Simulated Annealing Parameter Search

../../../../_images/annealing_hyperparam.svg

Hyperparameter search by simulated annealing.

Documentation

Uses simulated annealing to find the optimal parameters by considering a hyper cube of all possible indices to the given parameter table. Each column of the parameter table corresponds to one axis of this cube with a range corresponding to the non-masked rows of the parameter table. The radius for the annealing process assumes that all axes have unit length regardless of the number of non-masked rows. Note: This node should be considered _experimental_ and may change in the future.

Definition

Input ports

in-model
Type: model
Description: in-model
parameter space
Type: table
Description: param-space
X
Type: table
Description: X
Y
Type: table
Description: Y
cross-validation
Type: [(table,table)]
Description: cross-validation
Optional number of ports: 0–1 (default: 0)

Output ports

results
Type: table
Description: results
parameters
Type: table
Description: parameters
out-model
Type: model
Description: out-model

Configuration

Cooling method (cooling)

Method for lowering temperature

Cooling argument (cooling_arg)

Argument A to cooling method. Exponential: T=A^t Linear ignores A Logarithmic: T=A/log(1+t)

Cross validation splits (cv)

Number of fold in the default K-Fold cross validation. Ignored when cross-validation port is given

iterations (n_iter)

Number of randomized searches done

Implementation

class node_paramsearch.ParameterSearch_SimulatedAnnealing[source]