def test_get_loss(self, extended_simulator, t, metric, handle_CVodeError):
# Create some measurement objects from predictions, i.e. create some artifical data
predicitions = extended_simulator.simulate(t=t)
measurements = [
Measurement(name=_prediction.name,
timepoints=_prediction.timepoints,
values=_prediction.values,
errors=np.ones_like(_prediction.values))
for _prediction in predicitions
]
extended_simulator._get_loss(metric=metric, measurements=measurements)
# Loss will be nan in case no relevant measurements are provided, i.e. measurements for states that are not defined
assert np.isnan(
extended_simulator._get_loss(metric=metric,
measurements=[
Measurement(name='y1000',
timepoints=[100, 200],
values=[1, 2],
errors=[10, 20]),
]))
# Get loss for other parameters, as a minimizer would do several times
_params = extended_simulator.get_all_parameters()
different_params = {_p: _params[_p] * 0.95 for _p in _params}
extended_simulator._get_loss_for_minimzer(
metric=metric,
guess_dict=different_params,
measurements=measurements,
handle_CVodeError=handle_CVodeError,
verbosity_CVodeError=False,
)
class StaticHelpers():
data_single = [Measurement(name='y0', timepoints=[1, 2, 3], values=[10, 10 ,10], errors=[0.1, 0.2, 0.3])]
data_multi = [
Measurement(name='y0', timepoints=[1, 2, 3], values=[10, 10 ,10], errors=[0.1, 0.2, 0.3], replicate_id='1st'),
Measurement(name='y0', timepoints=[1, 5, 10], values=[10, 10 ,10], errors=[0.1, 0.2, 0.3], replicate_id='2nd'),
]
many_time_series_1 = [
[TimeSeries(name='T1', timepoints=[1, 2, 3, 4], values=[10, 10, 10, 10], replicate_id='1st')],
[TimeSeries(name='T1', timepoints=[1, 2, 3], values=[10, 10, 10], replicate_id='1st')],
]
unknowns = ['y00', 'y10']
bounds = [(-100, 100), (-100, 100)]
class StaticDatatypes():
timeseries = TimeSeries(
name=StaticHelpers.name,
timepoints=StaticHelpers.timepoints,
values=StaticHelpers.values,
replicate_id=StaticHelpers.replicate_id
)
modelstate = ModelState(
name=StaticHelpers.name,
timepoints=StaticHelpers.timepoints,
values=StaticHelpers.values,
replicate_id=StaticHelpers.replicate_id
)
measurement_wo_errs = Measurement(
name=StaticHelpers.name,
timepoints=StaticHelpers.timepoints,
values=StaticHelpers.values,
replicate_id=StaticHelpers.replicate_id
)
measurement_w_errs = Measurement(
name=StaticHelpers.name,
timepoints=StaticHelpers.timepoints,
values=StaticHelpers.values,
errors=StaticHelpers.errors,
replicate_id=StaticHelpers.replicate_id
)
observation = Observation(
name=StaticHelpers.name,
timepoints=StaticHelpers.timepoints,
values=StaticHelpers.values,
observed_state=StaticHelpers.state,
replicate_id=StaticHelpers.replicate_id
)
sensitivity = Sensitivity(
timepoints=StaticHelpers.timepoints,
values=StaticHelpers.values,
response=StaticHelpers.state,
parameter=StaticHelpers.parameter,
replicate_id=StaticHelpers.replicate_id
)
class StaticHelpers():
data_single = [
Measurement(name='y0',
timepoints=[1, 2, 3],
values=[10, 20, 30],
errors=[0.1, 0.2, 0.3])
]
data_multi = [
Measurement(name='y0',
timepoints=[1, 2, 3],
values=[10, 20, 30],
errors=[0.1, 0.2, 0.3],
replicate_id='1st')
]
unknowns = ['y00', 'y10']
bounds = [(-100, 100), (-100, 100)]
class StaticData():
measurements_wo_errors = Measurement(name='M1',
timepoints=[1, 2, 3],
values=[10, 20, 30],
replicate_id='1st')
measurements_w_errors = Measurement(name='M2',
timepoints=[2, 3, 4],
values=[20, 30, 40],
errors=[1 / 20, 1 / 30, 1 / 40],
replicate_id='1st')
time_series_1 = TimeSeries(name='TS1',
timepoints=[1, 2],
values=[10, 20],
