deepmodeling/dpdata

dpdata

Performance History

Latest Results

fix(deepmd/hdf5): skip empty optional frame arrays on dump (#996) The deepmd/hdf5 writer reshaped and wrote optional frame arrays (e.g. forces or virials that are empty when cal_force/cal_stress is disabled) as a (nframes, 0) dataset, which the loader then failed to reshape back to the full shape, raising 'cannot reshape array of size 0'. Skip empty optional arrays on dump, matching the existing deepmd/raw and deepmd/npy writers. Non-empty arrays are unaffected.

CedricConday:fix/hdf5-empty-optional-arrays

13 hours ago

fix: build pymatgen Molecule species from atom_types order (#994) (#1010) Fixes #994. ## Summary `PyMatgenMoleculeFormat.to_system()` built the `species` list by expanding grouped `atom_names`/`atom_numbs` (e.g. `H, H, O`), but passed `coords` in their `atom_types` order (e.g. `H, O, H`). When atoms are **not** grouped by type, the resulting `pymatgen.core.Molecule` assigns species to the wrong coordinates. ## Change Build `species` per-atom from `atom_types`, matching the coordinate order — the same construction the sibling Structure exporter already uses in this file: ```python species = [data["atom_names"][tt] for tt in data["atom_types"]] ``` This is the fix suggested in the issue. ## Testing - Added a regression test (`test_ungrouped_atom_types_species_match_coords`) using the reporter's reproducer: species now follow the coordinate order (`H, O, H`) and each site sits on its original coordinate. It **fails on master** and passes with this change. - All 50 pymatgen-related tests pass; `ruff check` and `ruff format --check` clean. --- *Disclosure: authored with AI assistance (Claude Code); reviewed by me and I'll shepherd it through review.*  ## Summary by CodeRabbit * **Bug Fixes** * Fixed molecule conversion so atom species now follow the actual per-atom type order. * Ensured converted molecule coordinates continue to match the source structure for each frame. * **Tests** * Added a regression test covering ungrouped atom types to verify species ordering and coordinate mapping during molecule conversion.

master

3 days ago

fix: build pymatgen Molecule species from atom_types order (#994) PyMatgenMoleculeFormat.to_system() built the species list by expanding grouped atom_names/atom_numbs, but passed coords in atom_types order. When atoms are not grouped by type, species were assigned to the wrong coordinates. Build species per-atom from atom_types, matching the coord order and the sibling Structure exporter (line 37). Adds a regression test that fails on master (['H','H','O'] vs ['H','O','H']) and passes with this change.

CedricConday:fix/pymatgen-molecule-species-order-994

3 days ago

feat(extxyz): log unit conversions when factor != 1.0 When energy-unit, force-unit, or stress-unit headers trigger a non-trivial conversion, emit an INFO-level log message showing the source unit and conversion factor for traceability.

SchrodingersCattt:devel

4 days ago

fix(extxyz): convert virial with energy-unit factor and add au/a.u. aliases - Apply e_factor to explicit virial= values (virial has energy dimensions) - Add 'au' and 'a.u.' as aliases for hartree in _ENERGY_UNIT_MAP - Handle force-unit=au/a.u. as hartree/bohr in _parse_force_unit - Add regression tests for non-eV virial and atomic-unit aliases

SchrodingersCattt:devel

4 days ago

fix: preserve atom order in map_atom_types (#1007) ## Summary - map each existing atom type entry through `atom_names` instead of expanding grouped `atom_numbs` - add a regression test for interleaved atom order (`[H, O, H]`) Fixes #984. ## Verification - `python -m pytest test_system_set_type.py -q` (from `tests/`): `3 passed in 0.20s` - `python -m ruff check dpdata/system.py tests/test_system_set_type.py`: `All checks passed!` - issue reproducer now prints `[0, 1, 0]`  ## Summary by CodeRabbit * **Bug Fixes** * Improved atom-type remapping so it now preserves the current atom order when converting to a new type map. * Ensured remapped atom types are generated consistently from existing atom-type indices. * **Tests** * Added coverage verifying atom-type remapping returns the expected order for a simple in-memory system.

master

4 days ago

fix: add missing elements 104-118 (Rf~Og) to periodic_table.json (#1009) The ELEMENTS list in periodic_table.py already contains all 118 element symbols, but periodic_table.json only had entries for elements 1-103 (H~Lr). This mismatch causes KeyError when dpdata.System() is constructed with a type_map that includes elements beyond Lr, and subsequently to_lammps_lmp() is called. The _get_lammps_masses() safety check passes (all names in ELEMENTS) but Element() lookup fails because _pdt dict lacks the entry. Added elements: Rf(104), Db(105), Sg(106), Bh(107), Hs(108), Mt(109), Ds(110), Rg(111), Cn(112), Nh(113), Fl(114), Mc(115), Lv(116), Ts(117), Og(118). Mass numbers from most stable isotopes.  ## Summary by CodeRabbit * **New Features** * Expanded the periodic table data to include elements 104–118, completing the later portion of the table. * Added the missing entry closure before the new elements to ensure the data is properly formatted.  Co-authored-by: guomingyu <guomingyu@users.noreply.github.com>

master

4 days ago

fix: add missing running_md.log fixture for no-force ABACUS test The test fixture at tests/abacus.md.noforce/ was missing running_md.log, which dpdata/formats/abacus/md.py:get_frame unconditionally opens. This caused FileNotFoundError when running test_no_force_columns. Add a minimal 2-frame running_md.log with matching energies to fix the test. Addresses CodeRabbit review comment on PR #1008

abhi-0203:fix/abacus-md-force-labels

5 days ago

Latest Branches

+1%

fix(deepmd/hdf5): skip empty optional frame arrays on dump (#996)#1011

13 hours ago

642a745

CedricConday:fix/hdf5-empty-optional-arrays

+3%

fix: build pymatgen Molecule species from atom_types order (#994)#1010

3 days ago

7e073a4

CedricConday:fix/pymatgen-molecule-species-order-994

feat(extxyz): add unit convert and tag synonym matching#678

4 days ago

a52b3e9

SchrodingersCattt:devel

Home Terms Privacy Docs