'superregional': 1,
'regional': 2,
'national': 3
}
if amp in mad_lvl_map.keys():
amp = sdata["mad_level_%s" %
mad_lvl_map[amp]].unique()[0] * 1.4826
else:
amp = float(amp)
# Run GPR
fit = gpr.fit_gpr(sdata,
amp,
obs_variable='ln_dr',
obs_var_variable='data_var',
scale=5,
draws=0,
diff_degree=2)
"""
Convert back to normal space
Approximate back-transformed variance using the delta method:
G(X) = G(mu) + (X-mu)G'(mu) (approximately)
Var(G(X)) = Var(X)*[G'(mu)]^2 (approximately)
Examples:
For G(X) = Invlogit(X)
... Invlogit'(p) = e^x/[(e^x+1)^2]
... so Var(Logit(X)) = Var(X) * e^2x / [(e^x+1)^4]
For G(X) = exp(X)
## Set Scale
scale = int(params['gpr_scale'])
## If parallel, subset
if parallel == 1:
df = df.loc[(df.location_id >= loc_start) & (df.location_id <= loc_end)]
#################################
### Run GPR
#################################
groups = df.groupby(by=['location_id', 'sex_id', 'age_group_id'])
df = groups.apply(lambda x: gpr.fit_gpr(x,
obs_variable='data',
obs_var_variable='variance',
mean_variable='st',
amp=x['st_amp'].values[0] * 1.4826,
scale=scale,
draws=draws))
#################################
### Save
#################################
## Clean
if draws == 0:
df = df[[
'location_id', 'year_id', 'age_group_id', 'sex_id', 'gpr_mean',
'gpr_lower', 'gpr_upper'
]]
else: