def parse_component_d2d_trait_args(rptraits, sptraits, wptraits):
if not rptraits:
raise RuntimeError("at least one rptrait is required")
rptraits = iterutils.tuplify(rptraits)
sptraits = iterutils.tuplify(sptraits) if sptraits is not None else ()
wptraits = iterutils.tuplify(wptraits) if wptraits is not None else ()
return dict(rptraits=rptraits, sptraits=sptraits, wptraits=wptraits)
def __init__(self, drivers, dmodels, gmodels):
drivers = iterutils.tuplify(drivers)
dmodels = iterutils.tuplify(dmodels)
gmodels = iterutils.tuplify(gmodels)
ndrivers = len(drivers)
ndmodels = len(dmodels)
ngmodels = len(gmodels)
if not (ndrivers == ndmodels == ngmodels):
raise RuntimeError(f"the number of "
f"drivers ({ndrivers}), "
f"dmodels ({ndmodels}), and "
f"gmodels ({ngmodels}) must be equal")
self._drivers = drivers
self._dmodels = dmodels
self._gmodels = gmodels
self._items = tuple([(drivers[i], dmodels[i], gmodels[i])
for i in range(self.nitems())])
# Preallocate some data structures
# It will make our life easier later on
self._h_model_data = [{
key: dict(d=None, m=None, w=None)
for key in dmodel.onames()
} for dmodel in self.dmodels()]
self._d_model_data = copy.deepcopy(self._h_model_data)
# Merge the pdescs of all models into a dict and ensure they
# have unique keys
self._pdescs, self._mappings = miscutils.merge_dicts_and_make_mappings(
[gmodel.params() for gmodel in self.gmodels()], 'model')
self._times_mdl_eval = []
self._times_mdl_d2h = []
self._time_samples = collections.defaultdict(list)
def parse_component_s3d_trait_args(rptraits, rhtraits, vptraits, vhtraits,
dptraits, dhtraits, zptraits, sptraits,
wptraits):
if not rptraits:
raise RuntimeError("at least one rptrait is required")
if not rhtraits:
raise RuntimeError("at least one rhtrait is required")
if not vptraits:
raise RuntimeError("at least one vptrait is required")
if not dptraits:
raise RuntimeError("at least one dptrait is required")
rptraits = iterutils.tuplify(rptraits)
rhtraits = iterutils.tuplify(rhtraits)
vptraits = iterutils.tuplify(vptraits)
vhtraits = iterutils.tuplify(vhtraits) \
if vhtraits else tuple([None] * len(vptraits))
dptraits = iterutils.tuplify(dptraits)
dhtraits = iterutils.tuplify(dhtraits) \
if dhtraits else tuple([None] * len(dptraits))
zptraits = iterutils.tuplify(zptraits) if zptraits else tuple()
sptraits = iterutils.tuplify(sptraits) if sptraits else tuple()
wptraits = iterutils.tuplify(wptraits) if wptraits else tuple()
if None in vhtraits:
vhtraits = iterutils.replace_items_and_copy(vhtraits, None,
traits.VHTraitOne())
if None in dhtraits:
dhtraits = iterutils.replace_items_and_copy(dhtraits, None,
traits.DHTraitOne())
rptraits_len = len(rptraits)
rhtraits_len = len(rhtraits)
vptraits_len = len(vptraits)
vhtraits_len = len(vhtraits)
dptraits_len = len(dptraits)
dhtraits_len = len(dhtraits)
if rptraits_len != rhtraits_len:
raise RuntimeError(
f"the number of rhtraits must be equal to "
f"the number of rptraits ({rhtraits_len} != {rptraits_len})")
if vptraits_len != vhtraits_len:
raise RuntimeError(
f"the number of vhtraits must be equal to "
f"the number of vptraits ({vhtraits_len} != {vptraits_len})")
if len(dptraits) != len(dhtraits):
raise RuntimeError(
f"the number of dhtraits must be equal to "
f"the number of dptraits ({dhtraits_len} != {dptraits_len})")
return dict(rptraits=rptraits,
