def run_phase(df, nid, extract_type_id, data_type_id, source,
representative_id, code_system_id):
"""Prep source data by location, year, age, sex, and garbage level."""
# set caching
configurator = Configurator('standard')
cache_dir = configurator.get_directory('db_cache')
cache_options = {
'block_rerun': True,
'cache_dir': cache_dir,
'force_rerun': False,
'cache_results': False
}
print_log_message("Pulling map from cause to detail level")
detail_level_map = get_map_to_package_metadata(code_system_id)
# merge incoming data with cause detail level df
print("Merging detail level onto data")
df = merge_with_detail_map(df, detail_level_map)
print_log_message("Determining national coverage")
# get national reprsentativeness
# in stata we used whether or not there was national
# or subnational in the source_type
# but since we no longer have that then the best we can
# do here is just use "representative_id"
df = assign_nationally_representative(
df, source, representative_id, data_type_id, cache_options
)
print_log_message("Collapsing data.")
df = df.groupby(["location_id", "year_id", "nid", "extract_type_id",
"source", "data_type_id", "detail_level_id",
"nationally_representative", "age_group_id",
"sex_id"], as_index=False)["deaths"].sum()
return df
def run_phase(df, cause_set_version_id, location_set_version_id, data_type_id,
env_run_id, source, nid, extract_type_id, remove_decimal,
code_map_version_id):
"""Run the full pipeline, chaining together CodProcesses."""
configurator = Configurator('standard')
cache_dir = configurator.get_directory('db_cache')
cache_options = {
'block_rerun': True,
'cache_dir': cache_dir,
'force_rerun': False,
'cache_results': False
}
# get cause hierarchy
cause_meta_df = get_current_cause_hierarchy(
cause_set_version_id=cause_set_version_id, **cache_options)
# get location hierarchy
location_meta_df = get_current_location_hierarchy(
location_set_version_id=location_set_version_id, **cache_options)
# get envelope
env_meta_df = get_env(env_run_id=env_run_id, **cache_options)
# get env with HIV
env_hiv_meta_df = get_env(env_run_id=env_run_id,
with_hiv=True,
**cache_options)
# get age groups
age_meta_df = get_ages(**cache_options)
code_system_id = int(
get_value_from_nid(nid,
'code_system_id',
extract_type_id=extract_type_id))
cause_map = get_cause_map(code_map_version_id=code_map_version_id,
**cache_options)
package_map = get_package_map(code_system_id=code_system_id,
**cache_options)
disagg_df = get_phase_output("disaggregation", nid, extract_type_id)
misdc_df = get_phase_output("misdiagnosiscorrection", nid, extract_type_id)
cause_package_hierarchy = get_cause_package_hierarchy(code_system_id)
if source == "Cancer_Registry":
df = prune_cancer_registry_data(df, location_meta_df)
# aggregate location
print_log_message("Aggregating location to country level")
location_aggregator = LocationAggregator(df, location_meta_df)
df = location_aggregator.get_computed_dataframe()
if data_type_id in POLICE_SURVEY_DATA_TYPE:
# special step to remove HIV from maternal data
print_log_message("Removing HIV from cc_code for maternal data.")
maternal_hiv_remover = MaternalHIVRemover(df, env_meta_df,
env_hiv_meta_df, source, nid)
df = maternal_hiv_remover.get_computed_dataframe()
print_log_message("Calculating sample size")
df = calc_sample_size(df)
print_log_message(log_statistic(df))
print_log_message("Converting to cause fractions")
df = df.loc[df['sample_size'] > 0]
df = convert_to_cause_fractions(
df, ['deaths', 'deaths_rd', 'deaths_corr', 'deaths_raw'])
print_log_message(log_statistic(df))
if data_type_id == VA_DATA_TYPE:
# run VA anemia adjusment
print_log_message("Running VA Anemia adjustment")
va_anemia_adjuster = AnemiaAdjuster()
df = va_anemia_adjuster.get_computed_dataframe(df)
if data_type_id in POLICE_SURVEY_DATA_TYPE:
cause_list = df.cause_id.unique()
square_me = (len(cause_list) == 2) & (CC_CODE in cause_list)
if (source in MATERNAL_SQUARED) or square_me:
print_log_message("Squaring maternal data")
df = square_maternal_sources(df, cause_meta_df, age_meta_df)
print_log_message("Dropping cc code")
df = drop_cc_code(df)
print_log_message(log_statistic(df))
print_log_message("Splitting locations.")
