def run_spacetime(location_id, lambdaa=1.0, omega=2, zeta=0.95):
# Results dir
try:
os.makedirs(stdir)
except:
pass
# Read in some data...
data = pd.read_csv("%s/linear_predictions.csv" % lindir)
results = []
for sex in [1, 2]:
sdata = data[data.sex_id == sex]
################################
# All country example
################################
# Initialize the smoother
s = st.Smoother(
sdata,
153,
datavar='ln_dr',
modelvar='ln_dr_predicted',
# snvar='sn_flag',
pred_age_group_ids=data.age_group_id.unique(),
pred_start_year=data.year_id.min())
# Set parameters (can additionally specify omega (age weight, positive
# real number) and zeta (space weight, between 0 and 1))
s.lambdaa = lambdaa
s.omega = omega
s.zeta = zeta
# s.sn_weight = 0.2
# Tell the smoother to calculate both time weights and age weights
s.time_weights()
s.age_weights()
# Run the smoother and write the results to a file
s.smooth(location_id)
# Using the "include_mad" will calculate the global / regional /
# national MAD estimates of the ST residuals, in case you need them
# for the GPR step... results = s.format_output(include_mad=True)
r = s.results
r['sex_id'] = sex
results.append(r)
results = pd.concat(results)
results.to_csv('%s/%s.csv' % (stdir, location_id), index=False)
return pd.read_csv(f).set_index(idx_cols)
results = [readf(f) for f in fs]
results = pd.concat(results, axis=1)
results = results.reset_index()
# Calculate MADs
forgpr = []
for sex in [1, 2]:
sdata = data[data.sex_id == sex]
sresults = results[results.sex_id == sex]
sresults = sresults.drop('sex_id', axis=1)
s = st.Smoother(sdata,
42,
datavar='ln_dr',
modelvar='ln_dr_predicted',
pred_age_group_ids=data.age_group_id.unique(),
pred_start_year=data.year_id.min())
s.results = sresults
forgpr.append(s.calculate_mad())
forgpr = pd.concat(forgpr)
"""
Convert data variances to log space
Use standard eror as the data variance. Approximate 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) = Logit(X)
def run_spacetime(location_id, df, age_start, age_end, year_start, year_end,
lambdaa, zeta, zeta_nodata, omega):
################################
## Setup
################################
## Detect level and parent
national_id = int(locs.level_3[locs.location_id == location_id])
level = int(locs.level[locs.location_id == location_id])
# Making sure that only borrowing strength from higher levels
columns_to_keep = list(df.columns.values)
df = df[((df.level <= 3) | (df.level_3 == national_id))
& (df.level <= level)]
df = df[columns_to_keep]
# Count the number of data (maximum number of data in an age group for that sex)
data_count = df.loc[df.location_id == location_id].groupby(
'age_group_id').agg('count')
data_count = np.max(data_count.data)
# If data count is less than threshold, pass a flag to ST
if data_count >= data_threshold:
zeta_threshold = 1
else:
zeta_threshold = 0
## If level > 3, set zeta to 0.5
if level > 3:
zeta = 0.5
################################
## Set weights
################################
# Initialize the smoother
s = st.Smoother(df,
location_set_version_id,
timevar='year_id',
agevar='age_group_id',
spacevar='location_id',
datavar='data',
modelvar='prior',
pred_age_group_ids=range(age_start, age_end + 1),
pred_start_year=year_start,
pred_end_year=year_end,
snvar='cv_subgeo')
# Set parameters (can additionally specify omega (age weight, positive real number) and zeta (space weight, between 0 and 1))
s.lambdaa = lambdaa
s.zeta = zeta
s.zeta_no_data = zeta_no_data
if 22 not in pd.unique(df['age_group_id']):
s.omega = omega
# Tell the smoother to calculate both time weights and age weights
s.time_weights()
if 22 not in pd.unique(df['age_group_id']):
s.age_weights()
################################
## Run Smoother
################################
s.smooth(locs=location_id, level=level, zeta_threshold=zeta_threshold)
results = pd.merge(df,
s.long_result(),
on=['age_group_id', 'year_id', 'location_id'],
how='right')
################################
## Clean
################################
cols = ['location_id', 'year_id', 'age_group_id', 'sex_id', 'st']
results = results[cols].drop_duplicates()
return results