Global

Methods

dummy_minimize(fnc, dimensions, n_callsopt) → {RandomOptimizer}

Minimize a function using a random algorithm. While naive, such approach is often surprisingly competitive for hyperparameter tuning purposes. Internally uses RandomOptimizer class to perform optimization.
Parameters:
Name Type Attributes Default Description
fnc function Function to be minimized.
dimensions Array An array of dimensions, that describe a search space for minimization, or an instance of Space object.
n_calls Number <optional>
 64 Function evaluation budget. The function will be evaluated for at most this number of times.
Source:
Returns:
The optimizer instance, that contains information about found minimum and explored arguments.
Type
RandomOptimizer

minimize_GradientDescent(fnc, grd, x0) → {Object}

Minimize an unconstrained function using first order gradient descent algorithm.
Parameters:
Name Type Description
fnc function Function to be minimized. This function takes array of size N as an input, and returns a scalar value as output, which is to be minimized.
grd function A gradient function of the objective.
x0 Array An array of values of size N, which is an initialization to the minimization algorithm.
Source:
Returns:
An object instance with two fields: argument, which denotes the best argument found thus far, and fncvalue, which is a value of the function at the best found argument.
Type
Object

minimize_L_BFGS(fnc, grd, x0) → {Object}

Minimize an unconstrained function using first order L-BFGS algorithm.
Parameters:
Name Type Description
fnc function Function to be minimized. This function takes array of size N as an input, and returns a scalar value as output, which is to be minimized.
grd function A gradient function of the objective.
x0 Array An array of values of size N, which is an initialization to the minimization algorithm.
Source:
Returns:
An object instance with two fields: argument, which denotes the best argument found thus far, and fncvalue, which is a value of the function at the best found argument.
Type
Object

minimize_Powell(fnc, x0) → {Object}

Minimize an unconstrained function using zero order Powell algorithm.
Parameters:
Name Type Description
fnc function Function to be minimized. This function takes array of size N as an input, and returns a scalar value as output, which is to be minimized.
x0 Array An array of values of size N, which is an initialization to the minimization algorithm.
Source:
Returns:
An object instance with two fields: argument, which denotes the best argument found thus far, and fncvalue, which is a value of the function at the best found argument.
Type
Object

rs_minimize(fnc, dimensions, n_callsopt, n_random_starts, mutation_rate) → {OMGOptimizer}

Minimize a function using a random algorithm. While naive, such approach is often surprisingly competitive for hyperparameter tuning purposes. Internally uses RandomOptimizer class to perform optimization.
Parameters:
Name Type Attributes Default Description
fnc function Function to be minimized.
dimensions Array An array of dimensions, that describe a search space for minimization, or an instance of Space object.
n_calls Number <optional>
 64 Function evaluation budget. The function will be evaluated for at most this number of times.
n_random_starts Integer Determines how many points wil be generated initially at random. The points are not generated at random after this number of evaluations has been reported to the optimizer.
mutation_rate Number A value in the range of (0.0, 1.0]
Source:
Returns:
The optimizer instance, that contains information about found minimum and explored arguments.
Type
OMGOptimizer

to_space(space_object) → {Space}

A convenience function for conversion of Array of dimensions into a single Space instance.
Parameters:
Name Type Description
space_object Object Either an array of dimension objects, or a Space instance.
Source:
Returns:
an instance of Space created out of the provided objects.
Type
Space