def model(par: Union[Sequence, Mapping]) -> Mapping:
"""The model function."""
# copy since we add fixed parameters
par = copy.deepcopy(par)
# convenience to allow calling model not only with dicts
if not isinstance(par, Mapping):
par = {key: val for key, val in zip(x_ids, par)}
# add fixed parameters
for key, val in zip(x_fixed_ids, x_fixed_vals):
par[key] = val
# simulate model
sim = simulate_petab(petab_problem=petab_problem,
amici_model=amici_model,
solver=amici_solver,
problem_parameters=par,
scaled_parameters=True)
# return values of interest
ret = {'llh': sim[LLH]}
if return_simulations:
for i_rdata, rdata in enumerate(sim[RDATAS]):
ret[f'y_{i_rdata}'] = rdata['y']
if return_rdatas:
ret[RDATAS] = sim[RDATAS]
return ret
def __call__(self, par: Union[Sequence, Mapping]) -> Mapping:
"""The model function.
Note: The parameters are assumed to be passed on prior scale.
"""
# copy since we add fixed parameters
par = copy.deepcopy(par)
# convenience to allow calling model not only with dicts
if not isinstance(par, Mapping):
par = {key: val for key, val in zip(self.x_free_ids, par)}
# add fixed parameters
for key, val in zip(self.x_fixed_ids, self.x_fixed_vals):
par[key] = val
# scale parameters whose priors are not on scale
for key in self.prior_scales.keys():
par[key] = rescale(
val=par[key],
origin_scale=self.prior_scales,
target_scale=self.scaled_scales,
)
# simulate model
sim = simulate_petab(
petab_problem=self.petab_problem,
amici_model=self.amici_model,
solver=self.amici_solver,
problem_parameters=par,
scaled_parameters=True,
)
# return values of interest
ret = {'llh': sim[LLH]}
if self.return_simulations:
for i_rdata, rdata in enumerate(sim[RDATAS]):
ret[f'y_{i_rdata}'] = rdata['y']
if self.return_rdatas:
ret[RDATAS] = sim[RDATAS]
return ret
def simulate_timecourse(
parent_petab_problem: petab.Problem,
timecourse_id: str,
solver_settings: Dict[str, Any],
problem_parameters: Dict[str, float] = None,
parameter_mapping=None,
sensi_orders: Tuple[int, ...] = (0, 1),
# FIXME Dict typehint
initial_states: Union[Dict, Tuple[float, ...]] = None,
model_settings: Dict[str, Any] = None,
#solver_customizer: Callable[[amici.Solver], None] = None
):
"""
Solver settings are a required attribute to ensure these are not
forgotten...
To use a default solver, supply an empty dictionary (`solver_settings={}`).
initial_states:
FIXME
Can supply a dictionary or a tuple.
"""
timecourse = get_timecourse(
petab_problem=parent_petab_problem,
timecourse_id=timecourse_id,
)
petab_problems = subset_petab_problem(
petab_problem=parent_petab_problem,
timecourse_id=timecourse_id,
)
#print([petab_problem.measurement_df for petab_problem in petab_problems])
results = []
#amici_model = import_petab_problem(petab_problems[0])
for index, petab_problem in enumerate(petab_problems):
if petab_problem.measurement_df.empty:
# FIXME resolve this properly. Is it due to timecourse pieces
# that are defined to occur after the last measured time
# point?
break
# FIXME uncomment, remove global `amici_model = ...` above
# Replace a controlled parameter with the ID of its timepoint-specific
# control parameter (hack to ensure sensitivities are computed correctly)
#print(1)
replace_timecourse_parameters = {
k: v
for k, v in petab_problem.condition_df.loc[timecourse_id].items()
if isinstance(v, str) and k != CONDITION_NAME
}
petab_problem.condition_df.drop(
replace_timecourse_parameters,
axis=1,
inplace=True,
)
for old_parameter_id, new_parameter_id in replace_timecourse_parameters.items(
):
if old_parameter_id in problem_parameters:
raise ValueError(
'A timecourse parameter was assigned a value via '
'`problem_parameters`, but should receive its value '
'from the timecourse information.')
