optimal_control_gradient_descent {seeds} | R Documentation |
estimating the optimal control using the dynamic elastic net
Description
estimating the optimal control using the dynamic elastic net
Usage
optimal_control_gradient_descent(
alphaStep,
armijoBeta,
x0,
parameters,
alpha1,
alpha2,
measData,
constStr,
SD,
modelFunc,
measFunc,
modelInput,
optW,
origAUC,
maxIteration,
plotEsti,
conjGrad,
eps,
nnStates,
verbose
)
Arguments
alphaStep |
starting value of the stepsize for the gradient descent, will be calculate to minimize the cost function by backtracking algorithm |
armijoBeta |
scaling of the alphaStep to find a approximately optimal value for the stepsize |
x0 |
initial state of the ode system |
parameters |
parameters of the ODE-system |
alpha1 |
L1 cost term scalar |
alpha2 |
L2 cost term scalar |
measData |
measured values of the experiment |
constStr |
a string that represents constrains, can be used to calculate a hidden input for a component that gradient is zero |
SD |
standard deviation of the experiment; leave empty if unknown; matrix should contain the timesteps in the first column |
modelFunc |
function that describes the ODE-system of the model |
measFunc |
function that maps the states to the outputs |
modelInput |
an dataset that describes the external input of the system |
optW |
vector that indicated at which knots of the network the algorithm should estimate the hidden inputs |
origAUC |
AUCs of the first optimization; only used by the algorithm |
maxIteration |
a upper bound for the maximal number of iterations |
plotEsti |
boolean that controls of the current estimates should be plotted |
conjGrad |
boolean that indicates the usage of conjugate gradient method over the normal steepest descent |
eps |
citeria for stopping the algorithm |
nnStates |
a bit vector indicating the states that should be non negative |
verbose |
Boolean indicating if an output in the console should be created to display the gradient descent steps |
Value
A list containing the estimated hidden inputs, the AUCs, the estimated states and resulting measurements and the cost function