optunity.solvers.Sobol module

class optunity.solvers.Sobol.Sobol(num_evals, seed=None, skip=None, **kwargs)[source]

Bases: optunity.solvers.util.Solver

Please refer to Sobol sequences for details on this algorithm.

Initializes a Sobol sequence solver.

Parameters:
  • num_evals (int) – number of evaluations to use
  • skip (int or None) – the number of initial elements of the sequence to skip, if None a random skip is generated
  • kwargs ({'name': [lb, ub], ..}) – box constraints for each hyperparameter

The search space is rescaled to the unit hypercube before the solving process begins.

static bitwise_xor(a, b)[source]

Returns the bitwise_xor of a and b as a bitstring.

Parameters:
  • a (int) – first number
  • b (int) – second number
 
>>> Sobol.bitwise_xor(13, 17)
28
>>> Sobol.bitwise_xor(31, 5)
26
bounds
desc_brief = 'sample the search space using a Sobol sequence'
desc_full = ['Generates a Sobol sequence of points to sample in the search space.', '', 'A Sobol sequence is a low discrepancy quasirandom sequence,', 'specifically designed to cover the search space evenly.', '', 'Details are available at: http://en.wikipedia.org/wiki/Sobol_sequence', 'This sampling method should be preferred over sampling uniformly at random.']
static i4_bit_hi1(n)[source]

Returns the position of the high 1 bit base 2 in an integer.

Parameters:n (int) – the integer to be measured
Returns:(int) the number of bits base 2

This was taken from http://people.sc.fsu.edu/~jburkardt/py_src/sobol/sobol.html Licensing:

This code is distributed under the MIT license.
Modified:
22 February 2011
Author:
Original MATLAB version by John Burkardt. PYTHON version by Corrado Chisari
static i4_bit_lo0(n)[source]

Returns the position of the low 0 bit base 2 in an integer.

Parameters:n (int) – the integer to be measured
Returns:(int) the number of bits base 2

This was taken from http://people.sc.fsu.edu/~jburkardt/py_src/sobol/sobol.html Licensing:

This code is distributed under the MIT license.
Modified:
22 February 2011
Author:
Original MATLAB version by John Burkardt. PYTHON version by Corrado Chisari
static i4_sobol(dim_num, seed)[source]

Generates a new quasi-random Sobol vector with each call.

Parameters:
  • dim_num (int) – number of dimensions of the Sobol vector
  • seed (int) – the seed to use to generate the Sobol vector
Returns:

the next quasirandom vector and the next seed to use

This was taken from http://people.sc.fsu.edu/~jburkardt/py_src/sobol/sobol.html Licensing:

This code is distributed under the MIT license.
Modified:
22 February 2011
Author:
Original MATLAB version by John Burkardt. PYTHON version by Corrado Chisari
static i4_sobol_generate(m, n, skip)[source]

Generates a Sobol sequence.

Parameters:
  • m (int) – the number of dimensions (our implementation supports up to 40)
  • n (int) – the length of the sequence to generate
  • skip (int) – the number of initial elements in the sequence to skip
Returns:

a list of length n containing m-dimensional points of the Sobol sequence
static maxdim()[source]

The maximum dimensionality that we currently support.

maximize(f, pmap=<built-in function map>)

Maximizes f.

Parameters:
  • f (callable) – the objective function
  • pmap (callable) – the map() function to use
Returns:

  • the arguments which optimize f
  • an optional solver report, can be None
minimize(f, pmap=<built-in function map>)

Minimizes f.

Parameters:
  • f (callable) – the objective function
  • pmap (callable) – the map() function to use
Returns:

  • the arguments which optimize f
  • an optional solver report, can be None
name = 'sobol'
num_evals
optimize(f, maximize=True, pmap=<built-in function map>)[source]

Optimizes f.

Parameters:
  • f (callable) – the objective function
  • maximize (boolean) – do we want to maximizes?
  • pmap (callable) – the map() function to use
Returns:

  • the arguments which optimize f
  • an optional solver report, can be None
skip
static suggest_from_box(num_evals, **kwargs)[source]

Create a configuration for a Sobol solver.

Parameters:
  • num_evals (int) – number of permitted function evaluations
  • kwargs ({'param': [lb, ub], ..}) – box constraints

Verify that we can effectively make a solver from box.

>>> s = Sobol.suggest_from_box(30, x=[0, 1], y=[-1, 0], z=[-1, 1])
>>> solver = Sobol(**s)