def test_jax_jit_enable_stitching(caplog, do_grad, return_fitted_val):
pyhf.set_backend(pyhf.tensor.jax_backend(precision='64b'), 'scipy')
pdf = pyhf.simplemodels.hepdata_like([50.0], [100.0], [10.0])
data = pyhf.tensorlib.astensor([125.0] + pdf.config.auxdata)
with caplog.at_level(logging.DEBUG, 'pyhf.optimize.opt_jax'):
pyhf.infer.mle.fixed_poi_fit(
1.0,
data,
pdf,
do_grad=do_grad,
do_stitch=False,
return_fitted_val=return_fitted_val,
) # jit
assert 'jitting function' in caplog.text
caplog.clear()
with caplog.at_level(logging.DEBUG, 'pyhf.optimize.opt_jax'):
pyhf.infer.mle.fixed_poi_fit(
1.0,
data,
pdf,
do_grad=do_grad,
do_stitch=True,
return_fitted_val=return_fitted_val,
) # jit
assert 'jitting function' in caplog.text
caplog.clear()
def test_interpcode(backend, interpcode, random_histosets_alphasets_pair):
pyhf.set_backend(backend)
if isinstance(pyhf.tensorlib, pyhf.tensor.tensorflow_backend):
tf.reset_default_graph()
pyhf.tensorlib.session = tf.Session()
histogramssets, alphasets = random_histosets_alphasets_pair
# single-float precision backends, calculate using single-floats
if isinstance(backend, pyhf.tensor.tensorflow_backend) or isinstance(
backend, pyhf.tensor.pytorch_backend):
histogramssets = np.asarray(histogramssets, dtype=np.float32)
alphasets = np.asarray(alphasets, dtype=np.float32)
slow_result = np.asarray(
pyhf.tensorlib.tolist(
pyhf.interpolate.interpolator(interpcode,
do_tensorized_calc=False)(
histogramssets=histogramssets,
alphasets=alphasets)))
fast_result = np.asarray(
pyhf.tensorlib.tolist(
pyhf.interpolate.interpolator(interpcode, do_tensorized_calc=True)(
histogramssets=pyhf.tensorlib.astensor(
histogramssets.tolist()),
alphasets=pyhf.tensorlib.astensor(alphasets.tolist()))))
assert pytest.approx(slow_result[~np.isnan(slow_result)].ravel().tolist(
)) == fast_result[~np.isnan(fast_result)].ravel().tolist()
def _fit_model_pyhf(
model: pyhf.pdf.Model,
data: List[float],
init_pars: Optional[List[float]] = None,
fix_pars: Optional[List[bool]] = None,
minos: Optional[Union[List[str], Tuple[str, ...]]] = None,
) -> FitResults:
"""Uses the ``pyhf.infer`` API to perform a maximum likelihood fit.
Parameters set to be fixed in the model are held constant. The ``init_pars``
argument allows to override the ``pyhf`` default initial parameter settings, and the
``fix_pars`` argument overrides which parameters are held constant.
Args:
model (pyhf.pdf.Model): the model to use in the fit
data (List[float]): the data to fit the model to
init_pars (Optional[List[float]], optional): list of initial parameter settings,
defaults to None (use ``pyhf`` suggested inits)
fix_pars (Optional[List[bool]], optional): list of booleans specifying which
parameters are held constant, defaults to None (use ``pyhf`` suggestion)
minos (Optional[Union[List[str], Tuple[str, ...]]], optional): runs the MINOS
algorithm for all parameters specified, defaults to None (does not run
MINOS)
Returns:
FitResults: object storing relevant fit results
"""
pyhf.set_backend("numpy", pyhf.optimize.minuit_optimizer(verbose=1))
result, result_obj = pyhf.infer.mle.fit(
data,
model,
init_pars=init_pars,
fixed_params=fix_pars,
return_uncertainties=True,
return_result_obj=True,
)
log.info(f"MINUIT status:\n{result_obj.minuit.fmin}")
bestfit = result[:, 0]
uncertainty = result[:, 1]
labels = model_utils.get_parameter_names(model)
corr_mat = result_obj.hess_inv.correlation()
best_twice_nll = float(result_obj.fun) # convert 0-dim np.ndarray to float
fit_results = FitResults(bestfit, uncertainty, labels, corr_mat,
best_twice_nll)
if minos is not None:
parameters_translated = []
for minos_par in minos:
par_index = model_utils._get_parameter_index(minos_par, labels)
if par_index != -1:
