Merge pull request #156 from scikit-hep/py312

enh: upgrade to Python312

.github/workflows/main.yml

@@ -31,19 +31,17 @@ jobs:
         os:
           - ubuntu-latest
         python-version:
-          - "3.8"
           - "3.9"
           - "3.10"
           - "3.11"
+          - "3.12"
         include:
           - os: windows-latest
-            python-version: "3.8"
-          - os: windows-latest
-            python-version: "3.11"
-          - os: macos-latest
-            python-version: "3.8"
+            python-version: "3.9"
           - os: macos-latest
-            python-version: "3.11"
+            python-version: "3.9"
+          - os: macos-14
+            python-version: "3.12"  # old versions not supported
     name: Check Python ${{ matrix.python-version }} ${{ matrix.os }}
     steps:
       - uses: actions/checkout@v4
@@ -68,14 +66,15 @@ jobs:
             ${{ runner.os }}-pip-
 
       - name: Install package
-        run: python -m pip install -e .[test] pytest-xdist   # for multiprocessing
-
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install -e .[test] pytest-xdist   # for multiprocessing
 
       - name: Test package
-        run: python -m pytest --doctest-modules --cov=hepstats --cov-report=xml -n3
+        run: python -m pytest --doctest-modules --cov=hepstats --cov-report=xml -n auto
 
       - name: Upload coverage to Codecov
-        if: matrix.python-version == '3.11' && matrix.os == 'ubuntu-latest'
+        if: matrix.python-version == '3.12' && matrix.os == 'ubuntu-latest'
         uses: codecov/codecov-action@v4
         with:
           token: ${{ secrets.CODECOV_TOKEN }}  # technically not needed, but prevents failures: https://community.codecov.com/t/upload-issues-unable-to-locate-build-via-github-actions-api/3954

environment.yml

@@ -6,11 +6,10 @@ dependencies:
   - numpy
   - scipy
   - iminuit
-  - tensorflow>=2.4
   - tensorflow-probability
-  #- zfit
+  #- zfit  # todo: conda-forge is 0.18.1, we need 0.20.0
   - asdf
   - matplotlib
   - pip:
     - .
-    - zfit >=0.6.4
+    - zfit >=0.20.0

setup.cfg

@@ -23,7 +23,6 @@ classifiers =
     Programming Language :: Python
     Programming Language :: Python :: 3
     Programming Language :: Python :: 3 :: Only
-    Programming Language :: Python :: 3.8
     Programming Language :: Python :: 3.9
     Programming Language :: Python :: 3.10
     Programming Language :: Python :: 3.11
@@ -43,7 +42,7 @@ install_requires =
     scipy
     tqdm
     uhi
-python_requires = >=3.8
+python_requires = >=3.9
 package_dir =
     = src
 
@@ -52,8 +51,9 @@ where = src
 
 [options.extras_require]
 dev =
-    black
-    zfit
+    %(docs)s
+    %(test)s
+    pre-commit
 docs =
     matplotlib
     pydata-sphinx-theme
@@ -65,7 +65,9 @@ test =
     pytest
     pytest-cov
     pytest-runner
-    zfit
+    zfit>=0.20.0
+zfit =
+    zfit>=0.20.0
 
 [tool:pytest]
 junit_family = xunit2

src/hepstats/hypotests/hypotests_object.py

@@ -4,7 +4,7 @@
 
 import numpy as np
 
-from ..utils.fit import get_nevents
+from ..utils.fit import get_nevents, set_values_once
 from ..utils.fit.api_check import is_valid_data, is_valid_fitresult, is_valid_loss, is_valid_minimizer, is_valid_pdf
 from .parameters import POI
 
@@ -121,8 +121,7 @@ def set_params_to_bestfit(self):
         """
         Set the values of the parameters in the models to the best fit values
         """
-        for param in self.parameters:
-            param.set_value(self.bestfit.params[param]["value"])
+        set_values_once(self.parameters, self.bestfit)
 
     def lossbuilder(self, model, data, weights=None, oldloss=None):
         """Method to build a new loss function.
@@ -160,7 +159,7 @@ def lossbuilder(self, model, data, weights=None, oldloss=None):
 
         if weights is not None:
             for d, w in zip(data, weights):
-                d.set_weights(w)
+                d = d.with_weights(w)
 
         if hasattr(oldloss, "create_new"):
             loss = oldloss.create_new(model=model, data=data, constraints=self.constraints)
@@ -196,15 +195,11 @@ def __init__(self, input, minimizer, sampler, sample):
         self._sample = sample
         self._toys_loss = {}
 
-    def sampler(self, floating_params=None):
+    def sampler(self):
         """
         Create sampler with models.
 
