def __init__(self, working_directory, name):
"""interface to the model
interface to the model
Parameters
----------
working_directory : str
working_directory for the model.
name : str
name of the modelInterface. The name should contain only
alpha-numerical characters.
Raises
------
EMAError if name contains non alpha-numerical characters
.. note:: Anything that is relative to `self.working_directory`
should be specified in `model_init` and not
in `__init__`. Otherwise, the code will not work when running
it in parallel. The reason for this is that the working
directory is being updated by parallelEMA to the worker's
separate working directory prior to calling `model_init`.
"""
super(VensimModelStructureInterface, self).__init__(working_directory,
name)
self.outcomes.append(Outcome('TIME' , time=True))
self.outcomes = list(self.outcomes)
self._lookup_uncertainties = []
debug("vensim interface init completed")
def run(self):
"""
read from the queue and write to the log handlers
The logging documentation says logging is thread safe, so there
shouldn't be contention between normal logging (from the main
process) and this thread.
Note that we're using the name of the original logger.
"""
while True:
try:
record = self.queue.get()
# get the logger for this record
if record is None:
ema_logging.debug("none received")
break
logger = logging.getLogger(record.name)
logger.callHandlers(record)
except (KeyboardInterrupt, SystemExit):
raise
except EOFError:
break
except:
traceback.print_exc(file=sys.stderr)
def _store_result(self, case_id, result):
for outcome in self.outcomes:
ema_logging.debug("storing {}".format(outcome))
try:
outcome_res = result[outcome]
except KeyError:
ema_logging.debug("%s not in msi" % outcome)
else:
try:
self.results[outcome][case_id,] = outcome_res
self.results[outcome].flush()
except KeyError:
data = np.asarray(outcome_res)
shape = data.shape
if len(shape) > 2:
raise ema_exceptions.EMAError(self.shape_error_msg.format(len(shape)))
shape = list(shape)
shape.insert(0, self.nr_experiments)
shape = tuple(shape)
fh = tempfile.TemporaryFile()
self.results[outcome] = np.memmap(fh, dtype=data.dtype, shape=shape)
self.results[outcome][:] = np.NAN
self.results[outcome][case_id,] = data
self.results[outcome].flush()
def model_init(self, policy, kwargs):
'''
Method called to initialize the model.
Parameters
----------
policy : dict
policy to be run.
kwargs : dict
keyword arguments to be used by model_intit. This
gives users to the ability to pass any additional
arguments.
'''
if not self.xl:
try:
ema_logging.debug("trying to start Excel")
self.xl = win32com.client.Dispatch("Excel.Application")
ema_logging.debug("Excel started")
except com_error as e:
raise EMAError(str(e))
ema_logging.debug("trying to open workbook")
self.wb = self.xl.Workbooks.Open(self.working_directory +
self.workbook)
ema_logging.debug("workbook opened")
ema_logging.debug(self.working_directory)
def _store_result(self, case_id, result):
for outcome in self.outcomes:
ema_logging.debug("storing {}".format(outcome))
try:
outcome_res = result[outcome]
except KeyError:
ema_logging.debug("%s not in msi" % outcome)
else:
try:
self.results[outcome][case_id, ] = outcome_res
self.results[outcome].flush()
except KeyError:
data = np.asarray(outcome_res)
shape = data.shape
if len(shape) > 2:
raise ema_exceptions.EMAError(
self.shape_error_msg.format(len(shape)))
shape = list(shape)
shape.insert(0, self.nr_experiments)
shape = tuple(shape)
fh = tempfile.TemporaryFile()
self.results[outcome] = np.memmap(fh,
dtype=data.dtype,
shape=shape)
self.results[outcome][:] = np.NAN
self.results[outcome][case_id, ] = data
self.results[outcome].flush()
def __call__(self, case_id, case, policy, name, result):
'''
Method responsible for storing results. The implementation in this
class only keeps track of how many runs have been completed and
logging this. Any extension of AbstractCallback needs to implement
this method. If one want to use the logging provided here, call it via
super.
Parameters
----------
case_id: int
the job id
case: dict
the case to be stored
policy: str
the name of the policy being used
name: str
the name of the model being used
result: dict
the result dict
'''
self.i+=1
ema_logging.debug(str(self.i)+" cases completed")
if self.i % self.reporting_interval == 0:
ema_logging.info(str(self.i)+" cases completed")
def model_init(self, policy, kwargs):
'''
Method called to initialize the model.
Parameters
----------
policy : dict
policy to be run.
kwargs : dict
keyword arguments to be used by model_intit. This
gives users to the ability to pass any additional
arguments.
'''
if not self.xl:
try:
ema_logging.debug("trying to start Excel")
self.xl = win32com.client.Dispatch("Excel.Application")
ema_logging.debug("Excel started")
except com_error as e:
raise EMAError(str(e))
ema_logging.debug("trying to open workbook")
self.wb = self.xl.Workbooks.Open(self.working_directory + self.workbook)
ema_logging.debug("workbook opened")
ema_logging.debug(self.working_directory)
def _store_result(self, case_id, result):
for outcome in self.outcomes:
ema_logging.debug("storing {}".format(outcome))
try:
outcome_res = result[outcome]
except KeyError:
ema_logging.debug("%s not specified as outcome in msi" %
outcome)
else:
try:
self.results[outcome][case_id, ] = outcome_res
except KeyError:
shape = np.asarray(outcome_res).shape
if len(shape) > 2:
raise ema_exceptions.EMAError(
self.shape_error_msg.format(len(shape)))
shape = list(shape)
shape.insert(0, self.nr_experiments)
self.results[outcome] = np.empty(shape)
self.results[outcome][:] = np.NAN
self.results[outcome][case_id, ] = outcome_res
def __init__(self, working_directory, name):
"""interface to the model
interface to the model
Parameters
----------
working_directory : str
working_directory for the model.
name : str
name of the modelInterface. The name should contain only
alpha-numerical characters.
Raises
------
EMAError if name contains non alpha-numerical characters
.. note:: Anything that is relative to `self.working_directory`
should be specified in `model_init` and not
in `__init__`. Otherwise, the code will not work when running
it in parallel. The reason for this is that the working
directory is being updated by parallelEMA to the worker's
separate working directory prior to calling `model_init`.
"""
super(VensimModelStructureInterface,
self).__init__(working_directory, name)
self.outcomes.append(Outcome('TIME', time=True))
self.outcomes = list(self.outcomes)
self._lookup_uncertainties = []
debug("vensim interface init completed")
def run_simulation(file_name):
'''
Convenient function to run a model and store the results of the run in
the specified .vdf file. The specified output file will be overwritten
by default
Parameters
----------
file_name : str
the file name of the output file relative to the working
directory
Raises
------
VensimError if running the model failed in some way.
'''
file_name = str(file_name)
try:
debug(" executing COMMAND: SIMULATE>RUNNAME|" + file_name + "|O")
command("SIMULATE>RUNNAME|" + file_name + "|O")
debug(r"MENU>RUN|o")
command(r"MENU>RUN|o")
except VensimWarning as w:
warning((str(w)))
raise VensimError(str(w))
def _first_get_population(self):
''' called only once to initialize some stuff, returns
a population. After the first call, _get_population is used instead.
'''
ema_logging.debug("Start of evolution")
new_pop = self.toolbox.population(self.pop_size)
# Evaluate the entire population
self.evaluate_population(new_pop, self.reporting_interval,
self.toolbox, self.ensemble)
self.pop = new_pop
# This is just to assign the crowding distance to the individuals
tools.emo.assignCrowdingDist(self.pop)
self.stats_callback(self.pop)
self.stats_callback.log_stats(self.called)
if self.caching:
self._update_cache(self.pop)
self.get_population = self._get_population
def _first_get_population(self):
''' called only once to initialize some stuff, returns
a population. After the first call, _get_population is used instead.
'''
ema_logging.debug("Start of evolution")
new_pop = self.toolbox.population(self.pop_size)
# Evaluate the entire population
self.evaluate_population(new_pop, self.reporting_interval, self.toolbox,
self.ensemble)
self.pop = new_pop
# This is just to assign the crowding distance to the individuals
tools.emo.assignCrowdingDist(self.pop)
self.stats_callback(self.pop)
self.stats_callback.log_stats(self.called)
if self.caching:
self._update_cache(self.pop)
self.get_population = self._get_population
def run_simulation(file_name):
'''
Convenient function to run a model and store the results of the run in
the specified .vdf file. The specified output file will be overwritten
by default
Parameters
----------
file_name : str
the file name of the output file relative to the working
directory
Raises
------
VensimError if running the model failed in some way.
