niftynet.engine.application_optimiser module
To customise optimisers including
new optimisation methods, learning rate decay schedule,
or customise other optional parameters of the optimiser:
create a newclass.py that has a class NewOptimisor and implement
get_instance().
and set config parameter in config file or from command line
specify –optimiser newclass.NewOptimisor
-
class
Adam
[source]
Bases: object
Adam optimiser with default hyper parameters
-
static
get_instance
(learning_rate)[source]
create an instance of the optimiser
-
class
Adagrad
[source]
Bases: object
Adagrad optimiser with default hyper parameters
-
static
get_instance
(learning_rate)[source]
create an instance of the optimiser
-
class
Momentum
[source]
Bases: object
Momentum optimiser with default hyper parameters
-
static
get_instance
(learning_rate)[source]
create an instance of the optimiser
-
class
NesterovMomentum
[source]
Bases: object
Nesterov Momentum optimiser with default hyper parameters
-
static
get_instance
(learning_rate)[source]
create an instance of the optimiser
-
class
RMSProp
[source]
Bases: object
RMSProp optimiser with default hyper parameters
-
static
get_instance
(learning_rate)[source]
create an instance of the optimiser
-
class
GradientDescent
[source]
Bases: object
Gradient Descent optimiser with default hyper parameters
-
static
get_instance
(learning_rate)[source]
create an instance of the optimiser