rhtraits=rhtraits,
vptraits=vptraits,
vhtraits=vhtraits,
dptraits=dptraits,
dhtraits=dhtraits,
zptraits=zptraits,
sptraits=sptraits,
wptraits=wptraits)
def __init__(self, dmodels, gmodels, drivers):
dmodels = iterutils.tuplify(dmodels)
gmodels = iterutils.tuplify(gmodels)
drivers = iterutils.tuplify(drivers)
ndmodels = len(dmodels)
ngmodels = len(gmodels)
ndrivers = len(drivers)
if not (ndmodels == ngmodels == ndrivers):
raise RuntimeError(f"could not create model; the number of "
f"gmodels ({ngmodels}), "
f"dmodels ({ndmodels}), and "
f"drivers ({ndrivers}) must be equal")
models = []
for dmodel, gmodel, driver in zip(dmodels, gmodels, drivers):
models.append(Model(dmodel, gmodel, driver))
def make_model_group_from_cmp(dmodels, gmodels, drivers):
dmodels = iterutils.tuplify(dmodels)
gmodels = iterutils.tuplify(gmodels)
drivers = iterutils.tuplify(drivers)
ndmodels = len(dmodels)
ngmodels = len(gmodels)
ndrivers = len(drivers)
if not (ndmodels == ngmodels == ndrivers):
raise RuntimeError(f"could not create model group; the number of "
f"gmodels ({ngmodels}), "
f"dmodels ({ndmodels}), and "
f"drivers ({ndrivers}) must be equal")
models = []
for dmodel, gmodel, driver in zip(dmodels, gmodels, drivers):
models.append(Model(dmodel, gmodel, driver))
return make_model_group_from_seq(models)
def make_param_symbol(name, indices):
indices = iterutils.tuplify(indices, False)
if not indices:
result = name
elif len(indices) == 1:
result = f'{name}{make_param_symbol_subscript_bindx(indices[0])}'
else:
result = f'{name}{make_param_symbol_subscript_aindx(indices)}'
return result
def __init__(self, components):
self._cmps = iterutils.tuplify(components)
self._size = [None, None]
self._wcube = None
self._dtype = None
self._driver = None
self._backend = None
(self._params,
self._mappings) = _detail.make_gmodel_2d_params(self._cmps)
def __init__(self,
components,
size_z=None,
step_z=None,
tauto=False,
tcomponents=None):
if tcomponents is None:
tcomponents = ()
self._cmps = iterutils.tuplify(components)
self._tcmps = iterutils.tuplify(tcomponents)
self._tauto = tauto
self._size = [None, None, size_z]
self._step = [None, None, step_z]
self._zero = [None, None, None]
self._tcube = None
self._wcube = None
self._dtype = None
self._driver = None
self._backend = None
(self._params, self._mappings,
self._tmappings) = _detail.make_gmodel_3d_params(
self._cmps, self._tcmps, self._tauto)
def parse_component_d3d_trait_args(rptraits, rhtraits, zptraits, sptraits,
wptraits):
if not rptraits:
raise RuntimeError("at least one rptrait is required")
if not rhtraits:
raise RuntimeError("at least one rhtrait is required")
rptraits = iterutils.tuplify(rptraits)
rhtraits = iterutils.tuplify(rhtraits)
zptraits = iterutils.tuplify(zptraits) if zptraits is not None else ()
sptraits = iterutils.tuplify(sptraits) if sptraits is not None else ()
wptraits = iterutils.tuplify(wptraits) if wptraits is not None else ()
rptraits_len = len(rptraits)
rhtraits_len = len(rhtraits)
if rptraits_len != rhtraits_len:
raise RuntimeError(
f"the number of rhtraits must be equal to "
f"the number of rptraits ({rhtraits_len} != {rptraits_len})")
return dict(rptraits=rptraits,
rhtraits=rhtraits,
zptraits=zptraits,
sptraits=sptraits,
wptraits=wptraits)
def __init__(
self, datasets, drivers, dmodels, gmodels,
wd=False, wp=0.0, wu=1.0):
