Profilers

Profiler

class desilike.profilers.base.Profiler(likelihood, posterior=True, rng=None, directory=None)

Profiler used to compute likelihood and posterior profiles.

Initialize the profiler.

Parameters:

• likelihood (BaseLikelihood) – Likelihood to profile.

• posterior (bool, optional) – If True, profile the posterior. Otherwise, profile the likelihood. Default is True.

• rng (numpy.random.Generator, int or None, optional) – Random number generator. Default is None.

• directory (str, Path, or None, optional) – Save samples to this folder. Default is None.

add_manual_grid(grid)

Manually add parameter grid to optimize.

Parameters:

grid (dict) – Parameter grid to profile, i.e., dict(a=[0, 1, 2]) implies that the maximum likelihood is found for \(a=0\), \(a=1\), and \(a=2\). If multiple parameters are specified, all combinations are profiled.

Raises:

ValueError – If no parameter is specificed or a parameter is not described in the likelihood.

add_optimize_all()

Add finding the global optimum.

add_single_sample(sample)

Add a parameter combination to optimize.

Parameters:

sample (dict) – Single parameter combination to profile.

Raises:

ValueError – If a parameter is not described in the likelihood.

run(n_per_iter=10, max_iter=10, tol=0.001, warm_start=False, max_init_attempts=100, optimizer=<function scipy_dual_annealing>, optimizer_kwargs=None)

Run the profiler.

Parameters:

• n_per_iter (int, optional) – Independent optimizations per sample at each iteration. Default is 10.

• max_iter (int, optional) – Maximum number of iterations. Default is 10.

• tol (float, optional) – Optimization stops if maximum improvement accross all samples drops below tol between optimizations. Default is 1e-2.

• warm_start (bool, optional) – If True, starting positions are derived from interpolating previous points. This can only be done if the profiler was run with warm_start=False before.

• max_init_attempts (int, optional) – Maximum number of attempts to initialize each sample. Default is 100.

• optimizer (callable, optional) – Optimizer function from desilike.profilers.optimizers. Default is desilike.profilers.scipy_dual_annealing.

• optimizer_kwargs (dict, optional) – Optional keyword arguments passed to the optimizer. Default is None.

Raises:

ValueError – If trying to run the profiler without having added samples.

Returns:

samples – Maxima found by the profiler.

Return type:

desilike.statistics.samples.Samples

Optimizers

Collection of wrappers for commonly used optimizers.

desilike.profilers.optimizers.scipy_dual_annealing(f, x_0, rng, **kwargs)

Optimize using scipy.dual_annealing().

Parameters:

• f (callable) – Function to optimize.

• x_0 (array-like) – Starting point.

• rng (numpy.random.Generator) – Random number generator.

• **kwargs – Additional keyword arguments passed to scipy.dual_annealing.

Returns:

• x_min (numpy.ndarray) – Coordinates of the minimum.

• f_min (float) – Value of the objective function at the minimum.

• success (bool) – Whether the optimizer finished successfully.

desilike.profilers.optimizers.scipy_minimize(f, x_0, rng, **kwargs)

Optimize using scipy.minimize().

Parameters:

• f (callable) – Objective function.

• x_0 (array-like) – Starting point.

• rng (numpy.random.Generator) – Unused. Present for API consistency.

• **kwargs – Additional keyword arguments passed to scipy.minimize.

Returns:

• x_min (numpy.ndarray) – Coordinates of the minimum.

• f_min (float) – Value of the objective function at the minimum.

• success (bool) – Whether the optimizer finished successfully.