def test_unbinned_nll(weights, sigma, options):
gaussian1, mu1, sigma1 = create_gauss1()
gaussian2, mu2, sigma2 = create_gauss2()
test_values = tf.constant(test_values_np)
test_values = zfit.Data.from_tensor(obs=obs1, tensor=test_values, weights=weights)
nll_object = zfit.loss.UnbinnedNLL(model=gaussian1, data=test_values, options=options)
minimizer = Minuit(tol=1e-5)
status = minimizer.minimize(loss=nll_object, params=[mu1, sigma1])
params = status.params
rel_error = 0.12 if weights is None else 0.1 # more fluctuating with weights
assert params[mu1]['value'] == pytest.approx(np.mean(test_values_np), rel=rel_error)
assert params[sigma1]['value'] == pytest.approx(np.std(test_values_np), rel=rel_error)
constraints = zfit.constraint.nll_gaussian(params=[mu2, sigma2],
observation=[mu_constr[0], sigma_constr[0]],
uncertainty=sigma())
nll_object = UnbinnedNLL(model=gaussian2, data=test_values,
constraints=constraints, options=options)
minimizer = Minuit(tol=1e-4)
status = minimizer.minimize(loss=nll_object, params=[mu2, sigma2])
params = status.params
if weights is None:
assert params[mu2]['value'] > np.average(test_values_np, weights=weights)
assert params[sigma2]['value'] < np.std(test_values_np)
def test_sweights():
minimizer = Minuit()
mass, p, loss, Nsig, Nbkg, sig_p, bkg_p = get_data_and_loss()
with pytest.raises(ModelNotFittedToData):
compute_sweights(loss.model[0], mass)
minimizer.minimize(loss)
model = loss.model[0]
assert is_sum_of_extended_pdfs(model)
yields = [Nsig, Nbkg]
sweights = compute_sweights(loss.model[0], mass)
assert np.allclose(
[np.sum(sweights[y]) / get_value(y.value()) for y in yields], 1.0)
nbins = 30
hist_conf = dict(bins=nbins, range=[0, 10])
hist_sig_true_p, _ = np.histogram(sig_p, **hist_conf)
sel = hist_sig_true_p != 0
hist_sig_true_p = hist_sig_true_p[sel]
hist_sig_sweights_p = np.histogram(p, weights=sweights[Nsig],
**hist_conf)[0][sel]
assert chisquare(hist_sig_sweights_p, hist_sig_true_p)[-1] < 0.01
hist_bkg_true_p, _ = np.histogram(bkg_p, **hist_conf)
sel = hist_bkg_true_p != 0
hist_bkg_true_p = hist_bkg_true_p[sel]
hist_bkg_sweights_p = np.histogram(p, weights=sweights[Nbkg],
**hist_conf)[0][sel]
assert chisquare(hist_bkg_sweights_p, hist_bkg_true_p)[-1] < 0.01
with pytest.warns(AboveToleranceWarning):
compute_sweights(
loss.model[0],
np.concatenate([mass, np.random.normal(0.8, 0.1, 100)]))
with pytest.raises(ModelNotFittedToData):
compute_sweights(
loss.model[0],
np.concatenate([mass, np.random.normal(0.8, 0.1, 1000)]))
def test_with_gauss_qtilde(n, min_x):
sigma_x = 0.032
minimizer = Minuit()
bounds = (-10, 10)
obs = zfit.Space("x", limits=bounds)
mean = zfit.Parameter("mean", n * sigma_x)
sigma = zfit.Parameter("sigma", 1.0)
model = zfit.pdf.Gauss(obs=obs, mu=mean, sigma=sigma)
data = model.sample(n=1000)
nll = UnbinnedNLL(model=model, data=data)
minimum = minimizer.minimize(loss=nll)
minimum.hesse()
x = minimum.params[mean]["value"]
x_err = minimum.params[mean]["minuit_hesse"]["error"]
x_min = x - x_err * 3
x_max = x + x_err * 3
x_min = max([x_min, min_x])
poinull = POIarray(mean, np.linspace(x_min, x_max, 50))
calculator = AsymptoticCalculator(nll, minimizer)
ci = ConfidenceInterval(calculator, poinull, qtilde=True)
ci.interval(alpha=0.05, printlevel=1)
def test_with_gauss_fluctuations():
x_true = -2.0
minimizer = Minuit()
bounds = (-10, 10)
obs = zfit.Space("x", limits=bounds)
mean = zfit.Parameter("mean", 0)
sigma = zfit.Parameter("sigma", 1.0)
model = zfit.pdf.Gauss(obs=obs, mu=mean, sigma=sigma)
npzfile = f"{notebooks_dir}/toys/FC_toys_{x_true}.npz"
data = zfit.data.Data.from_numpy(obs=obs, array=np.load(npzfile)["x"])
nll = UnbinnedNLL(model=model, data=data)
minimum = minimizer.minimize(loss=nll)
minimum.hesse()
toys_fname = f"{notebooks_dir}/toys/FC_toys_{x_true}.yml"
calculator = FrequentistCalculator.from_yaml(toys_fname, minimum,
minimizer)
keys = np.unique([k[0].value for k in calculator.keys()])
keys.sort()
poinull = POIarray(mean, keys)
ci = ConfidenceInterval(calculator, poinull, qtilde=False)
with pytest.warns(UserWarning):
ci.interval(alpha=0.05, printlevel=0)
ci = ConfidenceInterval(calculator, poinull, qtilde=True)
ci.interval(alpha=0.05, printlevel=0)
def test_unbinned_simultaneous_nll():
test_values = tf.constant(test_values_np)
test_values = zfit.Data.from_tensor(obs=obs1, tensor=test_values)
