def RDT_grabby(expname, rank, output_fname=None, plot_bairros=True) :
calib_folder = calib_base + expname +"/"
if not output_fname:
output_fname = calib_folder + "rank{}_rdt".format(rank)
LL_fname = calib_folder + "_plots/LL_all.csv"
LL_df = pd.read_csv(LL_fname)
LL_df.sort_values(by='total', ascending=False, inplace=True)
LL_df.reset_index(inplace=True)
sample = LL_df.loc[rank, 'sample']
iteration = LL_df.loc[rank, 'iteration']
start_date = "2009-01-01"
am = AnalyzeManager()
am.add_analyzer(PrevAnalyzer(start_date=start_date,
save_file=output_fname,
cait_output_mode=True,
plot_bairros=plot_bairros))
with open(calib_folder+"iter{}/IterationState.json".format(iteration)) as fin:
iteration_state = json.loads(fin.read())
siminfo = OrderedDict(iteration_state['simulations'])
for item in list(siminfo.items()) :
if item[1]['__sample_index__'] == sample :
simid = item[0]
# print("Sim ID: ",simid)
am.add_simulation(simid)
am.analyze()
def incidence_grabby(expname, hfca, rank, output_fname=None) :
calib_folder = calib_base + expname +"/"
if not output_fname:
output_fname = calib_folder + "rank{}_cases".format(rank)
LL_fname = calib_folder + "_plots/LL_all.csv"
LL_df = pd.read_csv(LL_fname)
LL_df.sort_values(by='total', ascending=False, inplace=True)
LL_df.reset_index(inplace=True)
sample = LL_df.loc[rank, 'sample']
iteration = LL_df.loc[rank, 'iteration']
am = AnalyzeManager()
# am.add_analyzer(IncidencePlotter(GriddedCalibSite(hfca),save_file=output_fname))
am.add_analyzer(IncidencePlotter(hfca, save_file=output_fname, save_csv=True))
with open(calib_folder+"iter{}/IterationState.json".format(iteration)) as fin:
iteration_state = json.loads(fin.read())
siminfo = OrderedDict(iteration_state['simulations'])
for item in list(siminfo.items()) :
if item[1]['__sample_index__'] == sample :
simid = item[0]
am.add_simulation(simid)
am.analyze()
def _analyze(self, experiment, analyzers, points_ran):
"""
This method is the in-common route for Resamplers to analyze simulations for liklihood.
:param experiment: the experiment to analyze, should be from self._run()
:param points_ran: Points objects that were just _run()
:return: The supplied points_ran with their .likelihood attribute set, AND the direct results of the analyzer
as a list.
"""
am = AnalyzeManager(analyzers=analyzers, exp_list=experiment)
am.analyze()
# The provided likelihood analyzer MUST set self.result to be a list of Point objects
# with the .likelihood attribute set to the likelihood value in its .finalize() method.
results = am.analyzers[0].result.tolist()
for i in range(len(results)):
# Add the likelihood
points_ran[i].likelihood = results[i]
# verify that the returned points all have a likelihood attribute set
likelihoods_are_missing = True in {
point.likelihood is None
for point in points_ran
}
if likelihoods_are_missing:
raise Exception(
'At least one Point object returned by the provided analyzer does not have '
'its .likelihood attribute set.')
return points_ran, results
def plot_vectors(exp_id, sample, save_file=None):
am = AnalyzeManager()
am.add_experiment(retrieve_experiment(exp_id))
am.add_analyzer(
VectorSpeciesReportAnalyzer(sample,
save_file=save_file,
channel='Daily HBR'))
am.analyze()
def analyze(args, unknownArgs, builtinAnalyzers):
# validate parameters
if args.config_name is None:
logger.error('Please provide Analyzer (-a or --config_name).')
exit()
# Retrieve what we need
itemids = args.itemids
batch_name = args.batch_name
# collect all experiments and simulations
exp_dict, sim_dict = collect_experiments_simulations(itemids)
# consider batch existing case
exp_dict, sim_dict = consolidate_experiments_with_options(exp_dict, sim_dict, batch_name)
# check status for each experiment
if not args.force:
check_status(exp_dict.values())
# collect all analyzers
analyzers = collect_analyzers(args, builtinAnalyzers)
if not exp_dict and not sim_dict:
# No experiment specified -> using latest experiment
latest = DataStore.get_most_recent_experiment()
exp_dict[latest.exp_id] = latest
# create instance of AnalyzeManager
analyzeManager = AnalyzeManager(exp_list=exp_dict.values(), sim_list=sim_dict.values(), analyzers=analyzers)
exp_ids_to_be_saved = list(set(exp_dict.keys()) - set(analyzeManager.experiments_simulations.keys()))
exp_to_be_saved = [exp_dict[exp_id] for exp_id in exp_ids_to_be_saved]
# if batch name exists, always save experiments
if batch_name:
# save/create batch
save_batch(batch_name, exp_to_be_saved, sim_dict.values())
# Only create a batch if we pass more than one experiment or simulation in total
elif len(exp_dict) + len(sim_dict) > 1:
# check if there is any existing batch containing the same experiments
batch_existing = check_existing_batch(exp_dict, sim_dict)
if batch_existing is None:
# save/create batch
save_batch(batch_name, exp_to_be_saved, sim_dict.values())
else:
# display the existing batch
logger.info('\nBatch: %s (id=%s)' % (batch_existing.name, batch_existing.id))
# start to analyze
analyzeManager.analyze()
# remove empty batches
clean_batch()
return analyzeManager
def analyze_iteration(self):
"""
Analyze the output of completed simulations by using the relevant analyzers by site.