replicate_id='1st')
time_series_2 = TimeSeries(name='TS2',
timepoints=[2, 3],
values=[20, 30],
replicate_id='2nd')
def test_with_model_enforcing_CVodeError(self):
name = 'model06'
modelclass = ModelLibrary.modelclasses[name]
model_parameters = ModelLibrary.model_parameters[name]
initial_values = ModelLibrary.initial_values[name]
extended_simulator = ExtendedSimulator(
bioprocess_model_class=modelclass,
model_parameters=model_parameters,
initial_values=initial_values)
# The chosen model will create an integration error for rate = 0. A RuntimeWarning is tehrefore raised before the CVodeError is raised
with pytest.warns(RuntimeWarning):
with pytest.raises(CVodeError):
extended_simulator.simulate(t=24, parameters={'rate0': 0})
with pytest.warns(RuntimeWarning):
with pytest.raises(CVodeError):
extended_simulator._get_loss_for_minimzer(
metric='negLL',
guess_dict={'rate0': 0},
measurements=[
Measurement(name='y0',
timepoints=[1, 2, 3],
values=[10, 20, 30]),
],
handle_CVodeError=False,
verbosity_CVodeError=False,
)
# For toxic parameters causing integration errors, CVodeError handling results in inf loss
with pytest.warns(RuntimeWarning):
assert np.isinf(
extended_simulator._get_loss_for_minimzer(
metric='negLL',
guess_dict={'rate0': 0},
measurements=[
Measurement(name='y0',
timepoints=[1, 2, 3],
values=[10, 20, 30]),
],
handle_CVodeError=True,
verbosity_CVodeError=True,
))
def test_init_datatypes(self, values, errors, info, replicate_id):
"""
To test typical instantiations of datatypes.
"""
# Standard instatiations
TimeSeries(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=values, info=info, replicate_id=replicate_id)
ModelState(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=values, info=info, replicate_id=replicate_id)
Measurement(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=values, errors=errors, info=info, replicate_id=replicate_id)
Observation(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=values, observed_state='y1', replicate_id=replicate_id)
Sensitivity(timepoints=StaticHelpers.timepoints, values=values, response='y1', parameter='p1', replicate_id=replicate_id)
Sensitivity(timepoints=StaticHelpers.timepoints, values=values, response='y1', parameter='p1', h=1e-8, replicate_id=replicate_id)
# Must provide timepoints
with pytest.raises(ValueError):
TimeSeries(name=StaticHelpers.name, timepoints=None, values=values)
# Must provide values
with pytest.raises(ValueError):
TimeSeries(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=None)
# Measurements can be created with error_models
Measurement(
name=StaticHelpers.name,
timepoints=StaticHelpers.timepoints,
values=values,
error_model=StaticErrorModelHelpers.constant_error_model,
error_model_parameters=StaticErrorModelHelpers.constant_error_model_parameters,
)
# Must provide a subclass of rvs.continous as p.d.f.
with pytest.raises(ValueError):
Measurement(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=values, error_distribution=scipy.stats.bernoulli)
# Error values must be >0
with pytest.raises(ValueError):
Measurement(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=values, errors=[0]*len(StaticHelpers.values))
with pytest.raises(ValueError):
Measurement(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=values, errors=[-1]*len(StaticHelpers.values))
def test_init_datatypes_masking_non_numeric(self, input_vector, masked_vector):
"""
Non-numeric values implicitly define a mask, which is in turn applied to all vectors of the corresponding datatypes.
