build: Update lower bound of pyhf to v0.7.0 (#367)

* Update lower bound of pyhf to v0.7.0
* Update model.config.par_names() to model.config.par_names due to pyhf API change

setup.cfg

@@ -23,7 +23,7 @@ packages = find:
 package_dir = =src
 python_requires = >=3.7
 install_requires =
-    pyhf[minuit]==0.7.0rc4  # model.config.suggested_fixed API change, staterror fix, set_poi(None)
+    pyhf[minuit]~=0.7.0  # model.config.suggested_fixed / .par_names API changes, set_poi(None)
     boost_histogram>=1.0.0  # subclassing with family, 1.02 for stdev scaling fix (currently not needed)
     awkward>=1.8  # _v2 API in submodule
     tabulate>=0.8.1  # multiline text

src/cabinetry/fit/__init__.py

@@ -109,7 +109,7 @@ def _fit_model_pyhf(
     uncertainty = np.where(
         result_obj.minuit.fixed, 0.0, pyhf.tensorlib.to_numpy(result[:, 1])
     )
-    labels = model.config.par_names()
+    labels = model.config.par_names
     corr_mat = pyhf.tensorlib.to_numpy(corr_mat)
     best_twice_nll = float(best_twice_nll)  # convert 0-dim np.ndarray to float
 
@@ -183,7 +183,7 @@ def _fit_model_custom(
     fix_pars = fix_pars or model.config.suggested_fixed()
     par_bounds = par_bounds or model.config.suggested_bounds()
 
-    labels = model.config.par_names()
+    labels = model.config.par_names
 
     def twice_nll_func(pars: np.ndarray) -> Any:
         """The objective for minimization: twice the negative log-likelihood.
@@ -560,7 +560,7 @@ def ranking(
             custom_fit=custom_fit,
         )
 
-    labels = model.config.par_names()
+    labels = model.config.par_names
     prefit_unc = model_utils.prefit_uncertainties(model)
 
     # use POI given by kwarg, fall back to POI specified in model
@@ -694,7 +694,7 @@ def scan(
         ScanResults: includes parameter name, scanned values and 2*log(likelihood)
         offset
     """
-    labels = model.config.par_names()
+    labels = model.config.par_names
 
     # get index of parameter with name par_name
     par_index = model_utils._parameter_index(par_name, labels)
@@ -837,7 +837,7 @@ def limit(
     # set POI name in model config to desired value, hypotest will pick this up
     # save original value to reset model later
     original_model_poi_name = model.config.poi_name
-    model.config.set_poi(model.config.par_names()[poi_index])
+    model.config.set_poi(model.config.par_names[poi_index])
 
     # show two decimals only if confidence level in percent is not an integer
     cl_label = (
@@ -1091,7 +1091,7 @@ def significance(
     # set POI name in model config to desired value, hypotest will pick this up
     # save original value to reset model later
     original_model_poi_name = model.config.poi_name
-    model.config.set_poi(model.config.par_names()[poi_index])
+    model.config.set_poi(model.config.par_names[poi_index])
 
     log.info(f"calculating discovery significance for {model.config.poi_name}")
     obs_p_val, exp_p_val = pyhf.infer.hypotest(

src/cabinetry/model_utils.py

@@ -425,7 +425,7 @@ def prediction(
         ModelPrediction: model, yields and uncertainties per bin and channel
     """
     if fit_results is not None:
-        if fit_results.labels != model.config.par_names():
+        if fit_results.labels != model.config.par_names:
             log.warning("parameter names in fit results and model do not match")
         # fit results specified, so they are used
         param_values = fit_results.bestfit
@@ -531,7 +531,7 @@ def _poi_index(
     """
     if poi_name is not None:
         # use POI given by kwarg if specified
-        poi_index = _parameter_index(poi_name, model.config.par_names())
+        poi_index = _parameter_index(poi_name, model.config.par_names)
         if poi_index is None:
             raise ValueError(f"parameter {poi_name} not found in model")
     elif model.config.poi_index is not None:
@@ -642,7 +642,7 @@ def match_fit_results(model: pyhf.pdf.Model, fit_results: FitResults) -> FitResu
 
     bestfit = asimov_parameters(model)  # Asimov parameter values for target model
     uncertainty = prefit_uncertainties(model)  # pre-fit uncertainties for target model
-    labels = model.config.par_names()  # labels for target model
+    labels = model.config.par_names  # labels for target model
 
     # indices of parameters in current fit results, or None if they are missing
     indices_for_corr: List[Optional[int]] = [None] * len(labels)

tests/test_model_utils.py

@@ -470,7 +470,7 @@ def test_match_fit_results(mock_pars, mock_uncs):
     )
 
     # remove par_a, flip par_b and par_c, add par_d
-    mock_model.config.par_names.return_value = ["par_c", "par_d", "par_b"]
+    mock_model.config.par_names = ["par_c", "par_d", "par_b"]
     matched_fit_res = model_utils.match_fit_results(mock_model, fit_results)
     assert mock_pars.call_args_list == [((mock_model,), {})]
     assert mock_uncs.call_args_list == [((mock_model,), {})]
@@ -484,7 +484,7 @@ def test_match_fit_results(mock_pars, mock_uncs):
     assert matched_fit_res.goodness_of_fit == 0.1
 
     # all parameters are new
-    mock_model.config.par_names.return_value = ["par_d", "par_e"]
+    mock_model.config.par_names = ["par_d", "par_e"]
     matched_fit_res = model_utils.match_fit_results(mock_model, fit_results)
     assert np.allclose(matched_fit_res.bestfit, [4.0, 5.0])
     assert np.allclose(matched_fit_res.uncertainty, [0.4, 0.5])
@@ -494,7 +494,7 @@ def test_match_fit_results(mock_pars, mock_uncs):
     assert matched_fit_res.goodness_of_fit == 0.1
 
     # fit results already match model exactly
-    mock_model.config.par_names.return_value = ["par_a", "par_b", "par_c"]
+    mock_model.config.par_names = ["par_a", "par_b", "par_c"]
     matched_fit_res = model_utils.match_fit_results(mock_model, fit_results)
     assert np.allclose(matched_fit_res.bestfit, [1.0, 2.0, 3.0])
     assert np.allclose(matched_fit_res.uncertainty, [0.1, 0.2, 0.3])