CosmoFit.ChainFile¶

class CosmoFit.ChainFile(path, name='mcmc')[source]¶

Bases: object

An HDF5 file holding one MCMC chain, plus the CosmoFit metadata describing it.

Wraps emcee.backends.HDFBackend – which owns the chain itself and needs no help – with the two things a chain has to carry to be re-usable months later: what posterior it came from, and how to check that against the fit in front of you.

Normally built for you by Fitter.run_mcmc(save="..."). Construct one directly to put several chains in one file (name=), or to inspect a file without a fitter.

Parameters:

path (str or Path) – The .h5 file. Parent directories are created on write.
name (str) – HDF5 group holding this chain, so one file can hold several (e.g. one per model in a comparison). Defaults to emcee’s own "mcmc".

Examples

>>> chain = ChainFile("chains/models.h5", name="CPL")
>>> chain.exists, chain.iteration
(True, 6000)

__init__(path, name='mcmc')[source]¶

Parameters:

path (str | PathLike)
name (str)

Methods

`__init__`(path[, name])
`backend`([read_only])	An `emcee.backends.HDFBackend` for this chain -- what gets handed to `emcee.EnsembleSampler(backend=...)`, and what every read below goes through.
`info`()	A short, printable description of the file: what's in it and how far it got.
`open`()	Read-only access to the stored chain, as a stand-in for the `emcee.EnsembleSampler` that produced it -- see `StoredSampler`.
`reset`(nwalkers, ndim)	Throw away this chain's stored steps and start it empty at `(nwalkers, ndim)`.
`write_metadata`(meta)	Store `meta` (JSON-encoded) next to the chain, replacing whatever was there.

Attributes

`exists`	Whether the file exists and already holds this chain's group.
`iteration`	Number of steps already stored (0 if the chain doesn't exist yet).
`metadata`	The CosmoFit metadata stored alongside the chain (an empty dict for a chain written by plain emcee, or one that doesn't exist yet).
`shape`	`(nwalkers, ndim)` of the stored chain, or None if there isn't one.