Source code for BigDFT.Datasets

"""Calculation datasets.

This module deals with the handling of series of calculations.
Classes and functions of this module are meant to simplify the approach to
ensemble calculations with the code, and to deal with parallel executions of
multiple instances of the code.
"""


from BigDFT.Calculators import Runner


[docs]def name_from_id(id): """Hash the id into a run name Construct the name of the run from the id dictionary Args: id (dict): id associated to the run Returns: str: name of the run associated to the dictionary ``id`` """ return ','.join([':'.join([k, str(id[k])]) for k in sorted(id)])
# keys = list(id.keys()) # keys.sort() # name = '' # for k in keys: # name += k + '=' + str(id[k]) + ',' # ','.join([k+'='+str(i)) # return name.rstrip(',')
[docs]def names_from_id(id): """ Hash the id into a list of run names to search with the function id_in_names and add the separator ',' to have the proper value of a key (to avoid 0.3 in 0.39) """ if id is None: return [''] else: return [name_from_id({k: v})+',' for k, v in id.items()]
[docs]class Dataset(Runner): """A set of calculations. Such class contains the various instances of a set of calculations with the code. The different calculations are labelled by parameter values and information that might then be retrieved for inspection and plotting. Args: label (str): The label of the dataset. It will be needed to identify the instance for example in plot titles or in the running directory. run_dir (str): path of the directory where the runs will be performed. input (dict): Inputfile to be used for the runs as default, can be overridden by the specific inputs of the run """ def __init__(self, label='BigDFT dataset', run_dir='runs', **kwargs): """ Set the dataset ready for appending new runs """ from copy import deepcopy newkwargs = deepcopy(kwargs) Runner.__init__(self, label=label, run_dir=run_dir, **newkwargs) self.runs = [] """ List of the runs which have to be treated by the dataset these runs contain the input parameter to be passed to the various runners. """ self.calculators = [] """ Calculators which will be used by the run method, useful to gather the inputs in the case of a multiple run. """ self.results = {} """ Set of the results of each of the runs. The set is not ordered as the runs may be executed asynchronously. """ self.ids = [] """ List of run ids, to be used in order to classify and fetch the results """ self.names = [] """ List of run names, needed for distinguishing the logfiles and input files. Each name should be unique to correctly identify a run. """ self._post_processing_function = None
[docs] def append_run(self, id, runner, **kwargs): """Add a run into the dataset. Append to the list of runs to be performed the corresponding runner and the arguments which are associated to it. Args: id (dict): the id of the run, useful to identify the run in the dataset. It has to be a dictionary as it may contain different keyword. For example a run might be classified as ``id = {'hgrid':0.35, 'crmult': 5}``. runner (Runner): the runner class to which the remaining keyword arguments will be passed at the input. Raises: ValueError: if the provided id is identical to another previously appended run. Todo: include id in the runs specification """ from copy import deepcopy name = name_from_id(id) if name in self.names: raise ValueError('The run id', name, ' is already provided, modify the run id.') self.names.append(name) # create the input file for the run, combining run_dict and input inp_to_append = deepcopy(self._global_options) inp_to_append.update(deepcopy(kwargs)) # get the number of this run irun = len(self.runs) # append it to the runs list self.runs.append(inp_to_append) # append id and name self.ids.append(id) # search if the calculator already exists found = False for calc in self.calculators: if calc['calc'] == runner: calc['runs'].append(irun) found = True break if not found: self.calculators.append({'calc': runner, 'runs': [irun]})
[docs] def process_run(self): """ Run the dataset, by performing explicit run of each of the item of the runs_list. """ self._run_the_calculations() return {}
def _run_the_calculations(self, selection=None): from copy import deepcopy for c in self.calculators: calc = c['calc'] # we must here differentiate between a taskgroup run and a # separate run for r in c['runs']: if selection is not None and r not in selection: continue inp = self.runs[r] name = self.names[r] local_inp = {k: v for k, v in self.local_options.items() if k in inp} if len(local_inp) == 0: tmp_inp = inp else: tmp_inp = deepcopy(inp) tmp_inp.update(local_inp) self.results[r] = calc.run(name=name, **tmp_inp)
[docs] def set_postprocessing_function(self, func): """Set the callback of run. Calls the function ``func`` after having performed the appended runs. Args: func (func): function that process the `inputs` `results` and returns the value of the `run` method of the dataset. The function is called as ``func(self)``. """ self._post_processing_function = func
[docs] def post_processing(self, **kwargs): """ Calls the Dataset function with the results of the runs as arguments """ if self._post_processing_function is not None: return self._post_processing_function(self) else: return self.results