super().__init__(drivers, dmodels, gmodels)
self._datasets = datasets = iterutils.tuplify(datasets)
if len(datasets) != len(dmodels):
raise RuntimeError(
f"the number of datasets and dmodels must be equal "
f"({len(datasets)} != {len(dmodels)})")
n = self.nitems()
self._d_dataset_d_vector = iterutils.make_list(n, None)
self._d_dataset_m_vector = iterutils.make_list(n, None)
self._d_dataset_e_vector = iterutils.make_list(n, None)
self._d_residual_vector = iterutils.make_list(n, None)
self._h_residual_vector = iterutils.make_list(n, None)
self._d_residual_nddata = iterutils.make_list(n, dict())
self._h_residual_nddata = iterutils.make_list(n, dict())
self._s_counts = iterutils.make_list(n, (None, None))
if not iterutils.is_sequence(wp):
wp = iterutils.make_tuple(n, wp)
if not iterutils.is_sequence(wu):
wu = iterutils.make_tuple(n, wu)
self._wd = wd
self._wp = wp
self._wu = wu
self._weights_d = iterutils.make_tuple(n, dict())
self._weights_p = iterutils.make_tuple(n, dict())
self._weights_u = iterutils.make_tuple(n, dict())
if len(wp) != n:
raise RuntimeError(
f"the length of wp and the number of datasets are not equal "
f"({len(wp)} != {n})")
if len(wu) != n:
raise RuntimeError(
f"the length of wu and the number of datasets are not equal "
f"({len(wu)} != {n})")
for i in range(n):
dataset = datasets[i]
dmodel = dmodels[i]
names_dat = tuple(dataset.keys())
names_mdl = tuple(dmodel.onames())
if set(names_dat) != set(names_mdl):
raise RuntimeError(
f"dataset and dmodel are incompatible "
f"for item #{i} "
f"({names_dat} != {names_mdl})")
if dataset.dtype != dmodel.dtype():
raise RuntimeError(
f"dataset and dmodel have incompatible dtypes "
f"for item #{i} "
f"({dataset.dtype} != {dmodel.dtype()})")
if dataset.size() != dmodel.size():
raise RuntimeError(
f"dataset and dmodel have incompatible sizes "
f"for item #{i} "
f"({dataset.size()} != {dmodel.size()})")
if dataset.step() != dmodel.step():
raise RuntimeError(
f"dataset and dmodel have incompatible steps "
f"for item #{i} "
f"({dataset.step()} != {dmodel.step()})")
if dataset.zero() != dmodel.zero():
raise RuntimeError(
f"dataset and dmodel have incompatible zeros "
f"for item #{i} "
f"({dataset.zero()} != {dmodel.zero()})")
for name in names_mdl:
min_ = np.nanmin(dataset[name].data())
max_ = np.nanmax(dataset[name].data())
self._weights_d[i][name] = 1 / (max_ - min_) if wd else 1.0
if isinstance(wp[i], type(None)):
self._weights_p[i][name] = 0.0
elif isinstance(wp[i], numbers.Real):
self._weights_p[i][name] = wp[i]
elif isinstance(wp[i], dict):
self._weights_p[i][name] = wp[i].get(name, 0.0)
if isinstance(wu[i], type(None)):
self._weights_u[i][name] = 1.0
elif isinstance(wu[i], numbers.Real):
self._weights_u[i][name] = wu[i]
elif isinstance(wu[i], dict):
self._weights_u[i][name] = wu[i].get(name, 1.0)
self._backends = iterutils.make_list(n, None)
self._prepared = False
def __init__(self, data, step):
step = iterutils.tuplify(step)
if len(step) == 1:
step = step + (step[0],)
self._data = data.copy()
self._step = step
def __init__(self, models):
self._models = models = iterutils.tuplify(models)
self._pdescs, self._mappings = miscutils.merge_dicts_and_make_mappings(
[model.pdescs() for model in models], 'model')
def make_model_group_from_seq(models):
return ModelGroup(iterutils.tuplify(models))
def load_moves_with_weights(info):
moves = iterutils.tuplify(info, False)