'''
file_name = str(file_name)
try:
debug(" executing COMMAND: SIMULATE>RUNNAME|"+file_name+"|O")
command("SIMULATE>RUNNAME|"+file_name+"|O")
debug(r"MENU>RUN|o")
command(r"MENU>RUN|o")
except VensimWarning as w:
warning((str(w)))
raise VensimError(str(w))
def cleanup(self):
''' cleaning up prior to finishing performing experiments. This will
close the workbook and close Excel'''
ema_logging.debug("cleaning up")
if self.wb:
self.wb.Close(False)
del self.wb
if self.xl:
self.xl.DisplayAlerts = False
self.xl.Quit()
del self.xl
self.xl = None
self.wb = None
def cleanup(self):
''' cleaning up prior to finishing performing experiments. This will
close the workbook and close Excel'''
ema_logging.debug("cleaning up")
if self.wb:
self.wb.Close(False)
del self.wb
if self.xl:
self.xl.DisplayAlerts = False
self.xl.Quit()
del self.xl
self.xl = None
self.wb = None
def set_engine_logger():
'''Updates EMA logging on the engines with an EngineLoggerAdapter
This adapter injects EMA as a topic into all messages
'''
logger = Application.instance().log
logger.setLevel(ema_logging.DEBUG)
for handler in logger.handlers:
if isinstance(handler, IPython.kernel.zmq.log.EnginePUBHandler): # @UndefinedVariable
handler.setLevel(ema_logging.DEBUG)
adapter = EngingeLoggerAdapter(logger, SUBTOPIC)
ema_logging._logger = adapter
ema_logging.debug('updated logger')
def set_working_directory(self, wd):
'''
Method for setting the working directory of the model interface. This
method is used in case of running models in parallel. In this case,
each worker process will have its own working directory, to avoid
having to share files across processes. This requires the need to
update the working directory to the new working directory.
Parameters
----------
wd : str
The new working directory.
'''
wd = os.path.abspath(wd)
debug('setting working directory to ' + wd)
self._working_directory = wd
def set_working_directory(self, wd):
"""
Method for setting the working directory of the model interface. This
method is used in case of running models in parallel. In this case,
each worker process will have its own working directory, to avoid
having to share files across processes. This requires the need to
update the working directory to the new working directory.
Parameters
----------
wd : str
The new working directory.
"""
wd = os.path.abspath(wd)
debug("setting working directory to " + wd)
self._working_directory = wd
def run_model(self, case):
"""
Method for running an instantiated model structure.
the provided implementation assumes that the keys in the
case match the variable names in the Vensim model.
If lookups are to be set specify their transformation from
uncertainties to lookup values in the extension of this method,
then call this one using super with the updated case dict.
if you want to use cin_files, set the cin_file, or cin_files in
the extension of this method to `self.cin_file`.
Parameters
----------
case : dict
keyword arguments for running the model. The case is a dict with
the names of the uncertainties as key, and the values to which
to set these uncertainties.
.. note:: setting parameters should always be done via run_model.
The model is reset to its initial values automatically after
each run.
"""
if self.cin_file:
try:
read_cin_file(self.working_directory + self.cin_file)
except VensimWarning as w:
debug(str(w))
else:
debug("cin file read successfully")
for lookup_uncertainty in self._lookup_uncertainties:
# ask the lookup to transform the retrieved uncertainties to the
# proper lookup value
case[lookup_uncertainty.name] = lookup_uncertainty.transform(case)
for key, value in case.items():
set_value(key, value)
def read_cin_file(file_name):
'''
read a .cin file
Parameters
----------
file_name : str
file name of cin file, relative to working directory
Raises
------
VensimWarning if the cin file cannot be read.
'''
debug("executing COMMAND: SIMULATE>READCIN|" + file_name)
try:
command(r"SIMULATE>READCIN|" + str(file_name))
except VensimWarning as w:
debug(str(w))
raise w
def read_cin_file(file_name):
'''
read a .cin file
Parameters
----------
file_name : str
file name of cin file, relative to working directory
Raises
------
VensimWarning if the cin file cannot be read.
'''
debug("executing COMMAND: SIMULATE>READCIN|"+file_name)
try:
command(r"SIMULATE>READCIN|"+str(file_name))
except VensimWarning as w:
debug(str(w))
raise w
def experiment_generator(designs, model_structures, policies):
'''
generator function which yields experiments
Parameters
----------
designs : iterable of dicts
model_structures : list
policies : list
Notes
-----
this generator is essentially three nested loops: for each model structure,
for each policy, for each experiment, run the experiment. This means
that designs should not be a generator because this will be exhausted after
the running the first policy on the first model.
'''
job_counter = itertools.count()
for msi in model_structures:
debug("generating designs for model %s" % (msi.name))
msi_uncs = {unc.name for unc in msi.uncertainties}
for policy in policies:
debug("generating designs for policy %s" % (policy['name']))
for design in designs:
# from the design only get the uncertainties that
# are valid for the current msi
keys = set(design.keys()).intersection(msi_uncs)
experiment = {unc: design[unc] for unc in keys}
# complete the design by adding the policy, model name
# and experiment id to it
experiment['policy'] = policy
experiment['model'] = msi.name
experiment['experiment id'] = six.next(job_counter)
yield experiment
def experiment_generator(designs, model_structures, policies):
'''
generator function which yields experiments
Parameters
----------
designs : iterable of dicts
model_structures : list
policies : list
Notes
-----
this generator is essentially three nested loops: for each model structure,
for each policy, for each experiment, run the experiment. This means
that designs should not be a generator because this will be exhausted after
the running the first policy on the first model.
'''
job_counter = itertools.count()
for msi in model_structures:
debug("generating designs for model %s" % (msi.name))
msi_uncs = {unc.name for unc in msi.uncertainties}
for policy in policies:
debug("generating designs for policy %s" % (policy['name']))
for design in designs:
# from the design only get the uncertainties that
# are valid for the current msi
keys = set(design.keys()).intersection(msi_uncs)
experiment = {unc:design[unc] for unc in keys}
# complete the design by adding the policy, model name
# and experiment id to it
experiment['policy'] = policy
experiment['model'] = msi.name
experiment['experiment id'] = six.next(job_counter)
yield experiment
def model_init(self, policy, kwargs):
'''
Method called to initialize the model.
Parameters
----------
policy : dict
policy to be run.
kwargs : dict
keyword arguments to be used by model_intit. This
gives users to the ability to pass any additional
arguments.
'''
self.policy = policy
self.netlogo = pyNetLogo.NetLogoLink()
debug("netlogo started")
path = self.working_directory+self.model_file
self.netlogo.load_model(path)
debug("model opened")
def __init__(self, gui=False, thd=False):
'''
Create a link with netlogo. Underneath, the netlogo jvm is started
through jpype.
:param gui: boolean, if true run netlogo with gui, otherwise run in
headless mode. Defaults to false.
:param thd: boolean, if thrue start netlogo in 3d mode. Defaults to
false
'''
if not jpype.isJVMStarted():
# netlogo jars
jars = [
NETLOGO_HOME + r'/lib/scala-library.jar',
NETLOGO_HOME + r'/lib/asm-all-3.3.1.jar',
NETLOGO_HOME + r'/lib/picocontainer-2.13.6.jar',
NETLOGO_HOME + r'/lib/log4j-1.2.16.jar',
NETLOGO_HOME + r'/lib/jmf-2.1.1e.jar',
NETLOGO_HOME + r'/lib/pegdown-1.1.0.jar',
NETLOGO_HOME + r'/lib/parboiled-core-1.0.2.jar',
NETLOGO_HOME + r'/lib/parboiled-java-1.0.2.jar',
NETLOGO_HOME + r'/lib/mrjadapter-1.2.jar',
NETLOGO_HOME + r'/lib/jhotdraw-6.0b1.jar',
NETLOGO_HOME + r'/lib/quaqua-7.3.4.jar',
NETLOGO_HOME + r'/lib/swing-layout-7.3.4.jar',
NETLOGO_HOME + r'/lib/jogl-1.1.1.jar', NETLOGO_HOME +
r'/lib/gluegen-rt-1.1.1.jar', NETLOGO_HOME + r'/NetLogo.jar',
PYNETLOGO_HOME + r'/external_files/netlogoLink.jar'
]
# format jars in right format for starting java virtual machine
# TODO the use of the jre here is only relevant under windows
# apparently
# might be solvable by setting netlogo home user.dir
joined_jars = jar_separator.join(jars)
jarpath = '-Djava.class.path={}'.format(joined_jars)
jvm_handle = jpype.getDefaultJVMPath()
jpype.startJVM(jvm_handle, jarpath, "-Xms128M", "-Xmx1024m")
jpype.java.lang.System.setProperty('user.dir', NETLOGO_HOME)
if sys.platform == 'darwin':
jpype.java.lang.System.setProperty("java.awt.headless", "true")
debug("jvm started")
link = jpype.JClass('netlogoLink.NetLogoLink')
debug('NetLogoLink class found')
if sys.platform == 'darwin' and gui:
info('on mac only headless mode is supported')
gui = False
self.link = link(gui, thd)
debug('NetLogoLink class instantiated')
def load_model(file_name):
'''
load the model
Parameters
----------
file_name : str
file name of model, relative to working directory
Raises
-------
VensimError if the model cannot be loaded.
.. note: only works for .vpm files
'''
debug("executing COMMAND: SIMULATE>SPECIAL>LOADMODEL|" + file_name)
try:
command("SPECIAL>LOADMODEL|" + str(file_name))
except VensimWarning as w:
warning(str(w))
raise VensimError("vensim file not found")
def test_log_messages(self):
ema_logging.log_to_stderr(ema_logging.DEBUG)
with mock.patch('util.ema_logging._logger') as mocked_logger:
message = 'test message'
ema_logging.debug(message)
mocked_logger.debug.assert_called_with(message)
ema_logging.info(message)
mocked_logger.info.assert_called_with(message)
ema_logging.warning(message)
mocked_logger.warning.assert_called_with(message)
ema_logging.error(message)
mocked_logger.error.assert_called_with(message)
ema_logging.exception(message)
mocked_logger.exception.assert_called_with(message)
ema_logging.critical(message)
mocked_logger.critical.assert_called_with(message)
def __init__(self, gui=False, thd=False):
'''
Create a link with netlogo. Underneath, the netlogo jvm is started
through jpype.
:param gui: boolean, if true run netlogo with gui, otherwise run in
headless mode. Defaults to false.
:param thd: boolean, if thrue start netlogo in 3d mode. Defaults to
false
'''
if not jpype.isJVMStarted():
# netlogo jars
jars = [NETLOGO_HOME + r'/lib/scala-library.jar',
NETLOGO_HOME + r'/lib/asm-all-3.3.1.jar',
NETLOGO_HOME + r'/lib/picocontainer-2.13.6.jar',
NETLOGO_HOME + r'/lib/log4j-1.2.16.jar',
NETLOGO_HOME + r'/lib/jmf-2.1.1e.jar',
NETLOGO_HOME + r'/lib/pegdown-1.1.0.jar',
NETLOGO_HOME + r'/lib/parboiled-core-1.0.2.jar',
NETLOGO_HOME + r'/lib/parboiled-java-1.0.2.jar',
NETLOGO_HOME + r'/lib/mrjadapter-1.2.jar',
NETLOGO_HOME + r'/lib/jhotdraw-6.0b1.jar',
NETLOGO_HOME + r'/lib/quaqua-7.3.4.jar',
NETLOGO_HOME + r'/lib/swing-layout-7.3.4.jar',
NETLOGO_HOME + r'/lib/jogl-1.1.1.jar',
NETLOGO_HOME + r'/lib/gluegen-rt-1.1.1.jar',
NETLOGO_HOME + r'/NetLogo.jar',
PYNETLOGO_HOME + r'/external_files/netlogoLink.jar']
# format jars in right format for starting java virtual machine
# TODO the use of the jre here is only relevant under windows
# apparently
# might be solvable by setting netlogo home user.dir
joined_jars = jar_separator.join(jars)
jarpath = '-Djava.class.path={}'.format(joined_jars)
jvm_handle = jpype.getDefaultJVMPath()
jpype.startJVM(jvm_handle, jarpath, "-Xms128M","-Xmx1024m")
jpype.java.lang.System.setProperty('user.dir', NETLOGO_HOME)
if sys.platform=='darwin':
jpype.java.lang.System.setProperty("java.awt.headless", "true");
debug("jvm started")
link = jpype.JClass('netlogoLink.NetLogoLink')
debug('NetLogoLink class found')
if sys.platform == 'darwin' and gui:
info('on mac only headless mode is supported')
gui=False
self.link = link(gui, thd)
debug('NetLogoLink class instantiated')
def load_model(file_name):
'''
load the model
Parameters
----------
file_name : str
file name of model, relative to working directory
Raises
-------
VensimError if the model cannot be loaded.
.. note: only works for .vpm files
'''
debug("executing COMMAND: SIMULATE>SPECIAL>LOADMODEL|"+file_name)
try:
command("SPECIAL>LOADMODEL|"+str(file_name))
except VensimWarning as w:
warning(str(w))
raise VensimError("vensim file not found")
def _store_result(self, case_id, result):
for outcome in self.outcomes:
ema_logging.debug("storing {}".format(outcome))
try:
outcome_res = result[outcome]
except KeyError:
ema_logging.debug("%s not specified as outcome in msi" % outcome)
else:
try:
self.results[outcome][case_id, ] = outcome_res
except KeyError:
shape = np.asarray(outcome_res).shape
if len(shape)>2:
raise ema_exceptions.EMAError(self.shape_error_msg.format(len(shape)))
shape = list(shape)
shape.insert(0, self.nr_experiments)
self.results[outcome] = np.empty(shape)
self.results[outcome][:] = np.NAN
self.results[outcome][case_id, ] = outcome_res
def model_init(self, policy, kwargs):
"""
Init of the model, The provided implementation here assumes
that `self.model_file` is set correctly. In case of using different
vensim models for different policies, it is recommended to extent
this method, extract the model file from the policy dict, set
`self.model_file` to this file and then call this implementation
through calling `super`.
Parameters
----------
policy : dict
policy to be run.
kwargs : dict
keyword arguments to be used by model_intit. This
gives users to the ability to pass any additional
arguments.
"""
load_model(self.working_directory+self.model_file) #load the model
debug("model initialized successfully")
be_quiet() # minimize the screens that are shown
try:
initialTime = get_val('INITIAL TIME')
finalTime = get_val('FINAL TIME')
timeStep = get_val('TIME STEP')
savePer = get_val('SAVEPER')
if savePer > 0:
timeStep = savePer
self.run_length = int((finalTime - initialTime)/timeStep +1)
except VensimWarning:
raise EMAWarning(str(VensimWarning))
def model_init(self, policy, kwargs):
"""
Init of the model, The provided implementation here assumes
that `self.model_file` is set correctly. In case of using different
vensim models for different policies, it is recommended to extent
this method, extract the model file from the policy dict, set
`self.model_file` to this file and then call this implementation
through calling `super`.
Parameters
----------
policy : dict
policy to be run.
kwargs : dict
keyword arguments to be used by model_intit. This
gives users to the ability to pass any additional
arguments.
"""
load_model(self.working_directory + self.model_file) #load the model
debug("model initialized successfully")
be_quiet() # minimize the screens that are shown
try:
initialTime = get_val('INITIAL TIME')
finalTime = get_val('FINAL TIME')
timeStep = get_val('TIME STEP')
savePer = get_val('SAVEPER')
if savePer > 0:
timeStep = savePer
self.run_length = int((finalTime - initialTime) / timeStep + 1)
except VensimWarning:
raise EMAWarning(str(VensimWarning))
def worker(inqueue,
outqueue,
model_interfaces,
model_kwargs=None):
#
# Code run by worker processes
#
ema_logging.debug("worker started")
if hasattr(inqueue, '_writer'):
inqueue._writer.close()
outqueue._reader.close()
msis = {msi.name: msi for msi in model_interfaces}
runner = ExperimentRunner(msis, model_kwargs)
while 1:
try:
task = inqueue.get()
except (EOFError, IOError):
ema_logging.debug('worker got EOFError or IOError -- exiting')
break
if task is None:
ema_logging.debug('worker got sentinel -- exiting')
break
_, experiment = task
experiment_id = experiment.get('experiment id')
try:
result = runner.run_experiment(experiment)
outqueue.put((experiment_id, (True, result)))
except EMAError as inst:
result = (False, inst)
outqueue.put((experiment_id, result))
except Exception:
result = (False, EMAParallelError("failure to initialize"))
outqueue.put((experiment_id, result))
def envelopes(results,
outcomes_to_show = [],
group_by = None,
grouping_specifiers = None,
density=None,
fill=False,
legend=True,
titles={},
ylabels={},
log=False):
''' Make envelop plots. An envelope shows over time the minimum and maximum
value for a set of runs over time. It is thus to be used in case of time
series data. The function will try to find a result labeled "TIME". If this
is present, these values will be used on the X-axis. In case of Vensim
models, TIME is present by default.
