def make_summaries(data):
"""
This function expects a DataFrame with the following
columns:
location_id
ihme_loc_id
year
sim
mort
"""
# Input
input_keep_cols = ['location_id', 'ihme_loc_id', 'year', 'sim', 'mort']
data = data[input_keep_cols]
# Format columns
data['year_id'] = data['year'].astype('int64')
data['sex_id'] = 3
data['age_group_id'] = 1
data['estimate_stage_id'] = 3
data['sim'] = data['sim'].astype('int64')
# Reshape draws wide
index_cols = [
'location_id', 'ihme_loc_id', 'year_id', 'year', 'sex_id',
'age_group_id', 'estimate_stage_id'
]
data = data.pivot_table(values="mort", index=index_cols, columns="sim")
data = data.reset_index()
data = data.rename(columns={x: 'draw_{}'.format(x) for x in range(1000)})
# Get the summary statistics
data = sm.get_estimates(data)
# Format for upload
keep_cols = index_cols + ['mean', 'lower', 'upper']
return data[keep_cols]
def summarize(self, arc_draws):
self.index_cols.extend(['year_start_id', 'year_end_id'])
self.index_cols.remove('year_id')
summaries_mean = arc_draws[self.index_cols + ['pct_change_means']]
summaries = get_estimates(arc_draws[self.index_cols + self.draw_cols])
summaries = summaries.merge(summaries_mean, on=self.index_cols)
summaries.rename(columns={'pct_change_means': 'val'}, inplace=True)
return summaries
def summarize_draws(df, index_cols):
"""Summarize the draws down to mean/lower/upper columns"""
col_order = [
'measure_id', 'year_id', 'location_id', 'sex_id', 'age_group_id',
'cause_id', 'rei_id', 'metric_id', 'mean', 'upper', 'lower'
]
sumdf = sm.get_estimates(df)
sumdf = sumdf.reset_index()
del sumdf['index']
del sumdf['median']
return sumdf[col_order]
def get_data_frame(self):
logger.info("BEGIN compute summaries")
self.validate_measure_and_metric(self.in_df, "incoming dataframe")
logger.debug("validated")
sumdf = sm.get_estimates(self.in_df)
sumdf = sumdf.reset_index()
del sumdf['index']
del sumdf['median']
if 'pct_change_means' in sumdf:
logger.info("replacing mean of pct change distribution with pct "
"change of means")
sumdf['mean'] = sumdf['pct_change_means']
sumdf = sumdf[self.write_out_columns]
return sumdf
def summarize(self, mmr_draws):
logger.info("Summarizing MMR draws")
summaries = get_estimates(mmr_draws)
summaries.drop('median', axis=1, inplace=True)
summaries.rename(columns={'mean': 'val'}, inplace=True)
return summaries
# standardize all inputs by transforming everything to rate space
df = define_metric(df, source)
if 1 in df.metric_id.unique():
df.loc[df.metric_id == 1] = transform_metric(df.loc[df.metric_id == 1],
to_id=3,
from_id=1)
# find index (non draw) columns
try:
df.drop(['envelope', 'pop'], axis=1, inplace=True)
except:
pass
draw_cols = list(df.filter(like='draw').columns)
index_cols = list(set(df.columns) - set(draw_cols + ['year_id']))
# calculate pct_change
if change_type == 'pct_change_num': # drop any 2's. transform only 3's.
df = transform_metric(df[df.metric_id == 3], to_id=1, from_id=3)
if change_type in ['pct_change_rate', 'pct_change_num']:
change_type = 'pct_change'
change_df = pct_change(df, start_year, end_year, change_type, index_cols)
# summarize
summ_df = get_estimates(change_df)
summ_df.drop('mean', axis=1, inplace=True)
summ_df.rename(columns={'pct_change_means': 'mean'}, inplace=True)
# stream results to sys.stdout for get_pct_change.ado to read in
# Use a dct because stata is faster at reading those
to_dct(df=summ_df, fname=sys.stdout, include_header=True)