bsmart.scans.MCMC

Simple MCMC scan

Simple MCMC scan implementing the Metropolis-Hastings algorithm

Launches a seperate scan for each core, with the option to merge the results at the end via “Merge Results”:”True” in “Setup”. Of course, will also run in single-core mode.

For variables we require a range and a variance for the jumps:

"Variables": {
     "lambda": { "RANGE": [0.01,1.0],
                 "VARIANCE": 0.2}
     },

Several possibilities for likelihood functions are provided – see the BSMlikelihood file.

E.g. gaussian, Log likelihood, “USER”, or “OFF” for observables that should not be taken into account for the likelihood calculation (but the user wants to retain them in the summary file for other purposes). These are set by providing e.g.:

"Observables": {
    "mh" : { "SLHA": ["MASS", [25]], "SCALING": "GAUSS",
            "MEAN": 125.0,
            "VARIANCE":3.0
          }

or “SCALING”:”LOG”, or “SCALING”:”OFF”

It is not necessary to specify the scaling explicitly: gaussian is taken by default

The results are summarised in a csv file (or several) in the Results subdirectory of the scan, which contains only the points retained by the scan (along with an ID for each point); the raw points will as usual be retained in the Spectrum_Files directory if so desired.

Information

BSMArt Name: MCMC

Requires:

• numpy

Settings:

• Points: Int

• Valid Points: Int

• Merge Results: Boolean

• Initial Sample: Path to initial sample file

• Sample LogLike Name: Column name for LogLike (if initial sample and relevant)

• Sample NLL Name: Column name for NLL (if initial sample and relevant)

• Sample Likelihood Name: Column name for Likelihood (if initial sample and relevant)

class bsmart.scans.MCMC.MCMC_onecore(ParentScan, CoreNumber)

Bases: object

add_valid_point()

run(all_list, valid_list)

class bsmart.scans.MCMC.NewScan(inputs, log)

Bases: Scan

Scanner class for multicore MCMC Scans

get_next_pointnumber()

get_previous_sample(sample_file)

Routine to load data from a previous sample and return a couple of numpy arrays of the values

get_random_point()

merge_results

For resuming, or using an initial sample, we need just to have a starting point per core.

The initial sample should also be copied to the results

postprocess(Point, observables, data_point, temp_dir, log, lock=None)

return the likelihood; we won’t get this far if the point failed to be generated

propose_point(last_point)

run()