Parameters
----------
results : tupule
return from :meth:`perform_experiments`.
outcomes_to_show : list of str, optional
list of outcome of interest you want to plot. If empty,
all outcomes are plotted. **Note**: just names.
group_by : str, optional
name of the column in the cases array to group results by.
Alternatively, `index` can be used to use indexing arrays as the
basis for grouping.
grouping_specifiers : iterable or dict, optional
set of categories to be used as a basis for grouping
by. Grouping_specifiers is only meaningful if
group_by is provided as well. In case of grouping by
index, the grouping specifiers should be in a
dictionary where the key denotes the name of the
group.
density : {None, HIST, KDE, VIOLIN, BOXPLOT}, optional
fill : bool, optional
legend : bool, optional
titles : dict, optional
a way for controlling whether each of the axes should have a
title. There are three possibilities. If set to None, no title
will be shown for any of the axes. If set to an empty dict,
the default, the title is identical to the name of the outcome of
interest. If you want to override these default names, provide a
dict with the outcome of interest as key and the desired title as
value. This dict need only contain the outcomes for which you
want to use a different title.
ylabels : dict, optional
way for controlling the ylabels. Works identical to titles.
log : bool, optional
log scale density plot
Returns
-------
Figure : Figure instance
axes : dict
dict with outcome as key, and axes as value. Density axes' are
indexed by the outcome followed by _density.
Note
----
the current implementation is limited to seven different categories in case
of group_by, categories, and/or discretesize. This limit is due to the colors
specified in COLOR_LIST.
Examples
--------
>>> import util as util
>>> data = util.load_results(r'1000 flu cases.cPickle')
>>> envelopes(data, group_by='policy')
will show an envelope for three three different policies, for all the
outcomes of interest. while
>>> envelopes(data, group_by='policy', categories=['static policy',
'adaptive policy'])
will only show results for the two specified policies, ignoring any results
associated with \'no policy\'.
'''
debug("generating envelopes")
results = copy.deepcopy(results)
prepared_data = prepare_data(results, outcomes_to_show, group_by,
grouping_specifiers)
outcomes, outcomes_to_show, time, grouping_labels = prepared_data
figure, grid = make_grid(outcomes_to_show, density)
# do the plotting
axes_dict = {}
for i, outcome_to_plot in enumerate(outcomes_to_show):
ax = figure.add_subplot(grid[i,0])
axes_dict[outcome_to_plot] = ax
ax_d= None
if density:
ax_d = figure.add_subplot(grid[i,1], sharey=ax)
axes_dict[outcome_to_plot+"_density"] = ax_d
if group_by:
group_by_envelopes(outcomes,outcome_to_plot, time, density,
ax, ax_d, fill, grouping_labels, log)
else:
single_envelope(outcomes, outcome_to_plot, time, density,
ax, ax_d, fill, log)
if ax_d:
for tl in ax_d.get_yticklabels():
tl.set_visible(False)
ax.set_xlabel(TIME_LABEL)
do_ylabels(ax, ylabels, outcome_to_plot)
do_titles(ax, titles, outcome_to_plot)
if legend and group_by:
gs1 = grid[0,0]
for ax in figure.axes:
gs2 = ax._subplotspec
if all((gs1._gridspec == gs2._gridspec,
gs1.num1 == gs2.num1,
gs1.num2 == gs2.num2)):
break
if fill:
make_legend(grouping_labels, ax, alpha=0.3,
legend_type=PATCH)
else:
make_legend(grouping_labels, ax, legend_type=LINE)
if TIGHT:
grid.tight_layout(figure)
return figure, axes_dict
def pairs_scatter(results,
outcomes_to_show = [],
group_by = None,
grouping_specifiers = None,
ylabels = {},
legend=True,
point_in_time=-1,
filter_scalar=True,
**kwargs):
'''
Generate a `R style pairs <http://www.stat.psu.edu/~dhunter/R/html/graphics/html/pairs.html>`_
scatter multiplot. In case of time-series data, the end states are used.
Parameters
----------
results : tuple
return from perform_experiments.
outcomes_to_show : list of str, optional
list of outcome of interest you want to plot.
group_by : str, optional
name of the column in the cases array to group results by.
Alternatively, `index` can be used to use indexing arrays as the
basis for grouping.
grouping_specifiers : dict, optional
dict of categories to be used as a basis for grouping
by. Grouping_specifiers is only meaningful if
group_by is provided as well. In case of grouping by
index, the grouping specifiers should be in a
dictionary where the key denotes the name of the
group.
ylabels : dict, optional
ylabels is a dictionary with the outcome names as keys, the
specified values will be used as labels for the y axis.
legend : bool, optional
if true, and group_by is given, show a legend.
point_in_time : float, optional
the point in time at which the scatter is to be made. If
None is provided (default), the end states are used.
point_in_time should be a valid value on time
filter_scalar: bool, optional
remove the non-time-series outcomes. Defaults to True.
Returns
-------
fig : Figure instance
the figure instance
axes : dict
key is tuple of names of outcomes, value is associated axes
instance
.. note:: the current implementation is limited to seven different
categories in case of column, categories, and/or discretesize.
This limit is due to the colors specified in COLOR_LIST.
'''
debug("generating pairwise scatter plot")
prepared_data = prepare_pairs_data(results, outcomes_to_show, group_by,
grouping_specifiers, point_in_time,
filter_scalar)
outcomes, outcomes_to_show, grouping_labels = prepared_data
grid = gridspec.GridSpec(len(outcomes_to_show), len(outcomes_to_show))
grid.update(wspace = 0.1,
hspace = 0.1)
#the plotting
figure = plt.figure()
axes_dict = {}
combis = [(field1, field2) for field1 in outcomes_to_show\
for field2 in outcomes_to_show]
for field1, field2 in combis:
i = list(outcomes_to_show).index(field1)
j = list(outcomes_to_show).index(field2)
ax = figure.add_subplot(grid[i,j])
axes_dict[(field1, field2)] = ax
if group_by:
for x, group in enumerate(grouping_labels):
y_data = outcomes[group][field1]
x_data = outcomes[group][field2]
facecolor = plotting_util.COLOR_LIST[x]
edgecolor = 'k'
if i==j:
facecolor = 'white'
edgecolor = 'white'
ax.scatter(x_data, y_data,
facecolor=facecolor, edgecolor=edgecolor)
else:
y_data = outcomes[field1]
x_data = outcomes[field2]
facecolor = 'b'
edgecolor = 'k'
if i==j:
facecolor = 'white'
edgecolor = 'white'
ax.scatter(x_data, y_data,
facecolor=facecolor, edgecolor=edgecolor)
do_text_ticks_labels(ax, i, j, field1, field2, ylabels,
outcomes_to_show)
if group_by and legend:
gs1 = grid[0,0]
for ax in figure.axes:
gs2 = ax._subplotspec
if all((gs1._gridspec == gs2._gridspec,
gs1.num1 == gs2.num1,
gs1.num2 == gs2.num2)):
break
make_legend(grouping_labels, ax, legend_type=SCATTER)
return figure, axes_dict
def run_model(self, case):
"""
Method for running an instantiated model structure.
the provided implementation assumes that the keys in the
case match the variable names in the Vensim model.
If lookups are to be set specify their transformation from
uncertainties to lookup values in the extension of this method,
then call this one using super with the updated case dict.
if you want to use cin_files, set the cin_file, or cin_files in
the extension of this method to `self.cin_file`.
Parameters
----------
case : dict
keyword arguments for running the model. The case is a dict with
the names of the uncertainties as key, and the values to which
to set these uncertainties.
.. note:: setting parameters should always be done via run_model.
The model is reset to its initial values automatically after
each run.
"""
if self.cin_file:
try:
read_cin_file(self.working_directory+self.cin_file)
except VensimWarning as w:
debug(str(w))
else:
debug("cin file read successfully")
for lookup_uncertainty in self._lookup_uncertainties:
# ask the lookup to transform the retrieved uncertainties to the
# proper lookup value
case[lookup_uncertainty.name] = lookup_uncertainty.transform(case)
for key, value in case.items():
set_value(key, value)
debug("model parameters set successfully")
debug("run simulation, results stored in " + self.working_directory+self.result_file)
try:
run_simulation(self.working_directory+self.result_file)
except VensimError:
raise
results = {}
error = False
for output in self.outcomes:
debug("getting data for %s" %output.name)
result = get_data(self.working_directory+self.result_file,
output.name
)
debug("successfully retrieved data for %s" %output.name)
if result is not None:
if result.shape[0] != self.run_length:
got = result.shape[0]
data = np.empty((self.run_length))
data[:] = np.NAN
data[0:result.shape[0]] = result
result = data
error = True
if not output.time:
result = result[-1]
else:
result = result[0::self.step]
try:
results[output.name] = result
except ValueError as e:
print("what")
raise e
self.output = results
if error:
raise CaseError("run not completed, got %s, expected %s" %
(got, self.run_length), case)
def pairs_lines(results,
outcomes_to_show=[],
group_by=None,
grouping_specifiers=None,
ylabels={},
legend=True,
**kwargs):
'''
Generate a `R style pairs <http://www.stat.psu.edu/~dhunter/R/html/graphics/html/pairs.html>`_
lines multiplot. It shows the behavior of two outcomes over time against
each other. The origin is denoted with a circle and the end is denoted
with a '+'.