# pyhf does not hand over parameter names, all parameters are known as
# x0, x1, etc.
parameters_translated.append(f"x{par_index}")
_run_minos(result_obj.minuit, parameters_translated, labels)
return fit_results
def test_interpcode_1(backend, do_tensorized_calc):
pyhf.set_backend(backend)
if isinstance(pyhf.tensorlib, pyhf.tensor.tensorflow_backend):
tf.reset_default_graph()
pyhf.tensorlib.session = tf.Session()
histogramssets = pyhf.tensorlib.astensor([[[[0.9], [1.0], [1.1]]]])
alphasets = pyhf.tensorlib.astensor([[-2, -1, 0, 1, 2]])
expected = pyhf.tensorlib.astensor([[[[0.9**2], [0.9], [1.0], [1.1],
[1.1**2]]]])
if do_tensorized_calc:
result_deltas = pyhf.interpolate.interpolator(
1, do_tensorized_calc=do_tensorized_calc)(histogramssets,
alphasets)
else:
result_deltas = pyhf.tensorlib.astensor(
pyhf.interpolate.interpolator(
1, do_tensorized_calc=do_tensorized_calc)(
pyhf.tensorlib.tolist(histogramssets),
pyhf.tensorlib.tolist(alphasets)))
# calculate the actual change
allsets_allhistos_noms_repeated = pyhf.tensorlib.einsum(
'sa,shb->shab', pyhf.tensorlib.ones(alphasets.shape),
histogramssets[:, :, 1])
results = allsets_allhistos_noms_repeated * result_deltas
assert pytest.approx(
np.asarray(
pyhf.tensorlib.tolist(results)).ravel().tolist()) == np.asarray(
pyhf.tensorlib.tolist(expected)).ravel().tolist()
def main(backend):
"""
Example of fitting a public workspace from HEPData. The fit selects a single
signal point from the signal patchset and then patches the background only
model to create a workspace.
"""
pyhf.set_backend(backend)
print(f"Backend set to: {backend}")
# Using the generator to cleanup automatically
tarfile = [tgz for tgz in electroweakinos_likelihoods_download()][0]
oneLbb_background, oneLbb_Wh_hbb_750_100_signal_patch = get_bkg_and_signal(
tarfile,
"1Lbb-likelihoods-hepdata",
(750, 100), # "C1N2_Wh_hbb_750_100(750, 100)"
)
print("\nStarting fit\n")
fit_start_time = datetime.now()
CLs_obs, CLs_exp = calculate_CLs(oneLbb_background,
oneLbb_Wh_hbb_750_100_signal_patch)
fit_end_time = datetime.now()
fit_time = fit_end_time - fit_start_time
print(f"fit C1N2_Wh_hbb_750_100 in {fit_time} seconds\n")
print(f"CLs_obs: {CLs_obs}")
print(f"CLs_exp: {CLs_exp}")
def test_diffable_backend(jitted):
pyhf.set_backend("jax", default=True)
def example_op(x):
y = pyhf.default_backend.astensor(x)
return 2 * y
if jitted:
assert jax.jacrev(jax.jit(example_op))([1.0]) == [2.0]
else:
assert jax.jacrev(example_op)([1.0]) == [2.0]
def example_op2(x):
y = pyhf.default_backend.power(x, 2)
z = pyhf.tensorlib.sum(y)
return z
if jitted:
assert jax.jacrev(jax.jit(example_op2))(pyhf.tensorlib.astensor(
[2.0, 3.0])).tolist() == [
4.0,
6.0,
]
else:
assert jax.jacrev(example_op2)(pyhf.tensorlib.astensor(
[2.0, 3.0])).tolist() == [
4.0,
6.0,
]
def infer_hypotest(workspace, metadata, patches, backend):
import time
import pyhf
pyhf.set_backend(backend)
tick = time.time()
model = workspace.model(
patches=patches,
modifier_settings={
"normsys": {
"interpcode": "code4"
},
"histosys": {
"interpcode": "code4p"
},
},
)
data = workspace.data(model)
test_poi = 1.0
return {
"metadata":
metadata,
"CLs_obs":
float(pyhf.infer.hypotest(test_poi, data, model, test_stat="qtilde")),
"Fit-Time":
time.time() - tick,
}
def reset_backend():
"""
This fixture is automatically run to reset the backend before and after a test function runs.