Cache the results that are returned by those analyzers.
"""
if self.results:
logger.info('Reloading results from cached iteration state.')
return self.results['total']
if not self.exp_manager:
self.exp_manager = ExperimentManagerFactory.from_experiment(
self.experiment_id)
from simtools.Analysis.BaseAnalyzers.BaseAnalyzer import BaseAnalyzer
from simtools.Analysis.AnalyzeManager import AnalyzeManager as am
if all(isinstance(a, BaseAnalyzer) for a in self.analyzer_list):
analyzerManager = am(exp_list=self.exp_manager.experiment,
analyzers=self.analyzer_list,
working_dir=self.iteration_directory,
verbose=True)
else:
analyzerManager = AnalyzeManager(
exp_list=self.exp_manager.experiment,
analyzers=self.analyzer_list,
working_dir=self.iteration_directory)
analyzerManager.analyze()
# Ask the analyzers to cache themselves
cached_analyses = {
a.uid if not callable(a.uid) else a.uid(): a.cache()
for a in analyzerManager.analyzers
}
logger.debug(cached_analyses)
# Get the results from the analyzers and ask the next point how it wants to cache them
results = pd.DataFrame({
a.uid if not callable(a.uid) else a.uid(): a.result
for a in analyzerManager.analyzers
})
cached_results = self.next_point_algo.get_results_to_cache(results)
# Store the analyzers and results in the iteration state
self.analyzers = cached_analyses
self.results = cached_results
# Set those results in the next point algorithm
self.next_point_algo.set_results_for_iteration(self.iteration, results)
# Update the summary table and all the results
self.all_results, self.summary_table = self.next_point_algo.update_summary_table(
self, self.all_results)
logger.info(self.summary_table)
def vector(expname, calib_stage, rank) :
calib_folder = calib_base + expname +"/"
output_fname = calib_folder + "rank{}_vectors".format(rank)
LL_fname = calib_folder + "_plots/LL_all.csv"
LL_df = pd.read_csv(LL_fname)
LL_df.sort_values(by='total', ascending=False, inplace=True)
LL_df.reset_index(inplace=True)
sample = LL_df.loc[rank, 'sample']
iteration = LL_df.loc[rank, 'iteration']
am = AnalyzeManager()
am.add_analyzer(VectorSpeciesReportAnalyzer(save_file=output_fname, channel='Daily HBR'))
with open(calib_folder+"iter{}/IterationState.json".format(iteration)) as fin:
iteration_state = json.loads(fin.read())
siminfo = OrderedDict(iteration_state['simulations'])
for item in list(siminfo.items()) :
if item[1]['__sample_index__'] == sample :
simid = item[0]
am.add_simulation(simid)
am.analyze()
def stdout(args, unknownArgs):
exp_manager = reload_experiment(args)
states, msgs = exp_manager.get_simulation_status()
if not exp_manager.status_succeeded(states):
logger.warning(
'WARNING: not all jobs have finished successfully yet...')
found = False
for sim_id, state in states.items():
if (state is SimulationState.Succeeded and args.succeeded) or\
(state is SimulationState.Failed and args.failed) or \
(not args.succeeded and not args.failed):
found = True
break
if not found:
print("No simulations found...")
else:
am = AnalyzeManager(exp_list=[exp_manager.experiment],
analyzers=StdoutAnalyzer([sim_id], args.error),
force_analyze=True,
verbose=False)
am.analyze()
selected = [
p.selected_data[id(self)] for p in parsers.values()
if id(self) in p.selected_data
]
self.data = pd.concat(selected)
def finalize(self):
import seaborn as sns
sns.set_style("darkgrid")
fig = plt.figure(figsize=(10, 6))
ax = fig.gca()
for a, adf in self.data.groupby('sim_id'):
for s, sdf in adf.groupby('species'):
ax.plot(sdf['date'], sdf[self.channel], label=s)
ax.legend()
plt.ylabel(self.channel)
if self.save_file:
plt.savefig(self.save_file + ".png")
plt.savefig(self.save_file + ".pdf")
else:
plt.show()
if __name__ == '__main__':
am = AnalyzeManager()
am.add_analyzer(VectorSpeciesReportAnalyzer())
am.add_simulation('4047a20f-b33d-e811-a2bf-c4346bcb7274')
am.analyze()
def plot_RDT(exp_id, sample, save_file=None, **kwargs):
am = AnalyzeManager()
am.add_experiment(retrieve_experiment(exp_id))
am.add_analyzer(
prevalence_plot_analyzer(catch, sample, save_file=save_file, **kwargs))
am.analyze()
def catalyst(args, unknownArgs):
"""
Catalyst run-and-analyze process as ported from the test team.