Parameters
----------
results : tuple
return from perform_experiments.
outcomes_to_show : list of str, optional
list of outcome of interest you want to plot.
group_by : str, optional
name of the column in the cases array to group results by.
Alternatively, `index` can be used to use indexing arrays as the
basis for grouping.
grouping_specifiers : dict, optional
dict of categories to be used as a basis for grouping
by. Grouping_specifiers is only meaningful if
group_by is provided as well. In case of grouping by
index, the grouping specifiers should be in a
dictionary where the key denotes the name of the
group.
ylabels : dict, optional
ylabels is a dictionary with the outcome names as keys, the
specified values will be used as labels for the y axis.
legend : bool, optional
if true, and group_by is given, show a legend.
point_in_time : float, optional
the point in time at which the scatter is to be made. If
None is provided (default), the end states are used.
point_in_time should be a valid value on time
Returns
-------
fig
the figure instance
dict
key is tuple of names of outcomes, value is associated axes
instance
'''
#unravel return from run_experiments
debug("making a pars lines plot")
prepared_data = prepare_pairs_data(results, outcomes_to_show, group_by,
grouping_specifiers, None)
outcomes, outcomes_to_show, grouping_labels = prepared_data
grid = gridspec.GridSpec(len(outcomes_to_show), len(outcomes_to_show))
grid.update(wspace=0.1, hspace=0.1)
#the plotting
figure = plt.figure()
axes_dict = {}
combis = [(field1, field2) for field1 in outcomes_to_show\
for field2 in outcomes_to_show]
for field1, field2 in combis:
i = list(outcomes_to_show).index(field1)
j = list(outcomes_to_show).index(field2)
ax = figure.add_subplot(grid[i, j])
axes_dict[(field1, field2)] = ax
if group_by:
for x, entry in enumerate(grouping_labels):
data1 = outcomes[entry][field1]
data2 = outcomes[entry][field2]
color = plotting_util.COLOR_LIST[x]
if i == j:
color = 'white'
simple_pairs_lines(ax, data1, data2, color)
else:
data1 = outcomes[field1]
data2 = outcomes[field2]
color = 'b'
if i == j:
color = 'white'
simple_pairs_lines(ax, data1, data2, color)
do_text_ticks_labels(ax, i, j, field1, field2, ylabels,
outcomes_to_show)
if group_by and legend:
gs1 = grid[0, 0]
for ax in figure.axes:
gs2 = ax._subplotspec
if all((gs1._gridspec == gs2._gridspec, gs1.num1 == gs2.num1,
gs1.num2 == gs2.num2)):
break
make_legend(grouping_labels, ax, legend_type=LINE)
return figure, axes_dict
def lines(results,
outcomes_to_show=[],
group_by=None,
grouping_specifiers=None,
density='',
legend=True,
titles={},
ylabels={},
experiments_to_show=None,
show_envelope=False,
log=False):
'''This function takes the results from :meth:`perform_experiments` and
visualizes these as line plots. It is thus to be used in case of time
series data. The function will try to find a result labeled "TIME". If this
is present, these values will be used on the X-axis. In case of Vensim
models, TIME is present by default.
Parameters
----------
results : tuple
return from :meth:`perform_experiments`.
outcomes_to_show : list of str, optional
list of outcome of interest you want to plot. If empty,
all outcomes are plotted. **Note**: just names.
group_by : str, optional
name of the column in the cases array to group results by.
Alternatively, `index` can be used to use indexing arrays as the
basis for grouping.
grouping_specifiers : iterable or dict, optional
set of categories to be used as a basis for grouping
by. Grouping_specifiers is only meaningful if
group_by is provided as well. In case of grouping by
index, the grouping specifiers should be in a
dictionary where the key denotes the name of the
group.
density : {None, HIST, KDE, VIOLIN, BOXPLOT}, optional
legend : bool, optional
titles : dict, optional
a way for controlling whether each of the axes should have a
title. There are three possibilities. If set to None, no title
will be shown for any of the axes. If set to an empty dict,
the default, the title is identical to the name of the outcome of
interest. If you want to override these default names, provide a
dict with the outcome of interest as key and the desired title as
value. This dict need only contain the outcomes for which you
want to use a different title.
ylabels : dict, optional
way for controlling the ylabels. Works identical to titles.
experiments_to_show : ndarray, optional
indices of experiments to show lines for,
defaults to None.
show_envelope : bool, optional
show envelope op outcomes. This envelope is the based on
the minimum at each column and the maximum at each column.
log : bool, optional
log scale density plot
Returns
-------
fig : Figure instance
axes : dict
dict with outcome as key, and axes as value. Density axes' are
indexed by the outcome followed by _density.
Note
----
the current implementation is limited to seven different categories in case
of group_by, categories, and/or discretesize. This limit is due to the colors
specified in COLOR_LIST.
'''
debug("generating line graph")
# make sure we have the data
results = copy.deepcopy(results)
if show_envelope:
return plot_lines_with_envelopes(
results,
outcomes_to_show=outcomes_to_show,
group_by=group_by,
legend=legend,
density=density,
grouping_specifiers=grouping_specifiers,
experiments_to_show=experiments_to_show,
titles=titles,
ylabels=ylabels,
log=log)
if experiments_to_show is not None:
experiments, outcomes = results
experiments = experiments[experiments_to_show]
new_outcomes = {}
for key, value in outcomes.items():
new_outcomes[key] = value[experiments_to_show]
results = experiments, new_outcomes
data = prepare_data(results, outcomes_to_show, group_by,
grouping_specifiers)
outcomes, outcomes_to_show, time, grouping_labels = data
figure, grid = make_grid(outcomes_to_show, density)
axes_dict = {}
# do the plotting
for i, outcome_to_plot in enumerate(outcomes_to_show):
ax = figure.add_subplot(grid[i, 0])
axes_dict[outcome_to_plot] = ax
ax_d = None
if density:
ax_d = figure.add_subplot(grid[i, 1], sharey=ax)
axes_dict[outcome_to_plot + "_density"] = ax_d
for tl in ax_d.get_yticklabels():
tl.set_visible(False)
if group_by:
group_by_lines(outcomes, outcome_to_plot, time, density, ax, ax_d,
grouping_labels, log)
else:
simple_lines(outcomes, outcome_to_plot, time, density, ax, ax_d,
log)
ax.set_xlabel(TIME_LABEL)
do_ylabels(ax, ylabels, outcome_to_plot)
do_titles(ax, titles, outcome_to_plot)
if legend and group_by:
gs1 = grid[0, 0]
for ax in figure.axes:
gs2 = ax._subplotspec
if all((gs1._gridspec == gs2._gridspec, gs1.num1 == gs2.num1,
gs1.num2 == gs2.num2)):
break
make_legend(grouping_labels, ax)
if TIGHT:
grid.tight_layout(figure)
return figure, axes_dict
def pairs_scatter(results,
outcomes_to_show=[],
group_by=None,
grouping_specifiers=None,
ylabels={},
legend=True,
point_in_time=-1,
filter_scalar=True,
**kwargs):
'''
Generate a `R style pairs <http://www.stat.psu.edu/~dhunter/R/html/graphics/html/pairs.html>`_
scatter multiplot. In case of time-series data, the end states are used.
Parameters
----------
results : tuple
return from perform_experiments.
outcomes_to_show : list of str, optional
list of outcome of interest you want to plot.
group_by : str, optional
name of the column in the cases array to group results by.
Alternatively, `index` can be used to use indexing arrays as the
basis for grouping.
grouping_specifiers : dict, optional
dict of categories to be used as a basis for grouping
by. Grouping_specifiers is only meaningful if
group_by is provided as well. In case of grouping by
index, the grouping specifiers should be in a
dictionary where the key denotes the name of the
group.
ylabels : dict, optional
ylabels is a dictionary with the outcome names as keys, the
specified values will be used as labels for the y axis.
legend : bool, optional
if true, and group_by is given, show a legend.
point_in_time : float, optional
the point in time at which the scatter is to be made. If
None is provided (default), the end states are used.
point_in_time should be a valid value on time
filter_scalar: bool, optional
remove the non-time-series outcomes. Defaults to True.
Returns
-------
fig : Figure instance
the figure instance
axes : dict
key is tuple of names of outcomes, value is associated axes
instance
.. note:: the current implementation is limited to seven different
categories in case of column, categories, and/or discretesize.
This limit is due to the colors specified in COLOR_LIST.
'''
debug("generating pairwise scatter plot")
prepared_data = prepare_pairs_data(results, outcomes_to_show, group_by,
grouping_specifiers, point_in_time,
filter_scalar)
outcomes, outcomes_to_show, grouping_labels = prepared_data
grid = gridspec.GridSpec(len(outcomes_to_show), len(outcomes_to_show))
grid.update(wspace=0.1, hspace=0.1)
#the plotting
figure = plt.figure()
axes_dict = {}
combis = [(field1, field2) for field1 in outcomes_to_show\
for field2 in outcomes_to_show]
for field1, field2 in combis:
i = list(outcomes_to_show).index(field1)
j = list(outcomes_to_show).index(field2)
ax = figure.add_subplot(grid[i, j])
axes_dict[(field1, field2)] = ax
if group_by:
for x, group in enumerate(grouping_labels):
y_data = outcomes[group][field1]
x_data = outcomes[group][field2]
facecolor = plotting_util.COLOR_LIST[x]
edgecolor = 'k'
if i == j:
facecolor = 'white'
edgecolor = 'white'
ax.scatter(x_data,
y_data,
facecolor=facecolor,
edgecolor=edgecolor)
else:
y_data = outcomes[field1]
x_data = outcomes[field2]
facecolor = 'b'
edgecolor = 'k'
if i == j:
facecolor = 'white'
edgecolor = 'white'
ax.scatter(x_data,
y_data,
facecolor=facecolor,
edgecolor=edgecolor)
do_text_ticks_labels(ax, i, j, field1, field2, ylabels,
outcomes_to_show)
if group_by and legend:
gs1 = grid[0, 0]
for ax in figure.axes:
gs2 = ax._subplotspec
if all((gs1._gridspec == gs2._gridspec, gs1.num1 == gs2.num1,
gs1.num2 == gs2.num2)):
break
make_legend(grouping_labels, ax, legend_type=SCATTER)
return figure, axes_dict
def pairs_density(results,
outcomes_to_show = [],
group_by = None,
grouping_specifiers = None,
ylabels = {},
point_in_time=-1,
log=True,
gridsize=50,
colormap='coolwarm',
filter_scalar=True):
'''
Generate a `R style pairs <http://www.stat.psu.edu/~dhunter/R/html/graphics/html/pairs.html>`_
hexbin density multiplot. In case of time-series data, the end states are
used.
hexbin makes hexagonal binning plot of x versus y, where x, y are 1-D
sequences of the same length, N. If C is None (the default), this is a
histogram of the number of occurences of the observations at (x[i],y[i]).
For further detail see `matplotlib on hexbin <http://matplotlib.sourceforge.net/api/pyplot_api.html#matplotlib.pyplot.hexbin>`_
Parameters
----------
results : tuple
return from perform_experiments.
outcomes_to_show : list of str, optional
list of outcome of interest you want to plot.
group_by : str, optional
name of the column in the cases array to group results by.
Alternatively, `index` can be used to use indexing arrays as the
basis for grouping.
grouping_specifiers : dict, optional
dict of categories to be used as a basis for grouping
by. Grouping_specifiers is only meaningful if
group_by is provided as well. In case of grouping by
index, the grouping specifiers should be in a
dictionary where the key denotes the name of the
group.
ylabels : dict, optional
ylabels is a dictionary with the outcome names as keys, the
specified values will be used as labels for the y axis.
point_in_time : float, optional
the point in time at which the scatter is to be made. If
None is provided (default), the end states are used.
point_in_time should be a valid value on time
log: bool, optional
indicating whether density should be log scaled. Defaults to True.
gridsize : int, optional
controls the gridsize for the hexagonal bining. (default = 50)
cmap : str
color map that is to be used in generating the hexbin. For details
on the available maps, see `pylab <http://matplotlib.sourceforge.net/examples/pylab_examples/show_colormaps.html#pylab-examples-show-colormaps>`_.
(Defaults = coolwarm)
filter_scalar: bool, optional
remove the non-time-series outcomes. Defaults to True.
Returns
-------
fig
the figure instance
dict
key is tuple of names of outcomes, value is associated axes
instance
'''
debug("generating pairwise density plot")
prepared_data = prepare_pairs_data(results, outcomes_to_show, group_by,
grouping_specifiers, point_in_time,
filter_scalar)
outcomes, outcomes_to_show, grouping_specifiers = prepared_data
if group_by:
#figure out the extents for each combination
extents = determine_extents(outcomes, outcomes_to_show)
axes_dicts = {}
figures = []
for key, value in outcomes.items():
figure, axes_dict = simple_pairs_density(value, outcomes_to_show,
log, colormap, gridsize, ylabels,
extents=extents, title=key)
axes_dicts[key] = axes_dict
figures.append(figure)
# harmonize the color scaling across figures
combis = [(field1, field2) for field1 in outcomes_to_show\
for field2 in outcomes_to_show]
for combi in combis:
vmax = -1
for entry in axes_dicts.values():
vmax = max(entry[combi].collections[0].norm.vmax, vmax)
for entry in axes_dicts.values():
ax = entry[combi]
ax.collections[0].set_clim(vmin=0, vmax=vmax)
del vmax
return figures, axes_dicts
else:
return simple_pairs_density(outcomes, outcomes_to_show, log, colormap,
gridsize, ylabels)
def envelopes(results,
outcomes_to_show=[],
group_by=None,
grouping_specifiers=None,
density=None,
fill=False,
legend=True,
titles={},
ylabels={},
log=False):
''' Make envelop plots. An envelope shows over time the minimum and maximum
value for a set of runs over time. It is thus to be used in case of time
series data. The function will try to find a result labeled "TIME". If this
is present, these values will be used on the X-axis. In case of Vensim
models, TIME is present by default.
Parameters
----------
results : tupule
return from :meth:`perform_experiments`.
outcomes_to_show : list of str, optional
list of outcome of interest you want to plot. If empty,
all outcomes are plotted. **Note**: just names.
group_by : str, optional
name of the column in the cases array to group results by.
Alternatively, `index` can be used to use indexing arrays as the
basis for grouping.
grouping_specifiers : iterable or dict, optional
set of categories to be used as a basis for grouping
by. Grouping_specifiers is only meaningful if
group_by is provided as well. In case of grouping by
index, the grouping specifiers should be in a
dictionary where the key denotes the name of the
group.
density : {None, HIST, KDE, VIOLIN, BOXPLOT}, optional
fill : bool, optional
legend : bool, optional
titles : dict, optional
a way for controlling whether each of the axes should have a
title. There are three possibilities. If set to None, no title
will be shown for any of the axes. If set to an empty dict,
the default, the title is identical to the name of the outcome of
interest. If you want to override these default names, provide a
dict with the outcome of interest as key and the desired title as
value. This dict need only contain the outcomes for which you
want to use a different title.
ylabels : dict, optional
way for controlling the ylabels. Works identical to titles.
log : bool, optional
log scale density plot
Returns
-------
Figure : Figure instance
axes : dict
dict with outcome as key, and axes as value. Density axes' are
indexed by the outcome followed by _density.
Note
----
the current implementation is limited to seven different categories in case
of group_by, categories, and/or discretesize. This limit is due to the colors
specified in COLOR_LIST.
Examples
--------
>>> import util as util
>>> data = util.load_results(r'1000 flu cases.cPickle')
>>> envelopes(data, group_by='policy')
will show an envelope for three three different policies, for all the
outcomes of interest. while
>>> envelopes(data, group_by='policy', categories=['static policy',
'adaptive policy'])
will only show results for the two specified policies, ignoring any results
associated with \'no policy\'.
'''
debug("generating envelopes")
results = copy.deepcopy(results)
prepared_data = prepare_data(results, outcomes_to_show, group_by,
grouping_specifiers)
outcomes, outcomes_to_show, time, grouping_labels = prepared_data
figure, grid = make_grid(outcomes_to_show, density)
# do the plotting
axes_dict = {}
for i, outcome_to_plot in enumerate(outcomes_to_show):
ax = figure.add_subplot(grid[i, 0])
axes_dict[outcome_to_plot] = ax
ax_d = None
if density:
ax_d = figure.add_subplot(grid[i, 1], sharey=ax)
axes_dict[outcome_to_plot + "_density"] = ax_d
if group_by:
group_by_envelopes(outcomes, outcome_to_plot, time, density, ax,
ax_d, fill, grouping_labels, log)
else:
single_envelope(outcomes, outcome_to_plot, time, density, ax, ax_d,
fill, log)
if ax_d:
for tl in ax_d.get_yticklabels():
tl.set_visible(False)
ax.set_xlabel(TIME_LABEL)
do_ylabels(ax, ylabels, outcome_to_plot)
do_titles(ax, titles, outcome_to_plot)
if legend and group_by:
gs1 = grid[0, 0]
for ax in figure.axes:
gs2 = ax._subplotspec
if all((gs1._gridspec == gs2._gridspec, gs1.num1 == gs2.num1,
gs1.num2 == gs2.num2)):
break
if fill:
make_legend(grouping_labels, ax, alpha=0.3, legend_type=PATCH)
else:
make_legend(grouping_labels, ax, legend_type=LINE)
if TIGHT:
grid.tight_layout(figure)
return figure, axes_dict
class AbstractCallback(object):
'''
Abstract base class from which different call back classes can be derived.
Callback is responsible for storing the results of the runs.
Parameters
----------
uncs : list
a list of the uncertainties over which the experiments
are being run.
outcomes : list
a list of outcomes
nr_experiments : int
the total number of experiments to be executed
reporting_interval : int, optional
the interval at which to provide progress information
via logging.
Attributes
----------
i : int
a counter that keeps track of how many experiments have been saved
reporting_interval : int
the frequency at which to log progress
'''
__metaclass__ = abc.ABCMeta
i = 0
reporting_interval = 100
def __init__(self,
uncertainties,
outcomes,
nr_experiments,
reporting_interval=100):
self.reporting_interval = reporting_interval
@abc.abstractmethod
def __call__(self, case_id, case, policy, name, result):
'''
Method responsible for storing results. The implementation in this
class only keeps track of how many runs have been completed and
logging this. Any extension of AbstractCallback needs to implement
this method. If one want to use the logging provided here, call it via
super.
Parameters
----------
case_id: int
the job id
case: dict
the case to be stored
policy: str
the name of the policy being used
name: str
the name of the model being used
result: dict
the result dict
'''
self.i += 1
ema_logging.debug(str(self.i) + " cases completed")
if self.i % self.reporting_interval == 0:
ema_logging.info(str(self.i) + " cases completed")
def pairs_density(results,
outcomes_to_show=[],
group_by=None,
grouping_specifiers=None,
ylabels={},
point_in_time=-1,
log=True,
gridsize=50,
colormap='coolwarm',
filter_scalar=True):
'''
Generate a `R style pairs <http://www.stat.psu.edu/~dhunter/R/html/graphics/html/pairs.html>`_
hexbin density multiplot. In case of time-series data, the end states are
used.
hexbin makes hexagonal binning plot of x versus y, where x, y are 1-D
sequences of the same length, N. If C is None (the default), this is a
histogram of the number of occurences of the observations at (x[i],y[i]).
For further detail see `matplotlib on hexbin <http://matplotlib.sourceforge.net/api/pyplot_api.html#matplotlib.pyplot.hexbin>`_
Parameters
----------
results : tuple
return from perform_experiments.
outcomes_to_show : list of str, optional
list of outcome of interest you want to plot.
group_by : str, optional
name of the column in the cases array to group results by.
Alternatively, `index` can be used to use indexing arrays as the
basis for grouping.
grouping_specifiers : dict, optional
dict of categories to be used as a basis for grouping
by. Grouping_specifiers is only meaningful if
group_by is provided as well. In case of grouping by
index, the grouping specifiers should be in a
dictionary where the key denotes the name of the
group.
ylabels : dict, optional
ylabels is a dictionary with the outcome names as keys, the
specified values will be used as labels for the y axis.
point_in_time : float, optional
the point in time at which the scatter is to be made. If
None is provided (default), the end states are used.
point_in_time should be a valid value on time
log: bool, optional
indicating whether density should be log scaled. Defaults to True.
gridsize : int, optional
controls the gridsize for the hexagonal bining. (default = 50)
cmap : str
color map that is to be used in generating the hexbin. For details
on the available maps, see `pylab <http://matplotlib.sourceforge.net/examples/pylab_examples/show_colormaps.html#pylab-examples-show-colormaps>`_.
(Defaults = coolwarm)
filter_scalar: bool, optional
remove the non-time-series outcomes. Defaults to True.
Returns
-------
fig
the figure instance
dict
key is tuple of names of outcomes, value is associated axes
instance
'''
debug("generating pairwise density plot")
prepared_data = prepare_pairs_data(results, outcomes_to_show, group_by,
grouping_specifiers, point_in_time,
filter_scalar)
outcomes, outcomes_to_show, grouping_specifiers = prepared_data
if group_by:
#figure out the extents for each combination
extents = determine_extents(outcomes, outcomes_to_show)
axes_dicts = {}
figures = []
for key, value in outcomes.items():
figure, axes_dict = simple_pairs_density(value,
outcomes_to_show,
log,
colormap,
gridsize,
ylabels,
extents=extents,
title=key)
axes_dicts[key] = axes_dict
figures.append(figure)
# harmonize the color scaling across figures
combis = [(field1, field2) for field1 in outcomes_to_show\
for field2 in outcomes_to_show]
for combi in combis:
vmax = -1
for entry in axes_dicts.values():
vmax = max(entry[combi].collections[0].norm.vmax, vmax)
for entry in axes_dicts.values():
ax = entry[combi]
ax.collections[0].set_clim(vmin=0, vmax=vmax)
del vmax
return figures, axes_dicts
else:
return simple_pairs_density(outcomes, outcomes_to_show, log, colormap,
gridsize, ylabels)
def run(self):
self._setupLogger()
p = multiprocessing.current_process()
ema_logging.debug('process %s with pid %s started' % (p.name, p.pid))
#call the run of the super, which in turn will call the worker function
super(LoggingProcess, self).run()
def run_experiment(self, experiment):
'''The logic for running a single experiment. This code makes
sure that model(s) are initialized correctly.
Parameters
----------
experiment : dict
Returns
-------
experiment_id: int
case : dict
policy : str
model_name : str
result : dict
Raises
------
EMAError
if the model instance raises an EMA error, these are reraised.
Exception
Catch all for all other exceptions being raised by the model.
These are reraised.
'''
policy = experiment.pop('policy')
model_name = experiment.pop('model')
experiment_id = experiment.pop('experiment id')
policy_name = policy['name']
ema_logging.debug("running policy {} for experiment {}".format(policy_name,
experiment_id))
# check whether we already initialized the model for this
# policy
if not (policy_name, model_name) in self.msi_initialization.keys():
try:
ema_logging.debug("invoking model init")
msi = self.msis[model_name]
msi.model_init(copy.deepcopy(policy),
copy.deepcopy(self.model_kwargs))
except EMAError as inst:
ema_logging.exception(inst)
self.cleanup()
raise inst
except Exception as inst:
ema_logging.exception("some exception occurred when invoking the init")
self.cleanup()
raise inst
ema_logging.debug("initialized model %s with policy %s" % (model_name,
policy_name))
self.msi_initialization = {(policy_name, model_name):self.msis[model_name]}
msi = self.msis[model_name]
case = copy.deepcopy(experiment)
try:
ema_logging.debug("trying to run model")
msi.run_model(case)
except CaseError as e:
ema_logging.warning(str(e))
ema_logging.debug("trying to retrieve output")
result = msi.retrieve_output()
ema_logging.debug("trying to reset model")
msi.reset_model()
return experiment_id, case, policy, model_name, result
def _handle_tasks(taskqueue, put, outqueue, pool):
thread = threading.current_thread()
for task in iter(taskqueue.get, None):
if thread._state:
ema_logging.debug('task handler found thread._state != RUN')
break
try:
put(task)
except IOError:
ema_logging.debug('could not put task on queue')
break
else:
continue
break
else:
ema_logging.debug('task handler got sentinel')
try:
# tell result handler to finish when cache is empty
ema_logging.debug('task handler sending sentinel to result handler')
outqueue.put(None)
# tell workers there is no more work
ema_logging.debug('task handler sending sentinel to workers')
for _ in range(2*len(pool)):
put(None)
except IOError:
ema_logging.debug('task handler got IOError when sending sentinels')
ema_logging.debug('task handler exiting')
def _handle_results(outqueue, get, cache, log_queue):
thread = threading.current_thread()
while 1:
try:
task = get()
except (IOError, EOFError):
ema_logging.debug('result handler got EOFError/IOError -- exiting')
return
if thread._state:
assert thread._state == pool.TERMINATE
ema_logging.debug('result handler found thread._state=TERMINATE')
break
if task is None:
ema_logging.debug('result handler got sentinel')
break
job, experiment = task
try:
cache[job]._set(experiment)
except KeyError:
pass
while cache and thread._state != pool.TERMINATE:
try:
task = get()
except (IOError, EOFError):
ema_logging.debug('result handler got EOFError/IOError -- exiting')
return
if task is None:
ema_logging.debug('result handler ignoring extra sentinel')
continue
job, obj = task
try:
cache[job]._set(obj)
except KeyError:
pass
if hasattr(outqueue, '_reader'):
ema_logging.debug('ensuring that outqueue is not full')
# If we don't make room available in outqueue then
# attempts to add the sentinel (None) to outqueue may
# block. There is guaranteed to be no more than 2 sentinels.
try:
for _ in range(10):
if not outqueue._reader.poll():
break
get()
except (IOError, EOFError):
pass
ema_logging.debug('result handler exiting: len(cache)=%s, thread._state=%s',
len(cache), thread._state)
log_queue.put(None)
def _terminate_pool(cls,
taskqueue,
inqueue,
outqueue,
pool,
task_handler,
result_handler,
cache,
working_dirs,
):
ema_logging.info("terminating pool")
# this is guaranteed to only be called once
ema_logging.debug('finalizing pool')
TERMINATE = 2
task_handler._state = TERMINATE
for p in pool:
taskqueue.put(None) # sentinel
time.sleep(1)
ema_logging.debug('helping task handler/workers to finish')
cls._help_stuff_finish(inqueue, task_handler, len(pool))
assert result_handler.is_alive() or len(cache) == 0
result_handler._state = TERMINATE
outqueue.put(None) # sentinel
if pool and hasattr(pool[0], 'terminate'):
ema_logging.debug('terminating workers')
for p in pool:
p.terminate()
ema_logging.debug('joining task handler')
task_handler.join(1e100)
ema_logging.debug('joining result handler')
result_handler.join(1e100)
if pool and hasattr(pool[0], 'terminate'):
ema_logging.debug('joining pool workers')
for p in pool:
p.join()
# cleaning up directories
# TODO investigate whether the multiprocessing.util tempdirectory
# functionality can be used instead
for directory in working_dirs:
ema_logging.debug("deleting "+str(directory))
shutil.rmtree(directory)
def pairs_lines(results,
outcomes_to_show = [],
group_by = None,
grouping_specifiers = None,
ylabels = {},
legend=True,
**kwargs):
'''
Generate a `R style pairs <http://www.stat.psu.edu/~dhunter/R/html/graphics/html/pairs.html>`_
lines multiplot. It shows the behavior of two outcomes over time against
each other. The origin is denoted with a circle and the end is denoted
with a '+'.
Parameters
----------
results : tuple
return from perform_experiments.
outcomes_to_show : list of str, optional
list of outcome of interest you want to plot.
group_by : str, optional
name of the column in the cases array to group results by.
Alternatively, `index` can be used to use indexing arrays as the
basis for grouping.
grouping_specifiers : dict, optional
dict of categories to be used as a basis for grouping
by. Grouping_specifiers is only meaningful if
group_by is provided as well. In case of grouping by
index, the grouping specifiers should be in a
dictionary where the key denotes the name of the
group.
ylabels : dict, optional
ylabels is a dictionary with the outcome names as keys, the
specified values will be used as labels for the y axis.
legend : bool, optional
if true, and group_by is given, show a legend.
point_in_time : float, optional
the point in time at which the scatter is to be made. If
None is provided (default), the end states are used.
point_in_time should be a valid value on time
Returns
-------
fig
the figure instance
dict
key is tuple of names of outcomes, value is associated axes
instance
'''
#unravel return from run_experiments
debug("making a pars lines plot")
prepared_data = prepare_pairs_data(results, outcomes_to_show, group_by,
grouping_specifiers, None)
outcomes, outcomes_to_show, grouping_labels = prepared_data
grid = gridspec.GridSpec(len(outcomes_to_show), len(outcomes_to_show))
grid.update(wspace = 0.1,
hspace = 0.1)
#the plotting
figure = plt.figure()
axes_dict = {}
combis = [(field1, field2) for field1 in outcomes_to_show\
for field2 in outcomes_to_show]
for field1, field2 in combis:
i = list(outcomes_to_show).index(field1)
j = list(outcomes_to_show).index(field2)
ax = figure.add_subplot(grid[i,j])
axes_dict[(field1, field2)] = ax
if group_by:
for x, entry in enumerate(grouping_labels):
data1 = outcomes[entry][field1]
data2 = outcomes[entry][field2]
color = plotting_util.COLOR_LIST[x]
if i==j:
color = 'white'
simple_pairs_lines(ax, data1, data2, color)
else:
data1 = outcomes[field1]
data2 = outcomes[field2]
color = 'b'
if i==j:
color = 'white'
simple_pairs_lines(ax, data1, data2, color)
do_text_ticks_labels(ax, i, j, field1, field2, ylabels,
outcomes_to_show)
if group_by and legend:
gs1 = grid[0,0]
for ax in figure.axes:
gs2 = ax._subplotspec
if all((gs1._gridspec == gs2._gridspec,
gs1.num1 == gs2.num1,
gs1.num2 == gs2.num2)):
break
make_legend(grouping_labels, ax, legend_type=LINE)
return figure, axes_dict
def __init__(self,
msis,
processes=None,
kwargs=None):
'''
Parameters
----------
msis : list
iterable of model structure interface instances
processes: int
nr. of processes to spawn, if none, it is set to equal the
nr. of cores
kwargs : dict
kwargs to be pased to :meth:`model_init`
'''
if processes is None:
try:
processes = multiprocessing.cpu_count()
except NotImplementedError:
processes = 1
ema_logging.info("nr of processes is "+str(processes))
# setup queues etc.
self._setup_queues()
self._taskqueue = queue.Queue(processes*2)
self._cache = {}
self._state = pool.RUN
# handling of logging
self.log_queue = multiprocessing.Queue()
h = ema_logging.NullHandler()
logging.getLogger(ema_logging.LOGGER_NAME).addHandler(h)
log_queue_reader = LogQueueReader(self.log_queue)
log_queue_reader.start()
# setup of the actual pool
self._pool = []
working_dirs = []
ema_logging.debug('generating workers')
worker_root = None
for i in range(processes):
ema_logging.debug('generating worker '+str(i))
workername = self._get_worker_name(i)
#setup working directories for parallel_ema
for msi in msis:
if msi.working_directory != None:
if worker_root == None:
wd = msis[0].working_directory
abs_wd = os.path.abspath(wd)
worker_root = os.path.dirname(abs_wd)
wd_name = workername + msi.name
working_directory = os.path.join(worker_root, wd_name)
working_dirs.append(working_directory)
shutil.copytree(msi.working_directory,
working_directory,
)
msi.set_working_directory(working_directory)
w = LoggingProcess(
self.log_queue,
level = logging.getLogger(ema_logging.LOGGER_NAME)\
.getEffectiveLevel(),
target=worker,
args=(self._inqueue,
self._outqueue,
msis,
kwargs
)
)
self._pool.append(w)
w.name = w.name.replace('Process', workername)
w.daemon = True
w.start()
ema_logging.debug(' worker '+str(i) + ' generated')
# thread for handling tasks
self._task_handler = threading.Thread(
target=CalculatorPool._handle_tasks,
name='task handler',
args=(self._taskqueue,
self._quick_put,
self._outqueue,
self._pool
)
)
self._task_handler.daemon = True
self._task_handler._state = pool.RUN
self._task_handler.start()
# thread for handling results
self._result_handler = threading.Thread(
target=CalculatorPool._handle_results,
name='result handler',
args=(self._outqueue,
self._quick_get,
self._cache,
self.log_queue)
)
self._result_handler.daemon = True
self._result_handler._state = pool.RUN
self._result_handler.start()
# function for cleaning up when finalizing object
self._terminate = Finalize(self,
self._terminate_pool,
args=(self._taskqueue,
self._inqueue,
self._outqueue,
self._pool,
self._task_handler,
self._result_handler,
self._cache,
working_dirs,
),
exitpriority=15
)
ema_logging.info("pool has been set up")