"""
pyhf.set_backend(pyhf.default_backend)
yield reset_backend
pyhf.set_backend(pyhf.default_backend)
def test_minuit_failed_optimization(monkeypatch, mocker,
has_reached_call_limit, is_above_max_edm):
class BadMinuit(iminuit.Minuit):
@property
def valid(self):
return False
@property
def fmin(self):
mock = mocker.MagicMock()
mock.has_reached_call_limit = has_reached_call_limit
mock.is_above_max_edm = is_above_max_edm
return mock
monkeypatch.setattr(iminuit, 'Minuit', BadMinuit)
pyhf.set_backend('numpy', 'minuit')
pdf = pyhf.simplemodels.hepdata_like([5], [10], [3.5])
data = [10] + pdf.config.auxdata
spy = mocker.spy(pyhf.optimize.minuit_optimizer, '_minimize')
with pytest.raises(pyhf.exceptions.FailedMinimization) as excinfo:
pyhf.infer.mle.fit(data, pdf)
assert isinstance(excinfo.value.result, OptimizeResult)
assert excinfo.match('Optimization failed')
assert 'Optimization failed' in spy.spy_return.message
if has_reached_call_limit:
assert excinfo.match('Call limit was reached')
assert 'Call limit was reached' in spy.spy_return.message
if is_above_max_edm:
assert excinfo.match('Estimated distance to minimum too large')
assert 'Estimated distance to minimum too large' in spy.spy_return.message
def test_tensorlib_setup(tensorlib, precision, mocker):
tb = getattr(pyhf.tensor, tensorlib)(precision=precision)
func = mocker.patch(f'pyhf.tensor.{tensorlib}._setup')
assert func.call_count == 0
pyhf.set_backend(tb)
assert func.call_count == 1
def test_custom_optimizer_name_supported():
class custom_optimizer(object):
def __init__(self, **kwargs):
self.name = "scipy"
with pytest.raises(AttributeError):
pyhf.set_backend(pyhf.tensorlib, custom_optimizer())
def test_jax_jit_enable_autograd(caplog, do_stitch, return_fitted_val):
pyhf.set_backend("jax", "scipy", precision="64b")
pdf = pyhf.simplemodels.uncorrelated_background([50.0], [100.0], [10.0])
data = pyhf.tensorlib.astensor([125.0] + pdf.config.auxdata)
with caplog.at_level(logging.DEBUG, 'pyhf.optimize.opt_jax'):
pyhf.infer.mle.fixed_poi_fit(
1.0,
data,
pdf,
do_grad=False,
do_stitch=do_stitch,
return_fitted_val=return_fitted_val,
) # jit
assert 'jitting function' in caplog.text
caplog.clear()
with caplog.at_level(logging.DEBUG, 'pyhf.optimize.opt_jax'):
pyhf.infer.mle.fixed_poi_fit(
1.0,
data,
pdf,
do_grad=True,
do_stitch=do_stitch,
return_fitted_val=return_fitted_val,
) # jit
assert 'jitting function' in caplog.text
caplog.clear()
def test_set_backend_by_bytestring(backend_name):
pyhf.set_backend(backend_name)
assert isinstance(
pyhf.tensorlib,
getattr(pyhf.tensor,
"{0:s}_backend".format(backend_name.decode("utf-8"))),
)
def test_set_optimizer_by_bytestring(optimizer_name):
pyhf.set_backend(pyhf.tensorlib, optimizer_name)
assert isinstance(
pyhf.optimizer,
getattr(pyhf.optimize,
"{0:s}_optimizer".format(optimizer_name.decode("utf-8"))),
)
def test_set_precision_by_string_wins(precision_level):
conflicting_precision = "32b" if precision_level == "64b" else "64b"
pyhf.set_backend(
pyhf.tensor.numpy_backend(precision=conflicting_precision),
precision=precision_level,
)
assert pyhf.tensorlib.precision == precision_level.lower()
def significance(model: pyhf.pdf.Model, data: List[float]) -> SignificanceResults:
"""Calculates the discovery significance of a positive signal.