Programmatic-only arguments:
args.mode : used by FidelityReportExperimentDefinition, default: 'prod'
args.report_label : attached to the experiment name
args.debug : True/False, passed into FidelityReportAnalyzer, default: False
:param args:
:param unknownArgs:
:return:
"""
from dtk.utils.builders.sweep import GenericSweepBuilder
from catalyst_report.fidelity_report_analyzer import FidelityReportAnalyzer
from catalyst_report.fidelity_report_experiment_definition import FidelityReportExperimentDefinition
import catalyst_report.utils as catalyst_utils
from simtools.Analysis.AnalyzeManager import AnalyzeManager
# we're going to do a dtk run, then a set-piece analysis. But first we need to do some overrides
# to get the run part to do the desired parameter sweep.
mod = args.loaded_module
# when run with 'dtk catalyst', run_sim_args['exp_name'] will have additional information appended.
mod.run_sim_args[
'exp_name'] = mod.run_sim_args['exp_name'] + '-development'
# lining up the arguments expected by FidelityReportExperimentDefinition
args.sweep = args.sweep_method
# hidden, programmatic arguments
args.mode = args.mode if hasattr(args, 'mode') else 'prod'
args.report_label = args.report_label if hasattr(args,
'report_label') else None
args.debug = args.debug if hasattr(args, 'debug') else False
# determine which report is being asked for. If not specified, default to what the config.json file says
# ck4, this should go somewhere else, on a Config object of some sort? (prob not the builder, though)
report_type_mapping = {
'DENGUE_SIM': 'dengue',
'GENERIC_SIM': 'generic',
'HIV_SIM': 'hiv',
'MALARIA_SIM': 'malaria',
'POLIO_SIM': 'polio',
'STI_SIM': 'sti',
'TB_SIM': 'tb',
'TYPHOID_SIM': 'typhoid',
'VECTOR_SIM': 'generic'
}
if args.report_type:
report_type = args.report_type
else:
sim_type = mod.run_sim_args['config_builder'].config['parameters'][
'Simulation_Type']
report_type = report_type_mapping.get(sim_type, None)
if not report_type:
raise KeyError(
'Default report type could not be determined for sim_type: %s. Report type must be specified'
' via -r flag.' % sim_type)
# Create and set a builder to sweep over population scaling or model timestep
reports = catalyst_utils.load_report_definitions(
definitions_filename=args.report_definitions)
if report_type in reports:
args.report_channel_list = reports[report_type]['inset_channel_names']
else:
raise Exception('Invalid report: %s. Available reports: %s' %
(report_type, sorted(reports.keys())))
catalyst_config = catalyst_utils.load_sweep_configs(
sweep_type=args.sweep_type, config_filename=args.sweep_definitions)
defn = FidelityReportExperimentDefinition(catalyst_config, args)
# redefine the experiment name so it doesn't conflict with the likely follow-up non-catalyst experiment
mod.run_sim_args['exp_name'] = 'Catalyst-' + mod.run_sim_args['exp_name']
# define the sweep to perform
sweep_dict = {
'Run_Number': range(1,
int(defn['nruns']) + 1),
defn['sweep_param']: defn['sweep_values']
}
mod.run_sim_args['exp_builder'] = GenericSweepBuilder.from_dict(sweep_dict)
# overwrite spatial output channels to those used in the catalyst report
spatial_channel_names = defn['spatial_channel_names']
if len(spatial_channel_names) > 0:
mod.run_sim_args['config_builder'].enable('Spatial_Output')
mod.run_sim_args['config_builder'].params[
'Spatial_Output_Channels'] = spatial_channel_names
else:
mod.run_sim_args['config_builder'].disable('Spatial_Output')
mod.run_sim_args['config_builder'].params[
'Spatial_Output_Channels'] = []
# now run if no preexisting experiment id was provided
if not args.experiment_id:
# we must always block so that we can run the analysis at the end; run and analyze!
args.blocking = True
experiment_manager = run(args, unknownArgs)
experiment = experiment_manager.experiment
print('Done running experiment: %s' % experiment.exp_id)
else:
experiment = retrieve_experiment(args.experiment_id)
# Create an analyze manager
am = AnalyzeManager(exp_list=[experiment], verbose=False)
# Add the TimeSeriesAnalyzer to the manager and do analysis
# ck4, is there a better way to specify the first 4 arguments? The DTKCase from Test-land might be nicer.
# After all, the names COULD be different
analyzer = FidelityReportAnalyzer(
'catalyst_report',
'config.json',
mod.run_sim_args['config_builder'].get_param(
'Demographics_Filenames')[0],
experiment_definition=defn,
experiment_id=experiment.exp_id,
experiment_name=experiment.exp_name,
label=args.report_label,
time_series_step_from=defn['step_from'],
time_series_step_to=defn['step_to'],
time_series_equal_step_count=True,
raw_data=True,
debug=args.debug)
am.add_analyzer(analyzer)
am.analyze()
import webbrowser
webbrowser.open_new("file:///{}".format(
os.path.join(os.getcwd(), "catalyst_report", "summary_report.html")))