ConditionalEmpiricalUniformPrior

class dysmalpy.parameters.ConditionalEmpiricalUniformPrior(f_cond=None, f_bounds=<function _f_boundaries_from_cond>)

Bases: Prior

Object for flat priors, but with boundaries that are conditional on other model parameters. Determined through empirical testing of f_cond, and bounds are then inferred based on iterating f_cond.

Initialize ConditionalEmpiricalUniformPrior instance.

Parameters:

• f_cond (function) – Function f_cond(param, modelset, self.f_bounds) that takes the parameter and model set as input. It must return True/False if the value does / does not satisfy the conditional requirements. If True, then the log prior will be 0., if False, then it will be -np.inf

• f_bounds (function, optional) –

  Function f_bounds(param, modelset, self.f_cond) that takes the parameter and model set as input. It must return a 2-element array with the lower, upper bounds for the parameter, for the given other model parameter settings. These will then be used to uniformly sample the parameter within these bounds.

  Note this will not be perfect, given the other parameters will be perturbed within their priors (and thus some of the sampled value tuples may end up in the bad region), but hopefully the MCMC walkers will still be able to work with this.

  If not set, it will default to a function that iterates to find boundaries based on self.f_cond.

Methods Summary

log_prior(param[, modelset])	Returns the log value of the prior given the parameter value
prior_unit_transform(param, u[, modelset])	Map a uniform random variable drawn from [0.,1.] to the prior of interest
sample_prior(param[, N, modelset])	Returns a random sample of parameter values distributed according to the prior

Methods Documentation

log_prior(param, modelset=None, **kwargs)

Returns the log value of the prior given the parameter value

Parameters:

• param (DysmalParameter) – DysmalParameter object with which the prior is associated

• modelset (ModelSet) – Current ModelSet, of which param is a part

Returns:

lprior – Log prior value. 0 if the parameter value is within the bounds otherwise -np.inf

Return type:

0 or -np.inf

prior_unit_transform(param, u, modelset=None, **kwargs)

Map a uniform random variable drawn from [0.,1.] to the prior of interest

sample_prior(param, N=1, modelset=None, **kwargs)

Returns a random sample of parameter values distributed according to the prior

Parameters:

• param (DysmalParameter) – DysmalParameter object with which the prior is associated

• N (int, optional) – Size of random sample. Default is 1.

Returns:

rsamp – Random sample of parameter values

Return type:

float or array