if new_parameter_id not in problem_parameters:
raise ValueError(
'Please supply a value for estimated timecourse '
'replacement parameters (parameter IDs in the rows) '
'of the condition table of a PEtab Timecourse problem.')
problem_parameters[old_parameter_id] = \
problem_parameters[new_parameter_id]
# Duplicate control parameter in parameter_df, new index value is the
# controlled parameter, such that sensitivities are output for the
# controlled parameter (and can be interpreted as the sensitivities for
# the control parameter)
petab_problem.parameter_df.loc[old_parameter_id] = \
petab_problem.parameter_df.loc[new_parameter_id]
#print(2)
amici_model = import_petab_problem(petab_problem)
amici_model.setT0(float(timecourse[index][0]))
#if initial_states is not None:
# breakpoint()
#print(f'\n\nTime: {amici_model.t0()}\n\n')
amici_edatas = create_edatas(amici_model, petab_problem)
if problem_parameters is not None:
# FIXME temp fix to add into parameters that AMICI automatically
# sets to be estimated from the SBML model (that weren't
# fixed like parameters in a condition table)
model_parameters = dict(
zip(amici_model.getParameterIds(),
amici_model.getParameters()))
#assert parameter_mapping is not None
#print(problem_parameters)
#print(petab_problem.measurement_df)
if parameter_mapping is None:
parameter_mapping = \
create_identity_parameter_mapping(amici_model, 1)
# Remove parameters from problem parameters if they are already
# specified by the timecourse.
# May break if a problem parameter is named `'conditionName'`.
subset_problem_parameters = {
parameter_id: parameter_value
for parameter_id, parameter_value in {
**model_parameters,
**problem_parameters
}.items()
#for parameter_id, parameter_value in problem_parameters.items()
if parameter_id not in petab_problem.condition_df.columns
}
removed_problem_parameters = \
set(problem_parameters).difference(subset_problem_parameters)
warnings.warn(
'The following parameters were removed from the supplied '
'`problem_parameters`, as they are already specified by the '
f'timecourse: {removed_problem_parameters}')
#notpositive = {
# k: v
# for k, v in subset_problem_parameters.items()
# if v <= 0
#}
#print(f'not positive parameters: {notpositive}') # FIXME
amici.parameter_mapping.fill_in_parameters(
edatas=amici_edatas,
problem_parameters=subset_problem_parameters,
scaled_parameters=True,
parameter_mapping=parameter_mapping,
amici_model=amici_model,
)
else:
subset_problem_parameters = None
#amici_model.setParameterById(problem_parameters)
#parameters = [
# problem_parameters[parameter_id]
# for parameter_id in amici_model.getParameterIds()
#]
#amici_edatas = create_edatas(amici_model, petab_problem)
#one(amici_edatas).parameters = parameters
#print(3)
if results:
one(amici_edatas).x0 = one(results[-1]['rdatas']).x[-1].flatten()
one(amici_edatas).sx0 = one(results[-1]['rdatas']).sx[-1].flatten()
elif initial_states is not None:
# TODO untested
#print(initial_states)
#print(amici_model.getStateIds())
if isinstance(initial_states, dict):
indexed_initial_states = [
initial_states[state_id]
for state_id in amici_model.getStateIds()
]
else:
indexed_initial_states = initial_states
#print(indexed_initial_states)
#print(amici_model.getStateIds())
#print(4)
one(amici_edatas).x0 = indexed_initial_states
#print(5)
amici_solver = amici_model.getSolver()
# TODO allow a user to specify these settings
if model_settings is not None:
for setter, value in model_settings.items():
