bump monty to use the monty.json import speedup patch, add import time regression test, lazy load some rarely used but costly modules

[DanielYang59]

Summary

• Organize optional dependencies in pyproject.toml to avoid duplicate
• Bump sympy >= 1.3 to resolve bump networkx to 2.7 to fix intermittent CI failure when torch install failed: ImportError: cannot import name 'Mapping' from 'collections' #4116 (comment)
• Avoid from numpy.testing import assert_allclose in production code

---

• bump monty to use the latest json import speedup patch, partially fix import monty.json slowing down core import #3793

• Have a quick look at other important modules (on cover those import other 3rd-party/non-core-pmg modules in this PR) and add import time test (profile: python -X importtime -c "from pymatgen.core.structure import Structure" 2> pmg.log && tuna pmg.log):

  • core.bonds:
  • core.composition:
  • core.interface (need attention)
  • core.ion:
  • core.lattice:
  • core.operations:
  • core.periodic_table:
  • core.sites:
  • core.spectrum:
  • core.structure
  • core.surface:
  • core.tensors:
  • core.trajectory (AseAtomsAdaptor lazy imported)
  • io.vasp.inputs
  • io.vasp.outputs:

• [For a follow up PR] looks like scipy across core need special attention

[DanielYang59]

@mkhorton Can I lazy import sympy (it's used by only one method symmetry.settings.JonesFaithfulTransformation.parse_transformation_string and would give us a ~15% speed up on import core.Structure)?

[DanielYang59]

AseAtomsAdaptor is only used in one of the many try-except branches (other branches also lazy import the corresponding modules), ~10% speed up.

pymatgen/src/pymatgen/core/trajectory.py

Lines 569 to 593 in 6992aee

	if fnmatch(filename, "*XDATCAR*"):
	from pymatgen.io.vasp.outputs import Xdatcar
	structures = Xdatcar(filename).structures
	elif fnmatch(filename, "vasprun*.xml*"):
	from pymatgen.io.vasp.outputs import Vasprun
	structures = Vasprun(filename).structures
	elif fnmatch(filename, "*.traj"):
	try:
	from ase.io.trajectory import Trajectory as AseTrajectory
	ase_traj = AseTrajectory(filename)
	# Periodic boundary conditions should be the same for all frames so just check the first
	pbc = ase_traj[0].pbc
	if any(pbc):
	structures = [AseAtomsAdaptor.get_structure(atoms) for atoms in ase_traj]
	else:
	molecules = [AseAtomsAdaptor.get_molecule(atoms) for atoms in ase_traj]
	is_mol = True
	except ImportError as exc:
	raise ImportError("ASE is required to read .traj files. pip install ase") from exc

[DanielYang59]

matplotlib is likely not a core module for pymatgen, and incurred noticeable overhead to core.interface:

After:

[DanielYang59]

@mkhorton I believe this PR is already for view, let me know if you have any comment, thank you!

Credit to @janosh for the motivation and helpful discussion!

[janosh]

thanks a lot for your extensive profiling work @DanielYang59! i expected there would be a some low-hanging fruit but not nearly as much as you found. you went above and beyond here. 👍
will be big help in resolving janosh/pymatviz#209