env_loc_splitter = EnvelopeLocationSplitter(df, env_meta_df, source)
df = env_loc_splitter.get_computed_dataframe()
print_log_message(log_statistic(df))
# aggregate causes
print_log_message("Aggregating causes")
cause_aggregator = CauseAggregator(df, cause_meta_df, source)
df = cause_aggregator.get_computed_dataframe()
print_log_message(log_statistic(df))
print_log_message("Adding parnt-mapped garbage to aggregated causes")
parent_gbg_adder = ParentMappedAggregatedGarbageAdder(
nid, extract_type_id, source, cause_package_hierarchy, cause_meta_df,
package_map, cause_map, remove_decimal, disagg_df, misdc_df)
df = parent_gbg_adder.get_computed_dataframe(df)
print_log_message("Applying hiv-prevalance in pregnancy adjustment to "
"maternal deaths")
hmp = HIVMatPAFs()
df = hmp.get_computed_dataframe(df, cause_meta_df, location_meta_df)
print_log_message(log_statistic(df))
print_log_message(
"Removing HIV and shocks from cause fraction denominator")
hiv_shock_remover = SampleSizeCauseRemover(cause_meta_df)
df = hiv_shock_remover.get_computed_dataframe(df)
print_log_message(log_statistic(df))
df = conform_one_like_cf_to_one(df)
print_log_message("Verifying cause fractions not null between 0 and 1")
assert_valid_cause_fractions(df)
if dataset_has_redistribution_variance(data_type_id, source):
# Determine the redistribution variance
rdvar = RedistributionVarianceEstimator(nid, extract_type_id,
cause_meta_df, remove_decimal,
code_system_id, cause_map,
package_map)
df = rdvar.get_computed_dataframe(df)
return df
class HIVMatPAFs(CodProcess):
calc_cf_col = 'cf'
all_cf_cols = ['cf', 'cf_raw', 'cf_corr', 'cf_rd']
def __init__(self):
self.configurator = Configurator('standard')
self.cache_dir = self.configurator.get_directory('db_cache')
self.maternal_hiv_props_path = \
self.configurator.get_directory('maternal_hiv_props')
# self.need_subnational_props = [51, 16, 86, 214, 165]
def get_computed_dataframe(self, df, cause_meta_df, location_meta_df):
restricted_maternal_df = \
self.restrict_to_maternal_data(df, cause_meta_df)
if restricted_maternal_df is None:
# nothing to do if there is no maternal data to adjust
return df
appended_pafs = self.append_maternal_pafs(
restricted_maternal_df.year_id.unique())
# no longer need this step since new PAFs have been created
# extra step to fix missing sub national proportions
# appended_pafs = self.duplicate_national_props(appended_pafs, location_meta_df)
merged_data = \
self.merge_data_and_proportions(restricted_maternal_df,
appended_pafs)
percent_maternal = self.generate_percentages(merged_data)
split_maternal = self.generate_splits(percent_maternal)
hiv_cfs = self.create_maternal_hiv_cfs(split_maternal)
cleaned = self.clean_adjusted_data(hiv_cfs)
final = \
self.append_adjusted_orig(df, restricted_maternal_df, cleaned)
group_cols = [
col for col in final.columns
if col not in self.all_cf_cols and col not in ['sample_size']
]
final = final.groupby(group_cols, as_index=False).agg({
'sample_size': 'mean',
'cf': 'sum',
'cf_raw': 'sum',
'cf_corr': 'sum',
'cf_rd': 'sum'
})
return final
def restrict_to_maternal_data(self, df, cause_meta_df):
"""Restrict incoming dataframe to only maternal data."""
df = df.copy()
# get age start and age end for maternal ages
maternal_metadata = cause_meta_df.loc[cause_meta_df['cause_id'] == 366]
age_start = maternal_metadata['yll_age_start']
assert len(age_start) == 1
age_start = age_start.iloc[0]
age_end = maternal_metadata.yll_age_end
assert len(age_end) == 1
age_end = age_end.iloc[0]
data = add_age_metadata(df,
add_cols=['simple_age'],
merge_col='age_group_id',
force_rerun=False,
block_rerun=True,
cache_results=False,
cache_dir=self.cache_dir)
data.rename(columns={'simple_age': 'age'}, inplace=True)
maternal_data = data.loc[(df['cause_id'] == 366)
& (data['age'] >= age_start) &
(data['age'] <= age_end) &
(data['sex_id'] == 2) &
(data['year_id'] >= 1980)]
maternal_data.drop('age', axis=1, inplace=True)
if len(maternal_data) == 0:
return None
else:
return maternal_data
def append_maternal_pafs(self, years):
"""Read in proportions."""