getattr(amici_model, setter)(value)
if solver_settings is not None:
for setter, value in solver_settings.items():
getattr(amici_solver, setter)(value)
#print(4)
#amici_solver.setSensitivityOrder(amici.SensitivityOrder_first)
#amici_solver.setSensitivityMethod(amici.SensitivityMethod_forward)
#amici_solver.setMaxSteps(int(1e6))
#solver_settings = {
# 'setSensitivityOrder': amici.SensitivityOrder_first,
# 'setSensitivityMethod': amici.SensitivityMethod_forward,
# 'setMaxSteps': int(1e6),
# 'setMaxTime': 60,
#}
#amici_solver.setAbsoluteTolerance(1e-8)
#amici_solver.setRelativeTolerance(1e-6)
#amici_solver.setAbsoluteToleranceFSA(1e-8)
#amici_solver.setRelativeToleranceFSA(1e-6)
#print(one(amici_edatas).parameters)
#import functools
#sp = lambda x: simulate_petab(
# petab_problem=petab_problem,
# amici_model=amici_model,
# solver=amici_solver,
# problem_parameters=x,
#)
results.append(
simulate_petab(
petab_problem=petab_problem,
amici_model=amici_model,
solver=amici_solver,
edatas=amici_edatas,
problem_parameters=subset_problem_parameters,
))
#print(5)
#print(problem_parameters)
#print(results)
if initial_states is not None:
pass
for old_parameter_id, new_parameter_id in replace_timecourse_parameters.items(
):
results[-1]['sllh'][new_parameter_id] = \
results[-1]['sllh'][old_parameter_id]
# Was artifically added, so remove now
del problem_parameters[old_parameter_id]
return results
yaml_file = os.path.join(case_dir, petabtests.problem_yaml_name(case))
problem = PysbPetabProblem.from_yaml(yaml_file,
flatten=case.startswith('0006'))
else:
raise ValueError(f"Unsupported model_type: {model_type}")
# compile amici model
if case.startswith('0006') and model_type != "pysb":
petab.flatten_timepoint_specific_output_overrides(problem)
model_output_dir = f'amici_models/model_{case}'
model = import_petab_problem(problem,
model_output_dir=model_output_dir,
force_compile=True)
# simulate
ret = simulate_petab(problem, model, log_level=logging.DEBUG)
rdatas = ret['rdatas']
chi2 = sum(rdata['chi2'] for rdata in rdatas)
llh = ret['llh']
simulation_df = rdatas_to_measurement_df(rdatas, model,
problem.measurement_df)
petab.check_measurement_df(simulation_df, problem.observable_df)
simulation_df = simulation_df.rename(
columns={petab.MEASUREMENT: petab.SIMULATION})
simulation_df[petab.TIME] = simulation_df[petab.TIME].astype(int)
solution = petabtests.load_solution(case, model_type)
gt_chi2 = solution[petabtests.CHI2]
gt_llh = solution[petabtests.LLH]
gt_simulation_dfs = solution[petabtests.SIMULATION_DFS]
if case.startswith('0006'):
def _test_case(case):
"""Run a single PEtab test suite case"""
case = petabtests.test_id_str(case)
logger.debug(f"Case {case}")
# load
case_dir = os.path.join(petabtests.CASES_DIR, case)
# import petab problem
yaml_file = os.path.join(case_dir, petabtests.problem_yaml_name(case))
problem = petab.Problem.from_yaml(yaml_file)
# compile amici model
model_output_dir = f'amici_models/model_{case}'
model = import_petab_problem(
problem, model_output_dir=model_output_dir)
# simulate
chi2s_match = llhs_match = simulations_match = False
ret = simulate_petab(problem, model, log_level=logging.DEBUG)
rdatas = ret['rdatas']
chi2 = sum(rdata['chi2'] for rdata in rdatas)
llh = ret['llh']
simulation_df = rdatas_to_measurement_df(rdatas, model,
problem.measurement_df)
petab.check_measurement_df(simulation_df, problem.observable_df)
simulation_df = simulation_df.rename(