Observed and expected p-values and significances are both calculated and reported.
Args:
model (pyhf.pdf.Model): model to use in fits
data (List[float]): data (including auxdata) the model is fit to
"""
pyhf.set_backend(pyhf.tensorlib, pyhf.optimize.minuit_optimizer(verbose=1))
log.info("calculating discovery significance")
obs_p_val, exp_p_val = pyhf.infer.hypotest(
0.0, data, model, test_stat="q0", return_expected=True
)
obs_p_val = float(obs_p_val)
exp_p_val = float(exp_p_val)
obs_significance = scipy.stats.norm.isf(obs_p_val, 0, 1)
exp_significance = scipy.stats.norm.isf(exp_p_val, 0, 1)
log.info(f"observed p-value: {obs_p_val:.8%}")
log.info(f"observed significance: {obs_significance:.3f}")
log.info(f"expected p-value: {exp_p_val:.8%}")
log.info(f"expected significance: {exp_significance:.3f}")
significance_results = SignificanceResults(
obs_p_val, obs_significance, exp_p_val, exp_significance
)
return significance_results
def test_set_optimizer_by_bytestring(optimizer_name):
pyhf.set_backend(pyhf.tensorlib, optimizer_name)
assert isinstance(
pyhf.optimizer,
getattr(pyhf.optimize,
f"{optimizer_name.decode('utf-8'):s}_optimizer"),
)
def test_pdf_eval():
tf_sess = tf.Session()
backends = [
numpy_backend(poisson_from_normal=True),
pytorch_backend(),
tensorflow_backend(session=tf_sess),
mxnet_backend()
]
values = []
for b in backends:
pyhf.set_backend(b)
source = {
"binning": [2, -0.5, 1.5],
"bindata": {
"data": [120.0, 180.0],
"bkg": [100.0, 150.0],
"bkgsys_up": [102, 190],
"bkgsys_dn": [98, 100],
"sig": [30.0, 95.0]
}
}
spec = {
'channels': [{
'name':
'singlechannel',
'samples': [{
'name':
'signal',
'data':
source['bindata']['sig'],
'modifiers': [{
'name': 'mu',
'type': 'normfactor',
'data': None
}]
}, {
'name':
'background',
'data':
source['bindata']['bkg'],
'modifiers': [{
'name': 'bkg_norm',
'type': 'histosys',
'data': {
'lo_data': source['bindata']['bkgsys_dn'],
'hi_data': source['bindata']['bkgsys_up']
}
}]
}]
}]
}
pdf = pyhf.hfpdf(spec)
data = source['bindata']['data'] + pdf.config.auxdata
v1 = pdf.logpdf(pdf.config.suggested_init(), data)
values.append(pyhf.tensorlib.tolist(v1)[0])
assert np.std(values) < 1e-6
def test_minuit_strategy_do_grad(mocker, backend):
"""
ref: gh#1172
When there is a user-provided gradient, check that one automatically sets
the minuit strategy=0. When there is no user-provided gradient, check that
one automatically sets the minuit strategy=1.
