def cause_age_sex_agg(death_df, true_paf_df, amen_paf_df, draw_parameters,
uhc_version_dir):
'''
Aggregate PAFs over age and sex, and collapse deaths.
'''
# convert to count space
true_paf_df = misc.draw_math([death_df, true_paf_df], specs.ID_COLS +
['age_group_id', 'sex_id', 'cause_id'],
specs.DRAW_COLS, '*')
amen_paf_df = misc.draw_math([death_df, amen_paf_df], specs.ID_COLS +
['age_group_id', 'sex_id', 'cause_id'],
specs.DRAW_COLS, '*')
# agg cause and sex
true_paf_df = true_paf_df.groupby(specs.ID_COLS + ['age_group_id'],
as_index=False)[specs.DRAW_COLS].sum()
amen_paf_df = amen_paf_df.groupby(specs.ID_COLS + ['age_group_id'],
as_index=False)[specs.DRAW_COLS].sum()
death_df = death_df.groupby(specs.ID_COLS + ['age_group_id'],
as_index=False)[specs.DRAW_COLS].sum()
# now that everything is both sex space, reassign sex_id
true_paf_df['sex_id'] = 3
amen_paf_df['sex_id'] = 3
death_df['sex_id'] = 3
# age-standardize
# set counts = False even though we are passing in counts... don't want to convert these to rates (yet)
true_paf_df = misc.age_standardize(true_paf_df,
specs.ID_COLS,
specs.DRAW_COLS,
draw_parameters,
uhc_version_dir,
counts=False)
amen_paf_df = misc.age_standardize(amen_paf_df,
specs.ID_COLS,
specs.DRAW_COLS,
draw_parameters,
uhc_version_dir,
counts=False)
death_df = misc.age_standardize(death_df,
specs.ID_COLS,
specs.DRAW_COLS,
draw_parameters,
uhc_version_dir,
counts=False)
# convert back to PAF space
true_paf_df = misc.draw_math([true_paf_df, death_df], specs.ID_COLS,
specs.DRAW_COLS, '/')
amen_paf_df = misc.draw_math([amen_paf_df, death_df], specs.ID_COLS,
specs.DRAW_COLS, '/')
return death_df, true_paf_df, amen_paf_df
def delete_risk(df, paf_df, index_cols, draw_cols):
'''
Remove effect of local risk exposure.
'''
# get local unattribtable
paf_df[draw_cols] = 1 - paf_df[draw_cols]
df = misc.draw_math([df, paf_df], index_cols, draw_cols, '*')
return df
def draw_divide(num_df, denom_df):
'''
Get the weight value for each uhc_id.
'''
# print uhc_id for ease of debugging if loop breaks
print(pd.unique(num_df['uhc_id']))
df = misc.draw_math([num_df, denom_df], specs.ID_COLS, specs.DRAW_COLS,
'/')
df['uhc_id'] = num_df['uhc_id'].tolist()
return df
def calc_counterfactual_burden():
'''
For a given service/population cell, get the risk-adjusted death rate for
relevant indicators.
'''
parser = argparse.ArgumentParser()
parser.add_argument('--uhc_id',
help='Indicates tracer-service_population',
type=int)
parser.add_argument('--uhc_version',
help='Version number for run.',
type=int)
parser.add_argument('--value_type', help='What are we storing,', type=str)
args = parser.parse_args()
# get tracers and service pop
uhc_version_dir = FILEPATH
# get efficacy tier
uhc_df = pd.read_excel(FILEPATH)
uhc_df = uhc_df.query("uhc_id == {}".format(args.uhc_id))
efficacy = 1. - 0.2 * uhc_df['efficacy_tier'].values.item() + 0.1
# read coverage and observed burden draws
coverage_df = pd.read_hdf(FILEPATH)
burden_df = pd.read_hdf(FILEPATH)
# perform calculation:
# counterfactual 0 is the burden we'd see if not for the intervention
# couterfactual0 = observed / (1 - coverage * efficacy)
# counterfactual 1 is the burden we'd see with 100% coverage of the intervention
# couterfactual1 = couterfactual0 * (1 - 1 * efficacy)
# the health gain weight is the difference
# health gain weight = couterfactual0 - couterfactual1
if len(coverage_df) + len(burden_df) == 0:
summary_df = pd.DataFrame(columns=specs.ID_COLS +
['mean', 'lower', 'upper'])
else:
coverage_df[specs.DRAW_COLS] = (
1 - coverage_df[specs.DRAW_COLS] * efficacy)
burden_df = misc.draw_math([burden_df, coverage_df], specs.ID_COLS,
specs.DRAW_COLS, '/')
burden_df[specs.DRAW_COLS] = burden_df[specs.DRAW_COLS] - (
burden_df[specs.DRAW_COLS] * (1 - efficacy))
burden_df['efficacy'] = efficacy
summary_df = misc.summarize(burden_df, specs.DRAW_COLS)
# store
burden_df.to_hdf(FILEPATH)
summary_df.to_csv(FILEPATH)
def fetch_ratio_draws(draw_parameters, uhc_version_dir, uhc_id, **kwargs):
'''
required kwargs:
gbd_id ([int]) = ids for `get_draws` call
gbd_id_type ([str]) = types associated w/ each id.