columns={petab.MEASUREMENT: petab.SIMULATION})
simulation_df[petab.TIME] = simulation_df[petab.TIME].astype(int)
solution = petabtests.load_solution(case)
gt_chi2 = solution[petabtests.CHI2]
gt_llh = solution[petabtests.LLH]
gt_simulation_dfs = solution[petabtests.SIMULATION_DFS]
tol_chi2 = solution[petabtests.TOL_CHI2]
tol_llh = solution[petabtests.TOL_LLH]
tol_simulations = solution[petabtests.TOL_SIMULATIONS]
chi2s_match = petabtests.evaluate_chi2(chi2, gt_chi2, tol_chi2)
llhs_match = petabtests.evaluate_llh(llh, gt_llh, tol_llh)
simulations_match = petabtests.evaluate_simulations(
[simulation_df], gt_simulation_dfs, tol_simulations)
logger.log(logging.DEBUG if chi2s_match else logging.ERROR,
f"CHI2: simulated: {chi2}, expected: {gt_chi2},"
f" match = {chi2s_match}")
logger.log(logging.DEBUG if simulations_match else logging.ERROR,
f"LLH: simulated: {llh}, expected: {gt_llh}, "
f"match = {llhs_match}")
logger.log(logging.DEBUG if simulations_match else logging.ERROR,
f"Simulations: match = {simulations_match}")
# FIXME case 7 fails due to #963
if case not in ['0007', '0016']:
check_derivatives(problem, model)
# FIXME case 7 fails due to #963
if not all([llhs_match, simulations_match]) \
or (not chi2s_match and case not in ['0007', '0016']):
# chi2s_match ignored until fixed in amici
logger.error(f"Case {case} failed.")
raise AssertionError(f"Case {case}: Test results do not match "
"expectations")
logger.info(f"Case {case} passed.")
def main():
"""Simulate the model specified on the command line"""
args = parse_cli_args()
if args.verbose:
logger.setLevel(logging.DEBUG)
logger.info(f"Simulating '{args.model_name}' "
f"({args.model_directory}) using PEtab data from "
f"{args.yaml_file_name}")
# load PEtab files
problem = petab.Problem.from_yaml(args.yaml_file_name)
petab.flatten_timepoint_specific_output_overrides(problem)
# load model
if args.model_directory:
sys.path.insert(0, args.model_directory)
model_module = importlib.import_module(args.model_name)
amici_model = model_module.getModel()
res = simulate_petab(petab_problem=problem,
amici_model=amici_model,
log_level=logging.DEBUG)
rdatas = res[RDATAS]
llh = res[LLH]
# create simulation PEtab table
sim_df = rdatas_to_measurement_df(rdatas=rdatas,
model=amici_model,
measurement_df=problem.measurement_df)
sim_df.rename(columns={petab.MEASUREMENT: petab.SIMULATION}, inplace=True)
if args.simulation_file:
sim_df.to_csv(index=False, sep="\t")
if args.plot:
try:
# visualize fit
axs = plot_petab_problem(petab_problem=problem, sim_data=sim_df)
# save figure
for plot_id, ax in axs.items():
fig_path = os.path.join(
args.model_directory,
args.model_name + "_" + plot_id + "_vis.png")
logger.info(f"Saving figure to {fig_path}")
ax.get_figure().savefig(fig_path, dpi=150)
except NotImplementedError:
pass
if args.check:
references_yaml = os.path.join(os.path.dirname(__file__),
"benchmark_models.yaml")
with open(references_yaml) as f:
refs = yaml.full_load(f)
try:
ref_llh = refs[args.model_name]["llh"]
logger.info(f"Reference llh: {ref_llh}")
if abs(ref_llh - llh) < 1e-3:
logger.info(f"Computed llh {llh} matches reference "
f"{ref_llh}. Absolute difference is "
f"{ref_llh - llh}.")
else:
logger.error(f"Computed llh {llh} does not match reference "
f"{ref_llh}. Absolute difference is "
f"{ref_llh - llh}."
f" Relative difference is {llh / ref_llh}")
sys.exit(1)
except KeyError:
logger.error("No reference likelihood found for "
f"{args.model_name} in {references_yaml}")