-        Args:
-            floating_params: floating parameters in the sampler
-
-        Example with `zfit`:
-            >>> sampler = calc.sampler(floating_params=[zfit.Parameter("mean")])
+        >>> sampler = calc.sampler()
         """
         self.set_params_to_bestfit()
         nevents = []
@@ -215,7 +210,7 @@ def sampler(self, floating_params=None):
             else:
                 nevents.append(nevents_data)
 
-        return self._sampler(self.loss.model, nevents, floating_params)
+        return self._sampler(self.loss.model, nevents)
 
     def sample(self, sampler, ntoys, poi: POI, constraints=None):
         """
@@ -230,7 +225,7 @@ def sample(self, sampler, ntoys, poi: POI, constraints=None):
 
         Example with `zfit`:
             >>> mean = zfit.Parameter("mean")
-            >>> sampler = calc.sampler(floating_params=[mean])
+            >>> sampler = calc.sampler()
             >>> sample = calc.sample(sampler, 1000, POI(mean, 1.2))
 
         Returns:
@@ -258,6 +253,6 @@ def toys_loss(self, parameter_name: str):
         """
         if parameter_name not in self._toys_loss:
             parameter = self.get_parameter(parameter_name)
-            sampler = self.sampler(floating_params=[parameter])
+            sampler = self.sampler()
             self._toys_loss[parameter.name] = self.lossbuilder(self.model, sampler)
         return self._toys_loss[parameter_name]

src/hepstats/hypotests/toyutils.py

@@ -2,11 +2,11 @@
 
 import warnings
 from collections.abc import Callable
-from contextlib import ExitStack
 from pathlib import Path
 
 import asdf
 import numpy as np
+import zfit.param
 from tqdm.auto import tqdm
 
 from ..utils import base_sample, base_sampler, pll
@@ -217,7 +217,7 @@ def ntoys(self, poigen: POI, poieval: POIarray) -> int:
         except KeyError:
             return 0
 
-    def generate_and_fit_toys(
+    def generate_and_fit_toys(  # TODO PROFILE THIS
         self,
         ntoys: int,
         poigen: POI,
@@ -263,6 +263,7 @@ def generate_and_fit_toys(
         ntrials = 0
 
         progressbar = tqdm(total=ntoys)
+        minimum = None
 
         for i in range(ntoys):
             ntrials += 1
@@ -280,13 +281,12 @@ def generate_and_fit_toys(
                     )
                     param_dict = next(samples)
 
-                with ExitStack() as stack:
-                    for param, value in param_dict.items():
-                        stack.enter_context(param.set_value(value))
-
+                with zfit.param.set_values(param_dict):
                     for _ in range(2):
                         try:
-                            minimum = minimizer.minimize(loss=toys_loss)
+                            minimum = minimizer.minimize(
+                                loss=toys_loss
+                            )  # TODO: , init=minimum use previous minimum as starting point for parameter uncertainties
                             converged = minimum.converged
                             if converged:
                                 break
@@ -301,7 +301,9 @@ def generate_and_fit_toys(
                         msg = f"{nfailures} out of {ntrials} fits failed or didn't converge."
                         warnings.warn(msg, FitFailuresWarning, stacklevel=2)
                     continue
-
+                if minimum is None:
+                    msg = "No minimum found."
+                    raise RuntimeError(msg)
                 bestfit[i] = minimum.params[param]["value"]
                 nll_bestfit[i] = pll(minimizer, toys_loss, POI(param, bestfit[i]))
 

