deepmd/npy/mixed format

deepmd/npy/mixed format#

Aliases#

deepmd/npy/mixed

Implementation: DeePMDMixedFormat

Overview#

Mixed-type NumPy dataset for DeePMD-kit.

DeePMD-kit is a deep learning package for training interatomic potentials. Unlike regular deepmd/npy, this layout can combine frames that have the same atom count but different formulas. Per-frame real atom types keep each composition recoverable for models that use type embeddings. Optional atom-count padding can reduce the number of output groups when a dpdata.MultiSystems contains many system sizes.

Examples

Dump a MultiSystems into a mixed type numpy directory:

>>> import dpdata
>>> dpdata.MultiSystems(*systems).to_deepmd_npy_mixed("mixed_dir")

Dump with atom_numb_pad to reduce the number of subdirectories. Systems are padded with virtual atoms (type -1) so that atom counts are rounded up to the nearest multiple of the given number:

>>> dpdata.MultiSystems(*systems).to_deepmd_npy_mixed("mixed_dir", atom_numb_pad=8)

Load a mixed type data into a MultiSystems:

>>> import dpdata
>>> dpdata.MultiSystems().load_systems_from_file("mixed_dir", fmt="deepmd/npy/mixed")

Quick examples#

The examples use the preferred alias deepmd/npy/mixed; any alias listed above is equivalent.

import dpdata

# Multiple compositions or calculation directories
systems = dpdata.MultiSystems.from_file("input_directory", fmt="deepmd/npy/mixed")

# Load or construct the system to write
system = dpdata.System("input_file")

# Write geometry-only data
system.to("deepmd/npy/mixed", "output_path")

# Write multiple systems
systems.to("deepmd/npy/mixed", "output_path")

Conversions#

Convert from System to this format#

dpdata.System.to(fmt: Literal['deepmd/npy/mixed'], file_name, set_size: 'int' = 2000, prec=<class 'numpy.float64'>, **kwargs)
dpdata.System.to_deepmd_npy_mixed(file_name, set_size: 'int' = 2000, prec=<class 'numpy.float64'>, **kwargs)

Dump the system in deepmd mixed type format (numpy binary) to folder.

The frames were already split to different systems, so these frames can be dumped to one single subfolders

named as folder/set.000, containing less than set_size frames.

Parameters:
file_namestr

The output folder

set_sizeint, default=2000

set size

prec{numpy.float32, numpy.float64}

The floating point precision of the compressed data

**kwargsdict

other parameters

Convert from LabeledSystem to this format#

dpdata.LabeledSystem.to(fmt: Literal['deepmd/npy/mixed'], file_name, set_size: 'int' = 2000, prec=<class 'numpy.float64'>, **kwargs)
dpdata.LabeledSystem.to_deepmd_npy_mixed(file_name, set_size: 'int' = 2000, prec=<class 'numpy.float64'>, **kwargs)

Dump the system in deepmd mixed type format (numpy binary) to folder.

The frames were already split to different systems, so these frames can be dumped to one single subfolders

named as folder/set.000, containing less than set_size frames.

Parameters:
file_namestr

The output folder

set_sizeint, default=2000

set size

prec{numpy.float32, numpy.float64}

The floating point precision of the compressed data

**kwargsdict

other parameters

Convert from this format to MultiSystems#

dpdata.MultiSystems.from_deepmd_npy_mixed(directory, **kwargs) → dpdata.system.MultiSystems

Find mixed-type DeePMD NumPy systems below a directory.

Parameters:
directorystr or os.PathLike

Root directory containing one or more mixed datasets.

**kwargsdict

Additional format arguments forwarded when each dataset is read.

Returns:
MultiSystems

converted system

Convert from MultiSystems to this format#

dpdata.MultiSystems.to(fmt: Literal['deepmd/npy/mixed'], directory, **kwargs) → dpdata.system.MultiSystems
dpdata.MultiSystems.to_deepmd_npy_mixed(directory, **kwargs) → dpdata.system.MultiSystems

Convert MultiSystems to this format.

Parameters:
directorystr

directory to save systems

Returns:
MultiSystems

this system