props = pd.DataFrame()
for year in years:
year = int(year)
props_path = "{}/maternal_hiv_props_{}.csv".format(
self.maternal_hiv_props_path, year)
data = pd.read_csv(props_path)
props = props.append(data)
props = props.rename(columns={'year': 'year_id'})
return props
def duplicate_national_props(self, props_df, loc_df):
"""Duplicate national proportions and fill sub national proportions.
Note: necessary in countries that we are now modeling sub nationally,
but since we weren't before there aren't any sub national proportions
for maternal hiv (yet).
"""
subnational = loc_df.loc[
loc_df['level'] > 3,
['location_id', 'parent_id', 'level', 'path_to_top_parent']]
# Russia sub nationals are level 5 while other countries are level 4
subnational.loc[
subnational['level'] == 5, 'parent_id'] = \
subnational['path_to_top_parent'].str.split(',').str[3].astype(int)
# only keep rows with the needed sub national locations
subnational = subnational.loc[subnational['parent_id'].isin(
self.need_subnational_props)]
# drop level 4 sub national location_ids for Russia
subnational = subnational.loc[~((subnational['parent_id'] == 62) &
(subnational['level'] == 4))]
subnational = subnational[['location_id', 'parent_id']]
subnational.rename(columns={
'location_id': 'child_location_id',
'parent_id': 'location_id'
},
inplace=True)
# create sub national maternal_hiv proportions from national
subnational = props_df.merge(subnational, on='location_id')
subnational.drop('location_id', axis=1, inplace=True)
subnational.rename(columns={'child_location_id': 'location_id'},
inplace=True)
props_df = pd.concat([props_df, subnational])
assert not props_df.duplicated().any(), 'please check maternal'\
' proportions, there are duplicates'
return props_df
def merge_data_and_proportions(self, data, props):
"""Merge restricted maternal data and proportions."""
merged_data = data.merge(props,
on=['location_id', 'age_group_id', 'year_id'],
how='left')
assert merged_data.notnull().values.all(), 'maternal proportions '\
'were not successfully merged with incoming data'
return merged_data
def generate_percentages(self, df):
"""Create new 'pct_maternal column'.
This is to prepare for calculating maternal hiv cause fractions
"""
df['pct_maternal'] = 1 - df['pct_hiv'] - df['pct_maternal_hiv']
df.loc[df['pct_maternal'].isnull(), 'pct_maternal'] = 1
df.loc[df['pct_hiv'].isnull(), 'pct_hiv'] = 0
df.loc[df['pct_maternal_hiv'].isnull(), 'pct_maternal_hiv'] = 0
assert all(x > 0 for x in df['pct_maternal'])
assert df[['pct_maternal', 'pct_hiv', 'pct_maternal_hiv'
]].notnull().values.any(), 'there are missing percentages'
assert all(
abs(df['pct_maternal'] + df['pct_hiv'] + df['pct_maternal_hiv']) -
1) < .0001
# proportion of maternal that is aggravated by hiv
# cannot be above 13% based on USERNAME's meta-analysis; otherwise
# this would suggest the percentage of maternal deaths that were
# hiv positive is >1
assert (df['pct_maternal_hiv_vr'] <= .13).all()
# maternal_hiv should not yet exist
assert not (df['cause_id'] == 741).any()
return df
''' '''
def generate_splits(self, df):
"""Create a column to indicate how the data should be split.