measure_id (int) = What is the measure of the indicator.
'''
nf_df = fetch_outputs_draws(draw_parameters, uhc_version_dir, uhc_id,
**kwargs)
kwargs['measure_id'] = 1
death_df = fetch_outputs_draws(draw_parameters, uhc_version_dir, uhc_id,
**kwargs)
# get ratio
df = misc.draw_math([death_df, nf_df], specs.ID_COLS, specs.DRAW_COLS, '/')
return df[specs.ID_COLS + specs.DRAW_COLS]
def add_global_risk(df, gpaf_df, index_cols, draw_cols):
'''
Add effect of global risk exposure.
'''
# get global unattribtable
assert index_cols == specs.ID_COLS, 'Assumes location and year as index'
# expand by location-year, average over year, then add year col back on
gpaf_df = gpaf_df.drop('location_id', axis=1)
gpaf_df = gpaf_df.merge(df[index_cols])
gpaf_df = gpaf_df.groupby('location_id',
as_index=False)[specs.DRAW_COLS].mean()
gpaf_df = gpaf_df.merge(df[index_cols])
gpaf_df[draw_cols] = 1 - gpaf_df[draw_cols]
df = misc.draw_math([df, gpaf_df], index_cols, draw_cols, '/')
return df
def fetch_mmr_draws(draw_parameters, uhc_version_dir, uhc_id, **kwargs):
'''
required kwargs:
gbd_id ([int]) = ids for `get_draws` call
gbd_id_type ([str]) = types associated w/ each id.
measure_id (int) = What is the measure of the indicator.
'''
# NOTE: need to break age/sex params, set up for all-ages (for because of met need)
draw_parameters['age_group_id'] = range(7, 16)
draw_parameters['sex_id'] = [2]
# load age-standardized draws for maternal deaths and births
death_df = fetch_outputs_draws(draw_parameters, uhc_version_dir, uhc_id,
**kwargs)
births_df = fetch_asfr_draws(draw_parameters, uhc_version_dir)
# calc age-standardized MMR
df = misc.draw_math([death_df, births_df], specs.ID_COLS, specs.DRAW_COLS,
'/')
return df[specs.ID_COLS + specs.DRAW_COLS]
def produce_uhc():
parser = argparse.ArgumentParser()
parser.add_argument('--uhc_id',
help='0 for UHC service coverage aggregate.',
type=int)
parser.add_argument('--uhc_version',
help='Version number for run.',
type=int)
parser.add_argument('--value_type', help='What are we storing,', type=str)
args = parser.parse_args()
# get service_proxys and service pop
uhc_version_dir = FILEPATH
# retrieve the IDs we need
uhc_ids = get_uhc_ids()
# calculate health gain weight fraction (health gain / sum of health gains)
count_dfs = uhc_io.compile_dfs('counterfactual_burden', uhc_ids,
uhc_version_dir)
total_df = pd.concat(count_dfs)
total_df = total_df.groupby(specs.ID_COLS,
as_index=False)[specs.DRAW_COLS].sum()
weight_dfs = [draw_divide(count_df, total_df) for count_df in count_dfs]
weight_df = pd.concat(weight_dfs)
weight_df['mean_weight'] = weight_df[specs.DRAW_COLS].mean(axis=1)
# output the unadjusted weights
summary_unadjusted_weights = misc.summarize(weight_df.copy(),
specs.DRAW_COLS)
summary_unadjusted_weights.to_csv(FILEPATH)
weight_df.to_csv(FILEPATH)
# adjust the weights -- take mean weight of indicator in a specified number of bands within a country and year
weight_df['mean_weight'] = weight_df['mean_weight'].replace(0, np.nan)
num_bands = 3
weight_df['weight_band'] = weight_df.groupby(
['location_id', 'year_id']).mean_weight.transform(
lambda x: pd.qcut(x, num_bands, labels=range(1, num_bands + 1)))
weight_df['weight_band'] = weight_df['weight_band'].replace(np.nan, 0)
weight_df[specs.DRAW_COLS] = weight_df.groupby(
['location_id', 'year_id',
'weight_band'])[specs.DRAW_COLS].transform('mean')
weight_df = weight_df.drop(['weight_band', 'mean_weight'], axis=1)
# apply weight
cov_dfs = uhc_io.compile_dfs('coverage', uhc_ids, uhc_version_dir)
cov_df = pd.concat(cov_dfs)
# weight_df = pd.concat(weight_dfs)
uhcw_df = misc.draw_math([cov_df, weight_df], specs.ID_COLS + ['uhc_id'],
specs.DRAW_COLS, '*')
uhcw_df = uhcw_df.groupby(specs.ID_COLS,
as_index=False)[specs.DRAW_COLS].sum()
uhca_df = cov_df.groupby(specs.ID_COLS,
as_index=False)[specs.DRAW_COLS].mean()
# summarize and store...
## WEIGHTS
summaryw_df = misc.summarize(weight_df, specs.DRAW_COLS)
weight_df.to_hdf(FILEPATH)
summaryw_df.to_csv(FILEPATH)
## WEIGHTED VALUES
summaryuhcw_df = misc.summarize(uhcw_df, specs.DRAW_COLS)
uhcw_df.to_hdf(FILEPATH)
summaryuhcw_df.to_csv(FILEPATH)
## AVERAGE VALUES
summaryuhca_df = misc.summarize(uhca_df, specs.DRAW_COLS)
uhca_df.to_hdf(FILEPATH)
summaryuhca_df.to_csv(FILEPATH)
def fetch_art_draws(draw_parameters, uhc_version_dir, uhc_id, **kwargs):
# load age- and sex-specific data
_fetch_art_draws_loc = functools.partial(
fetch_art_draws_loc, age_group_id=draw_parameters['age_group_id'])
pool = Pool(30)
dfs = pool.map(_fetch_art_draws_loc, draw_parameters['location_id'])
pool.close()
pool.join()
df = pd.concat(dfs)
# coverage is by sex, age, loc, year when we read it in. we need to take a
# few steps to get rid of that level of specificity, we need coverage by
# location and year, NOT by loc, year, age, and sex.
# load prevalence and aggregate
print("""
prev_df = get_draws(
source='como', gbd_round_id=GBD_ROUND, version_id={id}, num_workers=30,
metric_id=3,
location_id=draw_parameters['location_id'],
year_id=draw_parameters['year_id'],
age_group_id=draw_parameters['age_group_id'],
sex_id=[1, 2],
decomp_step=DECOMP_STEP,
**kwargs {kwargs}
)
""".format(id=COMO_VERSION_ID, kwargs=kwargs))
prev_df = get_draws(source='como',
gbd_round_id=GBD_ROUND,
version_id=COMO_VERSION_ID,
num_workers=30,
metric_id=3,
location_id=draw_parameters['location_id'],
year_id=draw_parameters['year_id'],
age_group_id=draw_parameters['age_group_id'],
sex_id=[1, 2],
decomp_step=DECOMP_STEP,
**kwargs)
prev_df = prev_df[['location_id', 'year_id', 'age_group_id', 'sex_id'] +
specs.DRAW_COLS]
pop_df = pd.read_hdf(FILEPATH)
prev_df = prev_df.merge(pop_df[[
'location_id', 'year_id', 'age_group_id', 'sex_id', 'population'
]])
# multiply prevalence (proportion) by population to get number of people with
# HIV/AIDS
prev_df[specs.DRAW_COLS] = (prev_df[specs.DRAW_COLS].values.transpose() *
prev_df['population'].values).transpose()
# multiply number of people with HIV/AIDS by coverage to get number of people
# covered
df = misc.draw_math([df, prev_df],
['location_id', 'year_id', 'age_group_id', 'sex_id'],
specs.DRAW_COLS, '*')
# get number of people covered and number of people with HIV/AIDS for each
# year and loc (sum up by sex and by age)
df = df.groupby(specs.ID_COLS, as_index=False)[specs.DRAW_COLS].sum()
prev_df = prev_df.groupby(specs.ID_COLS,
as_index=False)[specs.DRAW_COLS].sum()
# now divide number of people covered by number of people with HIV/AIDs to get
# back into coverage space
df = misc.draw_math([df, prev_df], specs.ID_COLS, specs.DRAW_COLS, '/')
draw_parameters['to_check'] = ['location_id', 'year_id']
param_check(df, draw_parameters)
return df[specs.ID_COLS + specs.DRAW_COLS]