test_values2 = tf.constant(test_values_np2)
test_values2 = zfit.Data.from_tensor(obs=obs1, tensor=test_values2)
gaussian1, mu1, sigma1 = create_gauss1()
gaussian2, mu2, sigma2 = create_gauss2()
gaussian2 = gaussian2.create_extended(zfit.Parameter('yield_gauss2', 5))
nll = zfit.loss.UnbinnedNLL(
model=[gaussian1, gaussian2],
data=[test_values, test_values2],
)
minimizer = Minuit(tolerance=1e-5)
status = minimizer.minimize(loss=nll, params=[mu1, sigma1, mu2, sigma2])
params = status.params
assert set(nll.get_params()) == {mu1, mu2, sigma1, sigma2}
assert params[mu1]['value'] == pytest.approx(np.mean(test_values_np),
rel=0.007)
assert params[mu2]['value'] == pytest.approx(np.mean(test_values_np2),
rel=0.007)
assert params[sigma1]['value'] == pytest.approx(np.std(test_values_np),
rel=0.007)
assert params[sigma2]['value'] == pytest.approx(np.std(test_values_np2),
rel=0.007)
def test_binned_nll(weights):
obs = zfit.Space("obs1", limits=(-15, 25))
gaussian1, mu1, sigma1 = create_gauss1(obs=obs)
gaussian2, mu2, sigma2 = create_gauss2(obs=obs)
test_values_np = np.random.normal(loc=mu_true, scale=4, size=(10000, 1))
test_values = tf.constant(test_values_np)
test_values = zfit.Data.from_tensor(obs=obs,
tensor=test_values,
weights=weights)
test_values_binned = test_values.create_hist(
converter=zfit.hist.histogramdd, bin_kwargs={"bins": 100})
nll_object = zfit.loss.BinnedNLL(model=gaussian1, data=test_values_binned)
minimizer = Minuit()
status = minimizer.minimize(loss=nll_object, params=[mu1, sigma1])
params = status.params
rel_error = 0.035 if weights is None else 0.15 # more fluctuating with weights
assert params[mu1]["value"] == pytest.approx(np.mean(test_values_np),
rel=rel_error)
assert params[sigma1]["value"] == pytest.approx(np.std(test_values_np),
rel=rel_error)
constraints = zfit.constraint.nll_gaussian(
params=[mu2, sigma2],
mu=[mu_constr[0], sigma_constr[0]],
sigma=[mu_constr[1], sigma_constr[1]],
)
nll_object = zfit.loss.BinnedNLL(
model=gaussian2,
data=test_values_binned,
fit_range=(-np.infty, np.infty),
constraints=constraints,
)
minimizer = Minuit()
status = minimizer.minimize(loss=nll_object, params=[mu2, sigma2])
params = status.params
if weights is None:
assert params[mu2]["value"] > np.mean(test_values_np)
assert params[sigma2]["value"] < np.std(test_values_np)
def test_extended_unbinned_nll():
test_values = z.constant(test_values_np)
test_values = zfit.Data.from_tensor(obs=obs1, tensor=test_values)
gaussian3, mu3, sigma3, yield3 = create_gauss3ext()
nll = zfit.loss.ExtendedUnbinnedNLL(model=gaussian3,
data=test_values,
fit_range=(-20, 20))
assert {mu3, sigma3, yield3} == nll.get_params()
minimizer = Minuit()
status = minimizer.minimize(loss=nll)
params = status.params
assert params[mu3]['value'] == pytest.approx(np.mean(test_values_np), rel=0.007)
assert params[sigma3]['value'] == pytest.approx(np.std(test_values_np), rel=0.007)
assert params[yield3]['value'] == pytest.approx(yield_true, rel=0.007)
def test_unbinned_simultaneous_nll():
mu1, mu2, nll, sigma1, sigma2 = create_simultaneous_loss()
minimizer = Minuit(tol=1e-5)
status = minimizer.minimize(loss=nll, params=[mu1, sigma1, mu2, sigma2])
params = status.params
assert set(nll.get_params()) == {mu1, mu2, sigma1, sigma2}
assert params[mu1]["value"] == pytest.approx(np.mean(test_values_np),
rel=0.007)
assert params[mu2]["value"] == pytest.approx(np.mean(test_values_np2),
rel=0.007)
assert params[sigma1]["value"] == pytest.approx(np.std(test_values_np),
rel=0.007)
assert params[sigma2]["value"] == pytest.approx(np.std(test_values_np2),
rel=0.007)
def test_extended_unbinned_nll(size):
if size is None:
test_values = z.constant(test_values_np)
size = test_values.shape[0]
else:
test_values = create_test_values(size)
test_values = zfit.Data.from_tensor(obs=obs1, tensor=test_values)
gaussian3, mu3, sigma3, yield3 = create_gauss3ext()
nll = zfit.loss.ExtendedUnbinnedNLL(model=gaussian3,
data=test_values,
fit_range=(-20, 20))
assert {mu3, sigma3, yield3} == nll.get_params()
minimizer = Minuit(tol=1e-4)
status = minimizer.minimize(loss=nll)
params = status.params
assert params[mu3]['value'] == pytest.approx(zfit.run(tf.math.reduce_mean(test_values)), rel=0.05)
assert params[sigma3]['value'] == pytest.approx(zfit.run(tf.math.reduce_std(test_values)), rel=0.05)
assert params[yield3]['value'] == pytest.approx(size, rel=0.005)