[docs] def fetch_results(self, id=None, attribute=None, run_if_not_present=True): """Retrieve some attribute from some of the results. Selects out of the results the objects which have in their ``id`` at least the dictionary specified as input. May return an attribute of each result if needed. Args: id (dict): dictionary of the retrieved id. Return a list of the runs that have the ``id`` argument inside the provided ``id`` in the order provided by :py:meth:`append_run`. If absent, then the entire list of runs is returned. attribute (str): if present, provide the attribute of each of the results instead of the result object run_if_not_present (bool): If the run has not yet been performed in the dataset then perform it. Example: >>> study=Dataset() >>> study.append_run(id={'cr': 3}, input={'dft':{'rmult':[3,8]}}) >>> study.append_run(id={'cr': 4}, input={'dft':{'rmult':[4,8]}}) >>> study.append_run(id={'cr': 3, 'h': 0.5}, >>> input={'dft':{'hgrids': 0.5, 'rmult':[4,8]}}) >>> #append other runs if needed >>> #run the calculations (optional if run_if_not_present=True) >>> study.run() >>> # returns a list of the energies of first and the third result >>> # in this example >>> data=study.fetch_results(id={'cr': 3}, attribute='energy') """ if id is None: selection_to_run = list(range(len(self.names))) fetch_indices = selection_to_run else: names = names_from_id(id) fetch_indices = [] selection_to_run = [] for irun, name in enumerate(self.names): # add the separator ',' to have the proper value of a key # (to avoid 0.3 in 0.39) if not all([(n in name+',') for n in names]): continue if run_if_not_present and irun not in self.results: selection_to_run.append(irun) fetch_indices.append(irun) if len(selection_to_run) > 0: self._run_the_calculations(selection=selection_to_run) data = [] for irun in fetch_indices: r = self.results[irun] data.append(r if attribute is None else getattr(r, attribute)) return data
[docs] def seek_convergence(self, rtol=1.e-5, atol=1.e-8, selection=None, **kwargs): """ Search for the first result of the dataset which matches the provided tolerance parameter. The results are in dataset order (provided by the :py:meth:`append_run` method) if `selection` is not specified. Employs the numpy :py:meth:`allclose` method for comparison. Args: rtol (float): relative tolerance parameter atol (float): absolute tolerance parameter selection (list): list of the id of the runs in which to perform the convergence search. Each id should be unique in the dataset. **kwargs: arguments to be passed to the :py:meth:`fetch_results` method. Returns: id,result (tuple): the id of the last run which matches the convergence, together with the result, if convergence is reached. Raises: LookupError: if the parameter for convergence were not found. The dataset has to be enriched or the convergence parameters loosened. """ from numpy import allclose from futile.Utils import write to_get = self.ids if selection is None else selection id_ref = to_get[0] write('Fetching results for id "', id_ref, '"') ref = self.fetch_results(id=id_ref, **kwargs) ref = ref[0] for id in to_get[1:]: write('Fetching results for id "', id, '"') val = self.fetch_results(id=id, **kwargs) val = val[0] if allclose(ref, val, rtol=rtol, atol=atol): res = self.fetch_results(id=id_ref) label = self.get_global_option('label') write('Convergence reached in Dataset "' + label+'" for id "', id_ref, '"') return (id_ref, res[0]) ref = val id_ref = id raise LookupError('Convergence not reached, enlarge the dataset' ' or change tolerance values')
[docs] def get_times(self): from os import path as p from futile import Time as T time_files = {} run_dir = self.get_global_option('run_dir') time_files = [p.join(run_dir, self.fetch_results(id=self.ids[c], attribute='data_directory')[0], 'time-' + self.names[c] + '.yaml' ) for c in self.results ] return T.TimeData(*time_files)
[docs] def wait(self): from IPython.display import display, clear_output from aiida.orm import load_node running = len(self.results) while(running != 0): import time time.sleep(1) running = len(self.results) for c in self.results: pk = self.results[c]['node'].pk node = load_node(pk) if(node.is_finished): running -= 1 # print(node.is_finished_ok) clear_output(wait=True) display(str(running)+" processes still running")
[docs] def get_logfiles(self): logfiles = {} for c in self.results: try: logfiles[c] = self.calculators[0]['calc'].get_logs( self.results[c]['node'].pk, self.names[c]) except ValueError: logfiles[c] = self.results[c] print("no logfile for " + str(c)) return logfiles
[docs]def combine_datasets(*args): """ Define a new instance of the dataset class that should provide as a result a list of the runs of the datasets """ full = Dataset(label='combined_dataset') # append the runs or each dataset for dt in args: for irun, runs in enumerate(dt.runs): calc = dt.get_runner(irun) id, dt.get_id(irun) full.append_run(id, calc, **runs) full.set_postprocessing_function(_combined_postprocessing_functions)
def _combined_postprocessing_functions(runs, results, **kwargs): pass