physical_samples_from_chain

ell1fit.eccentricity.physical_samples_from_chain(results_row, flat_chain, parameters, labels=None, strict=True)

Turn a flattened chain into physical samples of the named parameters.

The sampler works in local coordinates: column i of the chain holds dA1, say, the offset from the starting value in units of that parameter’s preconditioned scale. The fit records both numbers needed to undo that, dA1_initial and dA1_factor, so that

\[A1 = A1^{\mathrm{initial}} + \mathrm{d}A1 \times \mathrm{factor}.\]

Parameters:

results_rowastropy.table.Row or dict

One row of a *_results.ecsv table.

flat_chainnp.ndarray

Flattened chain, shape (nsamples, ndim), in local coordinates.

parametersiterable of str

Physical parameter names, e.g. ("A1", "EPS1", "EPS2").

labelslist of str, optional

Parameter name per column, as saved by ell1fit.mcmc_utils.save_flat_samples(). When absent – an HDF5 chain carries no names – the columns are identified by their recorded percentiles instead.

strictbool

Whether a parameter that was not fitted is an error. True, the default, suits a caller that needs a specific parameter and has nothing to say without it. False suits one that offers a menu – the orbital summary plot asks for all five orbital parameters and draws whichever the chain actually explored. Either way, a chain that disagrees with the table about a parameter both contain still raises: that means the two files are from different fits, which is never something to skip.

Returns:

dict

{parameter: np.ndarray}, in the order requested, holding the physical posterior samples. Parameters skipped under strict=False are simply absent.