ml4eft.limits.optimize_ns.Optimize#

class ml4eft.limits.optimize_ns.Optimize(config, coeff=None, rep=None)[source]#

Bases: object

Optimizer to obtain confidence level intervals in the SMEFT using either binned or unbinned (ML level) observables

__init__(config, coeff=None, rep=None)[source]#

Optimize constructor

Parameters
  • config (dict) – loaded json run card

  • coeff (array_like, optional) – Subset of (N,) ndarray EFT parameter names to include in the fit. Takes all EFT parameters by default.

  • rep (int, optional) – Run the optimiser on a specific replica numner, set to None by default, in which case the optimiser works with the median over all replicas.

Methods

__init__(config[, coeff, rep])

Optimize constructor

clean()

Removes raw NS output if no raw output is needed

cube_to_dict(cube)

Converts the prior volume (N, ) ndarray to a dict with the EFT parameter names as keys

log_like_binned(cube)

Implementaiton of the binned Poissonian log-likelihood

log_like_nn(cube)

Implementaiton of the unbinned extended log-likelihood using the ML models

log_like_truth(cube)

Implementaiton of the unbinned extended log-likelihood using the analytical differential cross-sections at the parton level

my_prior(cube)

Construct prior volume

run_sampling()

Runs the minimisation with Nested Sampling

save(result)

Save NS replicas to json inside a dictionary of the form {'coeff': [replicas values]}

clean()[source]#

Removes raw NS output if no raw output is needed

cube_to_dict(cube)[source]#

Converts the prior volume (N, ) ndarray to a dict with the EFT parameter names as keys

Parameters

cube (array_like) – prior volume

log_like_binned(cube)[source]#

Implementaiton of the binned Poissonian log-likelihood

Parameters

cube (array_like) – prior volume

Returns

Binned Poissonian log-likelihood

Return type

float

log_like_nn(cube)[source]#

Implementaiton of the unbinned extended log-likelihood using the ML models

Parameters

cube (array_like) – prior volume

Returns

Extended-log likelihood using the ML models

Return type

float

log_like_truth(cube)[source]#

Implementaiton of the unbinned extended log-likelihood using the analytical differential cross-sections at the parton level

Parameters

cube (array_like) – prior volume

Returns

Extended-log likelihood using the analytical predictions

Return type

float

my_prior(cube)[source]#

Construct prior volume

Parameters

cube (array_like,) – (N, ) ndarray unit hypercube of dim N

Returns

cube – prior volume

Return type

array_like

run_sampling()[source]#

Runs the minimisation with Nested Sampling

save(result)[source]#

Save NS replicas to json inside a dictionary of the form {‘coeff’: [replicas values]}

Parameters

result (dict) – result dictionary