters \(\theta\) for each batch dataset:

\[\theta^{new} = \theta^{old} - \eta\cdot\nabla_{\theta}J(x,y)\]

Parameters:params (list) – A list of TensorSharedVariable model parameters

update(cost, params)