"""
pyhf.set_backend(pyhf.tensorlib, 'minuit')
spy = mocker.spy(pyhf.optimize.minuit_optimizer, '_minimize')
m = pyhf.simplemodels.hepdata_like([50.0], [100.0], [10.0])
data = pyhf.tensorlib.astensor([125.0] + m.config.auxdata)
do_grad = pyhf.tensorlib.default_do_grad
pyhf.infer.mle.fit(data, m)
assert spy.call_count == 1
assert not spy.spy_return.minuit.strategy == do_grad
pyhf.infer.mle.fit(data, m, strategy=0)
assert spy.call_count == 2
assert spy.spy_return.minuit.strategy == 0
pyhf.infer.mle.fit(data, m, strategy=1)
assert spy.call_count == 3
assert spy.spy_return.minuit.strategy == 1
def test_pdf_eval_2():
tf_sess = tf.Session()
backends = [
numpy_backend(poisson_from_normal=True),
pytorch_backend(),
tensorflow_backend(session=tf_sess),
mxnet_backend()
]
values = []
for b in backends:
pyhf.set_backend(b)
source = {
"binning": [2, -0.5, 1.5],
"bindata": {
"data": [120.0, 180.0],
"bkg": [100.0, 150.0],
"bkgerr": [10.0, 10.0],
"sig": [30.0, 95.0]
}
}
pdf = hepdata_like(source['bindata']['sig'], source['bindata']['bkg'],
source['bindata']['bkgerr'])
data = source['bindata']['data'] + pdf.config.auxdata
v1 = pdf.logpdf(pdf.config.suggested_init(), data)
values.append(pyhf.tensorlib.tolist(v1)[0])
assert np.std(values) < 1e-6
def test_optimizer_unsupported_minimizer_options(optimizer):
pyhf.set_backend(pyhf.default_backend, optimizer())
m = pyhf.simplemodels.hepdata_like([5.0], [10.0], [3.5])
data = pyhf.tensorlib.astensor([10.0] + m.config.auxdata)
with pytest.raises(pyhf.exceptions.Unsupported) as excinfo:
pyhf.infer.mle.fit(data, m, unsupported_minimizer_options=False)
assert 'unsupported_minimizer_options' in str(excinfo.value)
def test_custom_backend_name_supported():
class custom_backend(object):
def __init__(self, **kwargs):
self.name = "pytorch"
with pytest.raises(AttributeError):
pyhf.set_backend(custom_backend())
def test_minimize_do_grad_autoconfig(mocker, backend, backend_new):
backend, do_grad = backend
backend_new, do_grad_new = backend_new
# patch all we need
from pyhf.optimize import mixins
shim = mocker.patch.object(mixins,
'shim',
return_value=({}, lambda x: True))
mocker.patch.object(OptimizerMixin, '_internal_minimize')
mocker.patch.object(OptimizerMixin, '_internal_postprocess')
# start with first backend
pyhf.set_backend(backend, 'scipy')
m = pyhf.simplemodels.hepdata_like([50.0], [100.0], [10.0])
data = pyhf.tensorlib.astensor([125.0] + m.config.auxdata)
assert pyhf.tensorlib.default_do_grad == do_grad
pyhf.infer.mle.fit(data, m)
assert shim.call_args[1]['do_grad'] == pyhf.tensorlib.default_do_grad
pyhf.infer.mle.fit(data, m, do_grad=not (pyhf.tensorlib.default_do_grad))
assert shim.call_args[1]['do_grad'] != pyhf.tensorlib.default_do_grad
# now switch to new backend and see what happens
pyhf.set_backend(backend_new)
m = pyhf.simplemodels.hepdata_like([50.0], [100.0], [10.0])
data = pyhf.tensorlib.astensor([125.0] + m.config.auxdata)
assert pyhf.tensorlib.default_do_grad == do_grad_new
pyhf.infer.mle.fit(data, m)