weights = [move.pop('weight', 1) for move in moves]
moves = move_parser.load(moves)
return tuple(zip(moves, weights))
def make_fitter_result(objective,
parameters,
posterior=None,
extra=None,
solutions=None):
# At least one solution or a global posterior is required
if not solutions:
if not posterior:
raise RuntimeError(
"at least one solution or a global posterior "
"is required in order to create a Fitter Result")
# If we have a global posterior but not solutions,
# transform the global posterior to a new solution
solutions = dict(posterior=posterior)
posterior = None
# Ensure solutions are iterable for convenience
solutions = iterutils.tuplify(solutions, False)
# Process the global posterior
if posterior:
posterior = make_fitter_result_posterior(posterior, parameters)
# Make some arrays with exploded param names for later use
enames_all = parameters.enames(True, True, True)
enames_free = parameters.enames(False, False, True)
enames_tied = parameters.enames(False, True, False)
enames_fixed = parameters.enames(True, False, False)
enames_varying = parameters.enames(False, True, True)
sols = []
# For each solution, create a FitterResultSolution
# and try to populate as many of its fields as possible.
for i, s in enumerate(solutions):
sol = FitterResultSolution()
# ...
if 'mode' in s:
eparams_free = dict(zip(enames_free, s['mode']))
eparams_varying = {p: None for p in enames_varying}
parameters.evaluate(eparams_free, eparams_varying, True)
sol.mode = np.array(list(eparams_varying.values()))
# Calculate statistical quantities from posterior
if 'posterior' in s:
posterior = make_fitter_result_posterior(s['posterior'],
parameters)
sol.covar = np.cov(posterior.samples, rowvar=False)
sol.mean = np.mean(posterior.samples, axis=0)
sol.std = np.std(posterior.samples, axis=0)
# If no
sol.mode = posterior.samples[np.argmax(posterior.logprobs), :]
if sol.mode is None:
sol.mode = posterior.samples[np.argmax(posterior.logprobs)]
sol.posterior = posterior
# Otherwise, salvage whatever is available
else:
if 'mean' in s:
sol.mean = np.full(len(enames_varying), np.nan)
for j, param in enumerate(enames_free):
sol.mean[enames_varying.index(param)] = s['mean'][j]
if 'std' in s:
sol.std = np.full(len(enames_varying), np.nan)
for j, param in enumerate(enames_free):
sol.std[enames_varying.index(param)] = s['std'][j]
if 'covar' in s:
pass
# Mode must be provided or recovered somehow
if sol.mode is None:
raise RuntimeError(
f"mode was not provided or could not be recovered "
f"for solution {i}")
# Calculate quantities using the mode
eparams_varying = dict(zip(enames_varying, sol.mode))
eparams_free = {n: eparams_varying[n] for n in enames_free}
print(eparams_free)
params = parameters.evaluate(eparams_free)
dof = 100
sol.model = objective.model_h(params)
sol.residual = objective.residual_nddata_h(params)
sol.wresidual = objective.residual_nddata_h(params)
sol.chisqr = 1.0
sol.rchisqr = sol.chisqr / (dof - len(enames_free))
sol.wchisqr = 1.0
sol.rwchisqr = sol.wchisqr / (dof - len(enames_free))
# ...
sols.append(sol)
# Sort solution by chi-squared
sols = sorted(sols, key=lambda x: x.chisqr)
param_names = dict(all=enames_all,
free=enames_free,
tied=enames_tied,
fixed=enames_fixed,
varying=enames_varying)
return FitterResult(objective.datasets(), parameters, param_names, sols,
posterior, extra)
