def get_relative_risks(config: Path, input_draw: int, random_seed: int, age_group_id: int) -> pd.DataFrame:
sim = InteractiveContext(config, setup=False)
artifact_path = sim.configuration.input_data.artifact_path
artifact = Artifact(artifact_path)
age_bins = artifact.load(data_keys.POPULATION.AGE_BINS).reset_index().set_index('age_group_id')
age_start = age_bins.loc[age_group_id, 'age_start']
age_end = age_bins.loc[age_group_id, 'age_end']
year_start = 2019
year_end = 2020
sim.configuration.update({
'input_data': {
'input_draw_number': input_draw,
},
'randomness': {
'random_seed': random_seed,
},
'population': {
'age_start': age_start,
'age_end': age_end
}
})
sim.setup()
pop = sim.get_population()
gestational_ages = sim.get_value('short_gestation.exposure')(pop.index)
birth_weights = sim.get_value('low_birth_weight.exposure')(pop.index)
interpolators = artifact.load(data_keys.LBWSG.RELATIVE_RISK_INTERPOLATOR)
def calculate_rr_by_sex(sex: str) -> float:
sex_mask = pop['sex'] == sex
row_index = (sex, age_start, age_end, year_start, year_end, 'diarrheal_diseases', 'excess_mortality_rate')
interpolator = pickle.loads(bytes.fromhex(
interpolators.loc[row_index, f'draw_{input_draw}']
))
rrs = np.exp(interpolator(gestational_ages[sex_mask], birth_weights[sex_mask], grid=False))
return rrs
lbwsg_rrs = pd.DataFrame({'relative_risk': 1.0}, index=pop.index)
lbwsg_rrs['sex'] = pop['sex']
lbwsg_rrs.loc[lbwsg_rrs['sex'] == 'Female', 'relative_risk'] = calculate_rr_by_sex('Female')
lbwsg_rrs.loc[lbwsg_rrs['sex'] == 'Male', 'relative_risk'] = calculate_rr_by_sex('Male')
return lbwsg_rrs
def test_acmr():
sim = InteractiveContext(
'src/vivarium_conic_lsff/model_specifications/india.yaml')
sim.step() # make sure everything is setup
pop = sim.get_population()
age_groups = pd.cut(pop.age,
bins=[0, 7 / 365, 28 / 365, 1, 5],
right=False)
mr_pipeline = sim.get_value('mortality_rate')
acmr_orig = mr_pipeline.source(pop.index).sum(axis=1)
acmr_w_risk = mr_pipeline(
pop.index).sum(axis=1) * 365 # convert back to "per person-year"
# confirm that they are *not* identical at the individual level
assert not np.allclose(
acmr_orig, acmr_w_risk,
rtol=.05), 'expect acmr to be quite different for some individuals'
# but close at pop level
assert np.allclose(
acmr_orig.groupby(age_groups).median(),
acmr_w_risk.groupby(age_groups).median(),
rtol=.1
), 'expect acmr to be within 10% of original at population level'