"""
_timeseries = TimeSeries(name=StaticHelpers.name, timepoints=input_vector, values=StaticHelpers.values)
assert all(_timeseries.timepoints) == all(masked_vector)
assert _timeseries.length == len(masked_vector)
_timeseries = TimeSeries(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=input_vector)
assert all(_timeseries.timepoints) == all(masked_vector)
assert _timeseries.length == len(masked_vector)
_measurement = Measurement(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=StaticHelpers.values, errors=input_vector)
assert all(_measurement.timepoints) == all(masked_vector)
assert _measurement.length == len(masked_vector)
def test_update_error_models_parameters(self, error_model, error_model_parameters):
"""
Updates error_models for existing Measurement objects
"""
# create measurement first
measurement = Measurement(name=StaticHelpers.name, timepoints=StaticHelpers.timepoints, values=StaticHelpers.values)
# To use a different (new) error_model, it must be passed with its corresponding error_model_parameters
measurement.update_error_model(error_model=error_model, error_model_parameters=error_model_parameters)
# Parameter values can be updated, as long as all parameters are present in the new dictionary
measurement.error_model_parameters = {_p : error_model_parameters[_p]*1.5 for _p in error_model_parameters}
# Incase the error model is applied, a warning can be given for overwriting the error vector
with pytest.warns(UserWarning):
measurement.apply_error_model(report_level=1)
# Setting new parameter values won't work
with pytest.raises(KeyError):
measurement.error_model_parameters = {'bad_parameter' : 1000}
def test_extended_simulator_with_observations(self):
# Get building blocks for BioprocessModel
name = 'model01'
modelclass = ModelLibrary.modelclasses[name]
initial_values = ModelLibrary.initial_values[name]
model_parameters = ModelLibrary.model_parameters[name]
# Get building blocks for ObservationFunctions
obsfun_name = 'obsfun01'
obsfun = ObservationFunctionLibrary.observation_functions[obsfun_name]
obsfun_parameters = ObservationFunctionLibrary.observation_function_parameters[
obsfun_name]
observed_state = ObservationFunctionLibrary.observed_states[
obsfun_name]
obsfuns_params = [(obsfun, {
**obsfun_parameters, 'observed_state':
observed_state
})]
# Set new values for parameters, using an extended simulator
extended_simulator = ExtendedSimulator(
bioprocess_model_class=modelclass,
model_parameters=model_parameters,
initial_values=initial_values,
observation_functions_parameters=obsfuns_params,
)
params = extended_simulator.get_all_parameters()
extended_simulator.set_parameters(
{_p: params[_p] * 1.05
for _p in params})
# Get some prediction to be used as artifical data
predicitions = extended_simulator.simulate(t=24)
measurements = [
Measurement(name=_prediction.name,
timepoints=_prediction.timepoints,
values=_prediction.values,
errors=np.ones_like(_prediction.values))
for _prediction in predicitions
]
extended_simulator._get_loss(metric='negLL', measurements=measurements)
def test_other_distributions(self):
# Create a Measurement object having an error t-distribution
measurement_t = Measurement(
name=StaticHelpers.name,
timepoints=StaticHelpers.timepoints,
values=StaticHelpers.values,
errors=StaticHelpers.values,
error_distribution=scipy.stats.t,
distribution_kwargs={'df' : 1},
replicate_id='1st'
)
# Get rvs values for, e.g. MC sampling
measurement_t._get_random_samples_values()
loss_1 = measurement_t.get_loss(metric='negLL', predictions=[StaticDatatypes.modelstate])
assert not np.isnan(loss_1)
loss_2 = measurement_t.get_loss(metric='negLL', predictions=[StaticDatatypes.modelstate], distribution_kwargs={'df' : 100})
assert not np.isnan(loss_2)
# Does not work in case there are no errors
with pytest.raises(AttributeError):
StaticDatatypes.measurement_wo_errs._get_random_samples_values()