assert shim.call_args[1]['do_grad'] == pyhf.tensorlib.default_do_grad
pyhf.infer.mle.fit(data, m, do_grad=not (pyhf.tensorlib.default_do_grad))
assert shim.call_args[1]['do_grad'] != pyhf.tensorlib.default_do_grad
def test_jax_jit(caplog, optimizer, do_grad, do_stitch, return_fitted_val):
pyhf.set_backend("jax", optimizer, precision="64b")
pdf = pyhf.simplemodels.hepdata_like([50.0], [100.0], [10.0])
data = pyhf.tensorlib.astensor([125.0] + pdf.config.auxdata)
with caplog.at_level(logging.DEBUG, 'pyhf.optimize.opt_jax'):
pyhf.infer.mle.fixed_poi_fit(
1.0,
data,
pdf,
do_grad=do_grad,
do_stitch=do_stitch,
return_fitted_val=return_fitted_val,
) # jit
assert 'jitting function' in caplog.text
caplog.clear()
with caplog.at_level(logging.DEBUG, 'pyhf.optimize.opt_jax'):
pyhf.infer.mle.fixed_poi_fit(
2.0,
data,
pdf,
do_grad=do_grad,
do_stitch=do_stitch,
return_fitted_val=return_fitted_val,
) # jit
assert 'jitting function' not in caplog.text
with caplog.at_level(logging.DEBUG, 'pyhf.optimize.opt_jax'):
pyhf.infer.mle.fit(
data,
pdf,
do_grad=do_grad,
do_stitch=do_stitch,
return_fitted_val=return_fitted_val,
) # jit
assert 'jitting function' in caplog.text
caplog.clear()
with caplog.at_level(logging.DEBUG, 'pyhf.optimize.opt_jax'):
pyhf.infer.mle.fit(
data,
pdf,
do_grad=do_grad,
do_stitch=do_stitch,
return_fitted_val=return_fitted_val,
) # jit
assert 'jitting function' not in caplog.text
with caplog.at_level(logging.DEBUG, 'pyhf.optimize.opt_jax'):
pyhf.infer.mle.fixed_poi_fit(
3.0,
data,
pdf,
do_grad=do_grad,
do_stitch=do_stitch,
return_fitted_val=return_fitted_val,
) # jit
assert 'jitting function' not in caplog.text
def test_solver_options_scipy(mocker):
optimizer = pyhf.optimize.scipy_optimizer(solver_options={'ftol': 1e-5})
pyhf.set_backend('numpy', optimizer)
assert pyhf.optimizer.solver_options == {'ftol': 1e-5}
model = pyhf.simplemodels.uncorrelated_background([50.0], [100.0], [10.0])
data = pyhf.tensorlib.astensor([125.0] + model.config.auxdata)
assert pyhf.infer.mle.fit(data, model).tolist()
def test_pdf_integration_normsys(backend):
pyhf.set_backend(backend)
if isinstance(pyhf.tensorlib, pyhf.tensor.tensorflow_backend):
tf.reset_default_graph()
pyhf.tensorlib.session = tf.Session()
schema = json.load(open('validation/spec.json'))
source = json.load(open('validation/data/2bin_histosys_example2.json'))
spec = {
'channels': [{
'name':
'singlechannel',
'samples': [{
'name':
'signal',
'data':
source['bindata']['sig'],
'modifiers': [{
'name': 'mu',
'type': 'normfactor',
'data': None
}]
}, {
'name':
'background',
'data':
source['bindata']['bkg'],
'modifiers': [{
'name': 'bkg_norm',
'type': 'normsys',
'data': {
'lo': 0.9,
'hi': 1.1
}
}]
}]
}]
}
jsonschema.validate(spec, schema)
pdf = pyhf.hfpdf(spec)
pars = [None, None]
pars[pdf.config.par_slice('mu')], pars[pdf.config.par_slice(
'bkg_norm')] = [[0.0], [0.0]]
assert np.allclose(
pyhf.tensorlib.tolist(pdf.expected_data(pars, include_auxdata=False)),
[100, 150])
pars[pdf.config.par_slice('mu')], pars[pdf.config.par_slice(
