Prior Distributions

class oktopus.prior.Prior

A base class for a prior distribution. Differently from Likelihood, a prior is a PDF that depends solely on the parameters, not on the observed data.

Attributes

name	A name associated with the prior

Methods

__call__(params)	Calls evaluate()
evaluate(params)	Evaluates the negative of the log of the PDF at params
fit([optimizer])	Minimizes the evaluate() function using scipy.optimize.minimize(), scipy.optimize.differential_evolution(), scipy.optimize.basinhopping(), or skopt.gp.gp_minimize().
gradient(params)	Returns the gradient of the loss function evaluated at params

evaluate(params)

Evaluates the negative of the log of the PDF at params

Parameters:

params : scalar or array-like

Value at which the PDF will be evaluated

Returns:

value : scalar

Value of the negative of the log of the PDF at params

name

A name associated with the prior

class oktopus.prior.JointPrior(*args)

Combine indepedent priors by summing the negative of the log of their distributions.

Notes

*args are stored in self.components.

Examples

>>> from oktopus import UniformPrior, GaussianPrior, JointPrior
>>> jp = JointPrior(UniformPrior(-0.5, 0.5), GaussianPrior(0., 1.))
>>> jp.evaluate((0., 0.))
0.0
>>> jp((0., 0.)) # jp is also a callable to evaluate()
0.0

Attributes

*args

(tuple of instances of Prior) Instances of Prior to be combined

Methods

__call__(params)	Calls evaluate()
evaluate(params)	Computes the sum of the log of each distribution given in args evaluated at params.
fit([optimizer])	Minimizes the evaluate() function using scipy.optimize.minimize(), scipy.optimize.differential_evolution(), scipy.optimize.basinhopping(), or skopt.gp.gp_minimize().
gradient(params)	Computes the gradient of the sum of the log of each distribution given in args evaluated at params.

evaluate(params)

Computes the sum of the log of each distribution given in args evaluated at params.

Parameters:

params : tuple

Value at which the JointPrior instance will be evaluated. This must have the same dimension as the number of priors used to initialize the object

Returns:

value : scalar

Sum of the negative of the log of each distribution given in args

gradient(params)

Computes the gradient of the sum of the log of each distribution given in args evaluated at params.

Parameters:

params : tuple

Value at which the JointPrior instance will be evaluated. This must have the same dimension as the number of priors used to initialize the object

Returns:

value : scalar

Gradient of the sum of the negative of the log of each distribution given in args

mean

class oktopus.prior.UniformPrior(lb, ub, name=None)

Computes the negative log pdf for a n-dimensional independent uniform distribution.

Examples

>>> from oktopus import UniformPrior
>>> unif = UniformPrior(0., 1.)
>>> unif(.5)
-0.0
>>> unif(1)
inf

Attributes

lb	(int or array-like of ints) Lower bounds (inclusive)
ub	(int or array-like of ints) Upper bounds (exclusive)

Methods

__call__(params)	Calls evaluate()
evaluate(params)
fit([optimizer])	Minimizes the evaluate() function using scipy.optimize.minimize(), scipy.optimize.differential_evolution(), scipy.optimize.basinhopping(), or skopt.gp.gp_minimize().
gradient(params)

evaluate(params)

gradient(params)

mean

Returns the mean of the uniform distributions

variance

Returns the variance of the uniform distributions

class oktopus.prior.GaussianPrior(mean, var, name=None)

Computes the negative log pdf for a n-dimensional independent Gaussian distribution.

Examples

>>> from oktopus import GaussianPrior
>>> prior = GaussianPrior(0, 1)
>>> prior(2.)
2.0

Attributes

mean	(scalar or array-like) Mean
var	(scalar or array-like) Variance

Methods

__call__(params)	Calls evaluate()
evaluate(params)
fit([optimizer])	Minimizes the evaluate() function using scipy.optimize.minimize(), scipy.optimize.differential_evolution(), scipy.optimize.basinhopping(), or skopt.gp.gp_minimize().
gradient(params)

evaluate(params)

gradient(params)

mean

variance

class oktopus.prior.LaplacianPrior(mean, var, name=None)

Computes the negative log pdf for a n-dimensional independent Laplacian random variable.

Examples

>>> from oktopus import LaplacianPrior
>>> prior = LaplacianPrior(0, 2)
>>> prior(1.)
1.0

Attributes

mean	(scalar or array-like) Mean
var	(scalar or array-like) Variance

Methods

__call__(params)	Calls evaluate()
evaluate(params)
fit([optimizer])	Minimizes the evaluate() function using scipy.optimize.minimize(), scipy.optimize.differential_evolution(), scipy.optimize.basinhopping(), or skopt.gp.gp_minimize().
gradient(params)	Returns the gradient of the loss function evaluated at params

evaluate(params)

mean

variance

Inheritance Diagram