[pattems]

Summary

• kliff/models/kim.py imports kimpy twice: once guarded (lines 21-27, try: import kimpy ... kimpy_avail = True / except ImportError: kimpy_avail = False) and once unguarded at the top (line 9, import kimpy). The unguarded one was added in 614c1b9 ("Working KIM trainer module") and is in v1.0.0, v1.0.1 and current main.
• kimpy is meant to be optional (it is only in the test extra, see kimpy is missing from install_requires #80), but the bare import defeats the kimpy_avail guard: without kimpy, import kliff.models fails with ModuleNotFoundError: No module named 'kimpy', and so does everything
  that goes through kliff/models/__init__.py (e.g. kliff.models.lennard_jones, kliff.models.parameter, kliff.trainer.kim_trainer, kliff.cmdline.model), even for users who never touch KIM models.
• Fix: delete the stray line. Nothing else in kim.py, or in the modules kliff.models imports, uses kimpy at module level; all other uses are inside methods that are reached only after the kimpy_avail check, so KIMModel(...) without kimpy now raises the intended ImportError: To use KIMModel, kimpy is need. Please install it first.

Additional dependencies introduced (if any)

None.

Testing

Python 3.12.13, 3.13.14 and 3.14.3 on Windows, each in a fresh venv with numpy, scipy, ase, loguru, monty, pyyaml, requests, pytest and WITHOUT kimpy, run with python -B and __pycache__ cleared.

The C++ extensions could not be built in this environment (MSVC rejects the variable-length array in kliff/neighbor/neighbor_list.cpp), so the four extension modules only - kliff.neighbor.neighlist,
kliff.legacy.descriptors.symmetry_function.sf, kliff.legacy.descriptors.bispectrum.bs and kliff.transforms.configuration_transforms.graphs.graph_module - were replaced by empty modules in sys.modules before importing kliff. kimpy was not stubbed except in the last check below.

Before (main, df2d4de), all three versions:

import kliff.models
-> ModuleNotFoundError: No module named 'kimpy'
pytest tests/models/test_parameter.py tests/models/test_parameter_transform.py
-> 2 errors during collection (ModuleNotFoundError: No module named 'kimpy')

After, all three versions:

import kliff.models                      -> OK, kliff.models.kim.kimpy_avail == False
KIMModel("SW_StillingerWeber_1985_Si__MO_405512056662_006")
                                         -> ImportError: To use `KIMModel`, `kimpy` is need. ...
with a stub `kimpy` module in sys.modules -> OK, kimpy_avail == True
pytest tests/models/test_parameter.py tests/models/test_parameter_transform.py
                                         -> 2 passed

Second commit: invalid escape sequences in docstrings

The math markup in three docstrings uses backslash sequences that are not valid string escapes, so Python 3.12 and later emit
SyntaxWarning: invalid escape sequence when the modules are compiled:

• kliff/utils.py:185 (stress_to_voigt): \sigma
• kliff/dataset/configuration.py:431 (Configuration.stress setter): \sigma
• kliff/uq/mcmc.py:271: \propto and \exp (the same line already writes \\theta)

The two \sigma docstrings become raw strings; in mcmc.py the backslashes are doubled, matching the existing \\theta. The docstring text is unchanged: every string constant in the three files is identical before and after (compared with ast). Compiling all of kliff/ gives 3 SyntaxWarnings before and 0 after, on Python 3.12.13, 3.13.14 and 3.14.3.