Source code for asaplib.hypers.hyper_soap

"""
tools for generating hyperparameters for SOAP descriptors
"""
import json

from .univeral_length_scales import uni_length_scales, system_pair_bond_lengths, round_sigfigs
from ..io import NpEncoder

"""
Automatically generate the hyperparameters of SOAP descriptors for arbitrary elements and combinations.

## Heuristics:
  * Get the length scales of the system from
    * maximum bond length (from equilibrium bond length in lowest energy 2D or 3D structure)
    * minimal bond length (from shortest bond length of any equilibrium structure, including dimer)
  * Apply a scaling for these length scales
    * largest soap cutoff = maximum bond length * 1.3
    * smallest soap cutoff = minimal bond length * 1.3
    * Add other cutoffs in between if requires more sets of SOAP descriptors 
    * The atom sigma is the `cutoff / 8`, divided by an optional `sharpness` factor

## Example

The command 
gen_default_soap_hyperparameters([5,32], soap_n=6, soap_l=6, multisoap=2, sharpness=1.0, scalerange=1.0, verbose=False)
will return length scales needed to define the SOAP descriptors for 
a system with boron (5) and germanium (32).
"""

[docs]def universal_soap_hyper(global_species, fsoap_param, dump=True): if fsoap_param == 'smart' or fsoap_param == 'Smart' or fsoap_param == 'SMART': soap_js = gen_default_soap_hyperparameters(list(global_species), multisoap=2, scalerange=1.2, soap_n=8, soap_l=4, sharpness=1.0) elif fsoap_param == 'minimal' or fsoap_param == 'Minimal' or fsoap_param == 'MINIMAL': soap_js = gen_default_soap_hyperparameters(list(global_species), multisoap=1, scalerange=0.85, soap_n=4, soap_l=3, sharpness=1.0) elif fsoap_param == 'longrange' or fsoap_param == 'Longrange' or fsoap_param == 'LONGRANGE': soap_js = gen_default_soap_hyperparameters(list(global_species), multisoap=2, scalerange=1.8, soap_n=8, soap_l=4, sharpness=1.2) else: raise IOError('Did not specify soap parameters. You can use [smart/minimal/longrange].') print(soap_js) if dump: with open('smart-soap-parameters', 'w') as jd: json.dump(soap_js, jd, cls=NpEncoder) return soap_js
[docs]def gen_default_soap_hyperparameters(Zs, multisoap=2, scalerange=1.0, soap_n=8, soap_l=4, sharpness=1.0, verbose=False): """ Parameters ---------- Zs : array-like, list of atomic species soap_n, soap_l: soap parameters multisoap: type=int, How many set of SOAP descriptors do you want to use? default=2 sharpness: type=float, sharpness factor for atom_gaussian_width, scaled to heuristic for GAP, default=1.0 range: type=float, the range of the SOAP cutoffs, scaled to heuristic for GAP, default=1.0 verbose: type=bool, default=False, more descriptions of what has been done. """ # check if the element is in the look up table # print(type(Zs)) for Z in Zs: if str(Z) not in uni_length_scales: raise RuntimeError("key Z {} not present in length_scales table".format(Z)) shortest_bond, longest_bond = system_pair_bond_lengths(Zs, uni_length_scales) if verbose: print(Zs, "range of bond lengths", shortest_bond, longest_bond) # factor between shortest bond and shortest cutoff threshold factor_inner = 1.3 * scalerange rcut_min = max(2.0, factor_inner * shortest_bond) # factor between longest bond and longest cutoff threshold factor_outer = 1.3 * scalerange rcut_max = max(rcut_min * 1.2, factor_outer * longest_bond) if verbose: print("Considering minimum and maximum cutoff", rcut_min, rcut_max) hypers = {} num_soap = 1 # first soap cutoff is just the rcut_max r_cut = rcut_max g_width = r_cut / 8.0 / sharpness hypers['soap' + str(num_soap)] = {'type': 'SOAP', 'species': Zs, 'cutoff': float(round_sigfigs(r_cut, 2)), 'n': soap_n, 'l': soap_l, 'atom_gaussian_width': float(round_sigfigs(g_width, 2))} if multisoap >= 2: # ratio between subsequent rcut values rcut_ratio = (rcut_max / rcut_min) ** (1. / (multisoap - 1)) while r_cut >= rcut_max * 0.99: num_soap += 1 r_cut /= rcut_ratio g_width = r_cut / 8.0 / sharpness hypers['soap' + str(num_soap)] = {'type': 'SOAP', "species": Zs, 'cutoff': float(round_sigfigs(r_cut, 2)), 'n': soap_n, 'l': soap_l, 'atom_gaussian_width': float(round_sigfigs(g_width, 2))} return hypers