Source code for psi4.driver.task_planner

#
# @BEGIN LICENSE
#
# Psi4: an open-source quantum chemistry software package
#
# Copyright (c) 2007-2022 The Psi4 Developers.
#
# The copyrights for code used from other parties are included in
# the corresponding files.
#
# This file is part of Psi4.
#
# Psi4 is free software; you can redistribute it and/or modify
# it under the terms of the GNU Lesser General Public License as published by
# the Free Software Foundation, version 3.
#
# Psi4 is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU Lesser General Public License for more details.
#
# You should have received a copy of the GNU Lesser General Public License along
# with Psi4; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
#
# @END LICENSE
#

__all__ = [
    "expand_cbs_methods",
    "task_planner",
    "TaskComputers",
]

import os
import copy
import logging
from typing import Dict, Tuple, Union

from qcelemental.models import DriverEnum
from psi4.driver import p4util, pp
from psi4.driver.task_base import AtomicComputer
from psi4.driver.driver_findif import FiniteDifferenceComputer
from psi4.driver.driver_nbody import ManyBodyComputer
from psi4.driver.driver_cbs import CompositeComputer, composite_procedures, cbs_text_parser
from psi4.driver.driver_util import negotiate_derivative_type, negotiate_convergence_criterion
from psi4 import core

logger = logging.getLogger(__name__)

TaskComputers = Union[AtomicComputer, CompositeComputer, FiniteDifferenceComputer, ManyBodyComputer]


[docs]def expand_cbs_methods(method: str, basis: str, driver: DriverEnum, **kwargs) -> Tuple[str, str, Dict]: """Sort out the user input method string into recognized fields. Handles cases like: (i) ``"mp2"`` -- passes through; (ii) ``"mp2/cc-pvdz"`` -- broken into method and basis fields; (iii) ``"mp2/cc-pv[d,t]z"`` -- processed into method="cbs" & CBSMetadata spec; (iv) ``method="cbs", cbsmeta=CBSMetadata`` -- passes through. Parameters ---------- method User first argument to driver function. A string hint of the method -- see cases above. basis User basis hint. driver The calling driver function. Note for finite difference that this is the target driver, not the means driver. """ if method == 'cbs' and kwargs.get('cbsmeta', False): return method, basis, kwargs['cbsmeta'] # Expand CBS methods if "/" in method: kwargs["ptype"] = driver cbsmeta = cbs_text_parser(method, **kwargs) # Single call detected if "cbs_metadata" not in cbsmeta: method = cbsmeta["method"] basis = cbsmeta["basis"] else: method = "cbs" else: cbsmeta = {} return method, basis, cbsmeta
[docs]def task_planner(driver: DriverEnum, method: str, molecule: core.Molecule, **kwargs) -> TaskComputers: """Plans a task graph of a complex computation. Canonical Task layering: - ManyBody - BSSE treatment, many-body expansion - FiniteDifference - derivatives through stencils - Composite - basis set extrapolation, focal-point methods - Atomic - analytic single-points Parameters ---------- driver The resulting type of computation: e/g/h. Note for finite difference that this should be the target driver, not the means driver. method A string representation of the method such as "HF" or "B3LYP". Special cases are: "cbs". molecule A Psi4 base molecule to use. kwargs User keyword arguments, often used to configure task computers. Returns ------- Union[AtomicComputer, CompositeComputer, FiniteDifferenceComputer, ManyBodyComputer] A simple (:class:`~psi4.driver.AtomicComputer`) or layered (:class:`~psi4.driver.driver_cbs.CompositeComputer`, :class:`~psi4.driver.driver_findif.FiniteDifferenceComputer`, :class:`~psi4.driver.driver_nbody.ManyBodyComputer`) task object. Layered objects contain many and multiple types of computers in a graph. """ # Only pull the changed options keywords = p4util.prepare_options_for_set_options() keywords["function_kwargs"] = {} if "external_potentials" in kwargs: keywords["function_kwargs"].update({"external_potentials": kwargs.pop("external_potentials")}) # Need to add full path to pcm file if "PCM__PCMSOLVER_PARSED_FNAME" in keywords.keys(): fname = keywords["PCM__PCMSOLVER_PARSED_FNAME"] keywords["PCM__PCMSOLVER_PARSED_FNAME"] = os.path.join(os.getcwd(), fname) # Pull basis out of kwargs, override globals if user specified basis = kwargs.pop("basis", keywords.pop("BASIS", "(auto)")) method = method.lower() # Expand CBS methods method, basis, cbsmeta = expand_cbs_methods(method, basis, driver, **kwargs) if method in composite_procedures: kwargs.update({'cbs_metadata': composite_procedures[method](**kwargs)}) method = 'cbs' pertinent_findif_kwargs = ['findif_irrep', 'findif_stencil_size', 'findif_step_size', 'findif_verbose'] current_findif_kwargs = {kw: kwargs.pop(kw) for kw in pertinent_findif_kwargs if kw in kwargs} # explicit: 'findif_mode' # Build a packet packet = {"molecule": molecule, "driver": driver, "method": method, "basis": basis, "keywords": keywords} # First check for BSSE type if kwargs.get("bsse_type", None) is not None: levels = kwargs.pop('levels', None) plan = ManyBodyComputer(**packet, **kwargs) original_molecule = packet.pop("molecule") # Add tasks for every nbody level requested if levels is None: levels = {plan.max_nbody: method} else: # rearrange bodies in order with supersystem last lest body count fail in organization loop below levels = dict(sorted(levels.items(), key=lambda item: 1000 if item[0] == "supersystem" else item[0])) # We define cp as being a correction to only interaction energies # If only doing cp, we need to ignore any user-specified 1st (monomer) level if 'cp' in kwargs.get("bsse_type", None) and 'nocp' not in kwargs.get("bsse_type", None): if 1 in levels.keys(): removed_level = levels.pop(1) logger.info("NOTE: User specified exclusively 'cp' correction, but provided level 1 details") logger.info(f"NOTE: Removing level {removed_level}") logger.info("NOTE: For total energies, add 'nocp' to bsse_list") # Organize nbody calculations into modelchem levels # * expand keys of `levels` into full lists of nbodies covered. save to plan, resetting max_nbody accordingly # * below, process values of `levels`, which are modelchem strings, into kwargs specs nbodies_per_mc_level = [] prev_body = 0 for nb in levels: nbodies = [] if nb == "supersystem": nbodies.append(nb) elif nb != (prev_body + 1): for m in range(prev_body + 1, nb + 1): nbodies.append(m) else: nbodies.append(nb) nbodies_per_mc_level.append(nbodies) prev_body += 1 plan.max_nbody = max(nb for nb in levels if nb != "supersystem") plan.nbodies_per_mc_level = nbodies_per_mc_level for mc_level_idx, mtd in enumerate(levels.values()): method, basis, cbsmeta = expand_cbs_methods(mtd, basis, driver, cbsmeta=cbsmeta, **kwargs) packet.update({'method': method, 'basis': basis}) # Tell the task builder which level to add a task list for # * see https://github.com/psi4/psi4/pull/1351#issuecomment-549948276 for discussion of where build_tasks logic should live if method == "cbs": # This CompositeComputer is discarded after being used for dermode. simplekwargs = copy.deepcopy(kwargs) simplekwargs.pop('dertype', None) simplecbsmeta = copy.deepcopy(cbsmeta) simplecbsmeta['verbose'] = 0 dummyplan = CompositeComputer(**packet, **simplecbsmeta, molecule=original_molecule, **simplekwargs) methods = [sr.method for sr in dummyplan.task_list] # TODO: pass more info, so fn can use for managed_methods -- ref, qc_module, fc/ae, conv/df dermode = negotiate_derivative_type(driver, methods, kwargs.pop('dertype', None), verbose=1) if dermode[0] == dermode[1]: # analytic logger.info("PLANNING MB(CBS): {mc_level_idx=} {packet=} {cbsmeta=} kw={kwargs}") plan.build_tasks(CompositeComputer, **packet, mc_level_idx=mc_level_idx, **cbsmeta, **kwargs) else: logger.info( f"PLANNING MB(FD(CBS): {mc_level_idx=} {packet=} {cbsmeta=} findif_kw={current_findif_kwargs} kw={kwargs}" ) plan.build_tasks(FiniteDifferenceComputer, **packet, mc_level_idx=mc_level_idx, findif_mode=dermode, computer=CompositeComputer, **cbsmeta, **current_findif_kwargs, **kwargs) else: dermode = negotiate_derivative_type(driver, method, kwargs.pop('dertype', None), verbose=1) if dermode[0] == dermode[1]: # analytic logger.info(f"PLANNING MB: {mc_level_idx=} {packet=}") plan.build_tasks(AtomicComputer, **packet, mc_level_idx=mc_level_idx, **kwargs) else: logger.info( f"PLANNING MB(FD): {mc_level_idx=} {packet=} findif_kw={current_findif_kwargs} kw={kwargs}" ) plan.build_tasks(FiniteDifferenceComputer, **packet, mc_level_idx=mc_level_idx, findif_mode=dermode, **current_findif_kwargs, **kwargs) return plan # Check for CBS elif method == "cbs": kwargs.update(cbsmeta) # This CompositeComputer is discarded after being used for dermode. Could have used directly for analytic except for excess printing with FD simplekwargs = copy.deepcopy(kwargs) simplekwargs['verbose'] = 0 dummyplan = CompositeComputer(**packet, **simplekwargs) methods = [sr.method for sr in dummyplan.task_list] # TODO: pass more info, so fn can use for managed_methods -- ref, qc_module, fc/ae, conv/df dermode = negotiate_derivative_type(driver, methods, kwargs.pop('dertype', None), verbose=1) if dermode[0] == dermode[1]: # analytic logger.info('PLANNING CBS: packet={packet} kw={kwargs}') plan = CompositeComputer(**packet, **kwargs) return plan else: # For FD(CBS(Atomic)), the CompositeComputer above is discarded after being used for dermode. logger.info( f'PLANNING FD(CBS): dermode={dermode} packet={packet} findif_kw={current_findif_kwargs} kw={kwargs}') plan = FiniteDifferenceComputer(**packet, findif_mode=dermode, computer=CompositeComputer, **current_findif_kwargs, **kwargs) return plan # Done with Wrappers -- know we want E, G, or H -- but may still be FD or AtomicComputer else: dermode = negotiate_derivative_type(driver, method, kwargs.pop('dertype', None), verbose=1) convcrit = negotiate_convergence_criterion(dermode, method, return_optstash=False) if dermode[0] == dermode[1]: # analytic logger.info(f'PLANNING Atomic: keywords={keywords}') return AtomicComputer(**packet, **kwargs) else: keywords.update(convcrit) logger.info( f'PLANNING FD: dermode={dermode} keywords={keywords} findif_kw={current_findif_kwargs} kw={kwargs}') return FiniteDifferenceComputer(**packet, findif_mode=dermode, **current_findif_kwargs, **kwargs)