src/hepstats/utils/fit/__init__.py

@@ -1,2 +1,10 @@
-from .diverse import get_value, eval_pdf, pll, array2dataset, get_nevents, set_values
+from .diverse import (
+    get_value,
+    eval_pdf,
+    pll,
+    array2dataset,
+    get_nevents,
+    set_values,
+    set_values_once,
+)
 from .sampling import base_sampler, base_sample

src/hepstats/utils/fit/diverse.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from contextlib import ExitStack, contextmanager
+from contextlib import ExitStack, contextmanager, suppress
 
 import numpy as np
 
@@ -14,6 +14,17 @@ def get_value(value):
     return np.array(value)
 
 
+def set_values_once(params, values):
+    with suppress(ImportError):
+        import zfit
+
+        return zfit.param.set_values(params, values)  # more efficient
+
+    for p, v in zip(params, values):
+        p.set_value(v)
+    return None
+
+
 def eval_pdf(model, x, params=None, allow_extended=False):
     """Compute pdf of model at a given point x and for given parameters values"""
 
@@ -33,8 +44,9 @@ def pdf(model, x):
         return pdf(model, x)
 
 
-def pll(minimizer, loss, pois) -> float:
+def pll(minimizer, loss, pois, init=None) -> float:
     """Compute minimum profile likelihood for fixed given parameters values."""
+    del init  # unused currently
 
     with ExitStack() as stack:
         for p in pois:
@@ -43,7 +55,7 @@ def pll(minimizer, loss, pois) -> float:
             param.floating = False
 
         if any(param_loss.floating for param_loss in loss.get_params()):
-            minimum = minimizer.minimize(loss=loss)
+            minimum = minimizer.minimize(loss=loss)  # TODO: add init?
             value = minimum.fmin
         else:
             value = get_value(loss.value())

src/hepstats/utils/fit/sampling.py

@@ -8,14 +8,13 @@
 from .diverse import get_value
 
 
-def base_sampler(models, nevents, floating_params=None):
+def base_sampler(models, nevents):
     """
     Creates samplers from models.
 
     Args:
         models (list(model)): models to sample
         nevents (list(int)): number of in each sampler
-        floating_params (list(parameter), optionnal): floating parameter in the samplers
 
     Returns:
         Samplers
@@ -24,24 +23,10 @@ def base_sampler(models, nevents, floating_params=None):
     assert all(is_valid_pdf(m) for m in models)
     assert len(nevents) == len(models)
 
-    if floating_params:
-        floating_params_names = [f.name for f in floating_params]
-
     samplers = []
-    fixed_params = []
-    for m in models:
-
-        def to_fix(p):
-            if floating_params:
-                return p.name in floating_params_names
-            else:
-                return False
-
-        fixed = [p for p in m.get_params() if not to_fix(p)]
-        fixed_params.append(fixed)
 
-    for i, (m, p) in enumerate(zip(models, fixed_params)):
-        sampler = m.create_sampler(n=nevents[i], fixed_params=p)
+    for i, m in enumerate(models):
+        sampler = m.create_sampler(n=nevents[i])
         samplers.append(sampler)
 
     return samplers
@@ -72,13 +57,10 @@ def base_sample(samplers, ntoys, parameter=None, value=None, constraints=None):
                 continue
 
     for i in range(ntoys):
-        if not (parameter is None or value is None):
-            with parameter.set_value(value):
-                for s in samplers:
-                    s.resample()
-        else:
-            for s in samplers:
-                s.resample()
+        params = None if parameter is None or value is None else {parameter: value}
+
+        for s in samplers:
+            s.resample(params=params)
 
         if constraints is not None:
             yield {param: value[i] for param, value in sampled_constraints.items()}

tests/hypotests/test_calculators.py

@@ -187,7 +187,7 @@ def test_frequentist_calculator_one_poi(constraint):
     assert calc.ntoysnull == 100
     assert calc.ntoysalt == 100
 
-    samplers = calc.sampler(floating_params=[mean])
+    samplers = calc.sampler()
     assert all(is_valid_data(s) for s in samplers)
     loss = calc.toys_loss(mean.name)
     assert is_valid_loss(loss)