'bkg_norm')] = [[0.0], [1.0]]
assert np.allclose(
pyhf.tensorlib.tolist(pdf.expected_data(pars, include_auxdata=False)),
[100 * 1.1, 150 * 1.1])
pars[pdf.config.par_slice('mu')], pars[pdf.config.par_slice(
'bkg_norm')] = [[0.0], [-1.0]]
assert np.allclose(
pyhf.tensorlib.tolist(pdf.expected_data(pars, include_auxdata=False)),
[100 * 0.9, 150 * 0.9])
def test_custom_optimizer_name_notsupported():
class custom_optimizer(object):
def __init__(self, **kwargs):
self.name = "notsupported"
optimizer = custom_optimizer()
assert pyhf.optimizer.name != optimizer.name
pyhf.set_backend(pyhf.tensorlib, optimizer)
assert pyhf.optimizer.name == optimizer.name
def test_bad_solver_options_scipy(mocker):
optimizer = pyhf.optimize.scipy_optimizer(
solver_options={'arbitrary_option': 'foobar'})
pyhf.set_backend('numpy', optimizer)
assert pyhf.optimizer.solver_options == {'arbitrary_option': 'foobar'}
model = pyhf.simplemodels.hepdata_like([50.0], [100.0], [10.0])
data = pyhf.tensorlib.astensor([125.0] + model.config.auxdata)
assert pyhf.infer.mle.fit(data, model).tolist()
def test_custom_backend_name_notsupported():
class custom_backend(object):
def __init__(self, **kwargs):
self.name = "notsupported"
backend = custom_backend()
assert pyhf.tensorlib.name != backend.name
pyhf.set_backend(backend)
assert pyhf.tensorlib.name == backend.name
def fit(
model: pyhf.pdf.Model,
data: List[float],
init_pars: Optional[List[float]] = None,
fixed_pars: Optional[List[bool]] = None,
asimov: bool = False,
minuit_verbose: bool = False,
) -> FitResults:
"""Performs a maximum likelihood fit, reports and returns the results.
The asimov flag allows to fit the Asimov dataset instead of observed
data.
Args:
model (pyhf.pdf.Model): Model to be used in the fit.
data (List[float]): Data to fit the model to.
init_pars (Optional[List[float]], optional): Initial parameter settings.
Setting to none uses pyhf suggested ones.. Defaults to None.
fixed_pars (Optional[List[bool]], optional): List of parameters to set to be
fixed in the fit. Defaults to None.
asimov (bool, optional): Asimov data or not. Defaults to False.
minuit_verbose (bool, optional): Set minuit verbosity. Defaults to False.
Returns:
FitResults: Object containing fit results.
"""
log.info("performing maximum likelihood fit")
pyhf.set_backend("numpy",
pyhf.optimize.minuit_optimizer(verbose=minuit_verbose))
result, result_obj = pyhf.infer.mle.fit(
data,
model,
init_pars=init_pars,
fixed_params=fixed_pars,
return_uncertainties=True,
return_result_obj=True,
)
bestfit = result[:, 0]
uncertainty = result[:, 1]
labels = model_tools.get_parameter_names(model)
types = model_tools.get_parameter_types(model)
corr_mat = result_obj.corr
cov_mat = result_obj.hess_inv
best_twice_nll = float(result_obj.fun)
fit_result = FitResults(bestfit, uncertainty, labels, types, cov_mat,
corr_mat, best_twice_nll)
log.debug(print_results(fit_result)) # type: ignore
log.debug(
f"-2 log(L) = {fit_result.best_twice_nll:.6f} at the best-fit point")
return fit_result
