Ctrl+K

wfl.fit.gap package

wfl.fit.gap package#

Submodules#

wfl.fit.gap.glue_2b module#

wfl.fit.gap.glue_2b.construct_glue_2b(inputs, energy_info_key, cutoff=None, do_e0=True, filename=None)#

Construct a glue potential from dimer data, return the isolated atom energies as well

Parameters

  • inputs (ConfigSet) –

  • energy_info_key (dict_key) – info key for energy, will not use the calculator’s energy directly

  • cutoff (float / dict, default None) – cutoff to set for all pairs, can be a float for uniform cutoff, or a dict with keys for pairs in ‘Z1_Z2’ format where Z1,Z2 is sorted. Dict with key missing for pair having data defaults to maximum separation possible with the given data

  • do_e0 (bool, default True) – calculate e0 values, only implemented to be done from homonuclear pairs

  • filename (path_like, default None) – if given then xml file is saved to this path

Returns

  • xml_str (str) – xml param str generated

  • e0_by_symbol (dict, None) – dict if do_e0, otherwise None

wfl.fit.gap.multistage module#

wfl.fit.gap.multistage.GAP_xml_modify_label(GAPfile, new_label=None)#

fix internal GAP name and return updated label

Parameters

  • GAPfile (str) – gap xml file

  • new_label (str / None, default None) – new label, if None then old label is returned

Returns

label – new label or old if not changed

Return type

str

wfl.fit.gap.multistage.fit(fitting_configs, GAP_name, params, ref_property_prefix='REF_', seeds=None, skip_if_present=False, run_dir='.', num_committee=0, committee_extra_seeds=None, committee_name_postfix='.committee_', verbose=False, remote_info=None, remote_label=None, wait_for_results=True)#

Fit a GAP iteratively, setting delta from error relative to previous stage

Parameters

  • fitting_configs (ConfigSet) – input fitting configurations

  • GAP_name (str) – name of GAP label, also used as part of xml file

  • params (dict) – parameters controlling each stage of fit, typically read in from YAML file

  • ref_property_prefix (str, default 'REF_') – string prefix added to atoms.info/arrays keys (energy, forces, virial, stress, hessian)

  • seeds (list(int)) – random seeds for each stage of fitting

  • verbose (bool, default False) – verbose output

  • skip_if_present (bool, default False) – skip if final GAP file exists in expected place

  • run_dir (str, default '.') – directory to run fitting in

  • num_committee (int, default 0) – number of models to create as a committee of GAP models refits the last stage a total of this many times

  • committee_extra_seeds (list(int) / None) – random seeds to use for committee of models after 0th

  • committee_name_postfix (str, default ".committee_") – str to add to name of committee models in the format: “{GAP_name}{committee_name_postfix}{num}.xml”

  • remote_info (dict or wfl.autoparallelize.utils.RemoteInfo, or '_IGNORE' or None) – If present and not None and not ‘_IGNORE’, RemoteInfo or dict with kwargs for RemoteInfo constructor which triggers running job in separately queued job on remote machine. If None, will try to use env var WFL_EXPYRE_INFO used (see below). ‘_IGNORE’ is for internal use, to ensure that remotely running job does not itself attempt to spawn another remotely running job.

  • remote_label (str, default None) – label to match in WFL_EXPYRE_INFO

  • wait_for_results (bool, default True) – wait for results of remotely executed job, otherwise return after starting job

  • Variables (Environment) –

  • --------------------- –

  • WFL_EXPYRE_INFO (JSON dict or name of file containing JSON with kwargs for RemoteInfo) – contructor to be used to run fitting in separate queued job

  • WFL_GAP_FIT_OMP_NUM_THREADS (number of threads to set for OpenMP of gap_fit) –

Returns

string name of final GAP xml file, and name of corresponding GAP (xml label)

Return type

final_GAP_file, final_GAP_name

wfl.fit.gap.multistage.max_cutoff(params)#
wfl.fit.gap.multistage.prep_params(Zs, length_scales, GAP_template, spacing=1.5, no_extra_inner=False, no_extra_outer=False, sharpness=1.0)#

prepare parameters for multistage fitting based on YAML template file

Parameters

  • Zs (list(int)) – atomic numbers

  • length_scales (dict(Z: dict('bond_len': bond_len))) – dict with bond lengths for each atomic number Z

  • GAP_template (dict or str) – dict with settings template (to get descriptors auto-filled from universal SOAP length scales) or filename with input template YAML

wfl.fit.gap.relocate module#

wfl.fit.gap.relocate.gap_relocate(old_file, new_file, extra_filename_glob=None, delete_old=False)#
relocate a GAP from one file to another, changing tags and attributes as needed to

be consistent with new name, and renaming additional files (referred to in the xml or just with same leading part of name

Parameters

  • old_file (str) – original GAP xml filename

  • new_file (str) – new GAP xml filename

  • extra_filename_glob (str, default '*') – glob to add to end of old_file to generate list of files that need to be renamed

  • delete_old (bool, default False) – delete the old_file and all the other related files

wfl.fit.gap.simple module#

wfl.fit.gap.simple.dict_to_gap_fit_string(param_dict)#

converts dictionary with gap_fit parameters to string for calling gap_fit.

  • booleans -> “T”/”F”

  • lists -> { v1 v2 v3 … }

  • {‘key1’:[v1, v2, v3], ‘key2’:[v1, v2, v3]}

    -> ‘{key1:v1:v2:v3:key2:v1:v2:v3}’

  • strings with spaces get enclosed in quotes

  • otherwise {key:val} -> key=val

  • asserts that mandatory parameters are given

  • descriptors are passed in pram_dict[‘_gap’], which is a list of dictionaries, one dictionary per descriptor

wfl.fit.gap.simple.run_gap_fit(fitting_configs, fitting_dict, stdout_file, gap_fit_command=None, verbose=True, do_fit=True, remote_info=None, remote_label=None, skip_if_present=False, **kwargs)#

Runs gap_fit

Parameters

  • fitting_configs (ConfigSet) – set of configurations to fit

  • fitting_dict (dict) – dict of keys to turn into command line for gap_fit

  • stdout_file (str / Path) – filename to pass standard output to

  • gap_fit_command (str, default "gap_fit") – executable for gap_fit. Alternatively set by WFL_GAP_FIT_COMMAND environment variable, which overrides this gap_fit_command argument.

  • verbose (bool, default True) –

  • do_fit (bool, default True) – carry out the fit, otherwise only print fitting command

  • remote_info (dict or wfl.autoparallelize.utils.RemoteInfo, or '_IGNORE' or None) – If present and not None and not ‘_IGNORE’, RemoteInfo or dict with kwargs for RemoteInfo constructor which triggers running job in separately queued job on remote machine. If None, will try to use env var WFL_EXPYRE_INFO used (see below). ‘_IGNORE’ is for internal use, to ensure that remotely running job does not itself attempt to spawn another remotely running job.

  • remote_label (str, default None) – remote label to patch in WFL_EXPYRE_INFO

  • skip_if_present (bool, default False) – skip fitting if output is already present

  • kwargs – any key:val pair will be added to the fitting str

  • Variables (Environment) –

  • --------------------- –

  • WFL_EXPYRE_INFO (JSON dict or name of file containing JSON with kwargs for RemoteInfo) – contructor to be used to run fitting in separate queued job

  • WFL_GAP_FIT_OMP_NUM_THREADS (number of threads to set for OpenMP of gap_fit) –

  • WFL_GAP_FIT_COMMAND (executable for gap_fit. Overrides the gap_fit_command argument.) –

Module contents#

Contents