(depends on source type)
"""
df = add_nid_metadata(
df,
add_cols='data_type_id',
block_rerun=True,
cache_dir=self.cache_dir,
force_rerun=False,
)
df.loc[df['data_type_id'].isin([7, 5]), 'split_maternal'] = 1
df.loc[df['split_maternal'].isnull(), 'split_maternal'] = 0
df.loc[df['split_maternal'] == 0, 'pct_maternal'] = 1
df.loc[df['split_maternal'] == 0,
'pct_maternal_hiv'] = df['pct_maternal_hiv_vr']
df.loc[df['split_maternal'] == 0, 'pct_hiv'] = 0
df.drop('pct_maternal_hiv_vr', axis=1, inplace=True)
return df
def create_maternal_hiv_cfs(self, df):
"""Create cause fractions for maternal hiv."""
df = df.copy()
maternal_hiv_df = df.copy()
maternal_hiv_df['cf'] = maternal_hiv_df['cf'] * \
maternal_hiv_df['pct_maternal_hiv']
maternal_hiv_df['cause_id'] = 741
maternal_hiv_df['cf_raw'] = 0
maternal_hiv_df['cf_corr'] = 0
maternal_hiv_df['cf_rd'] = 0
maternal_df = df.copy()
maternal_df['cf'] = maternal_df['cf'] * maternal_df['pct_maternal']
maternal_df['cause_id'] = 366
df = pd.concat([maternal_hiv_df, maternal_df], ignore_index=True)
return df
def clean_adjusted_data(self, df):
"""Clean up adjusted data to add on to the original dataset.
Add maternal_hiv to maternal, keep the maternal_hiv,
split_maternal 0 observations and call them maternal
"""
va_vr = df.loc[df['split_maternal'] == 0]
if len(va_vr) > 0:
assert set([741, 366]) == set(va_vr.cause_id.unique())
va_vr = va_vr.loc[va_vr['cause_id'] != 366]
va_vr['cause_id'] = 366
df = pd.concat([df, va_vr], ignore_index=True)
df = df.groupby([
'nid', 'extract_type_id', 'location_id', 'year_id', 'site_id',
'age_group_id', 'sex_id', 'sample_size', 'cause_id'
],
as_index=False)[self.all_cf_cols].sum()
# it is possible that, using this method, cause fractions exceed 1.
# this is meaningless and breaks noise reduction, so cap it
# make sure that cf isn't something absurd, though
assert (df['cf'] < 1.1).all()
df.loc[df['cf'] > 1, 'cf'] = 1
return df
def append_adjusted_orig(self, orig, maternal_data, adjusted):
"""Remove original maternal data and append on adjusted."""
data = orig.merge(maternal_data, how='left', indicator=True)
data = data.loc[data['_merge'] != 'both']
data.drop('_merge', axis=1, inplace=True)
data = data.append(adjusted, ignore_index=True)
return data
class GBDCauseMapper(CodProcess):
"""Convert cause codes into cause_ids.
Arguments:
id_cols (list):
data_col (list):
unique_cols (list):
Returns:
df, a pandas DataFrame with addition of cause_id
diag_df, a pandas DataFrame: assesses the difference
between different mapping versions
"""
id_cols = ['nid', 'extract_type_id', 'location_id', 'year_id',
'age_group_id', 'sex_id', 'cause_id', 'code_id',
'site_id']
data_col = ['deaths']
unique_cols = ['nid', 'extract_type_id', 'location_id', 'year_id',
'age_group_id', 'sex_id',
'cause_id', 'code_id', 'site_id']
# These are acauses 'sub_total', and '_sb'
unnecessary_causes = [920, 744]
cache_dir = str()
def __init__(self, cause_set_version_id, code_map):
self.cg = Configurator("standard")
self.cache_dir = self.cg.get_directory('db_cache')
self.cause_set_version_id = cause_set_version_id
self.code_map = code_map
def get_computed_dataframe(self, df, code_system_id):
# make special cause adjustments
df = self.special_cause_reassignment(df, code_system_id)
"""Map code id to cause id."""
print_log_message("Merging with cause map")
# get code metadata from a file already cached
df = add_code_metadata(
df, ['cause_id'], code_system_id,
code_map=self.code_map
)
report_if_merge_fail(df, 'cause_id', 'code_id')
# Make sure the mappings are good!
print("Asserting it's all good")
self.assert_valid_mappings(df, code_system_id)
df = self.drop_unnecessary_causes(df, self.unnecessary_causes)
print("Collapsing")
df = self.collapse_and_sum_by_deaths(df)
return df
def drop_unnecessary_causes(self, df, unnecessary_causes):
# Drops causes set as unnecessary, subtotal and stillbirth
df = df.copy()
df = df[~df['cause_id'].isin(unnecessary_causes)]
return df
def special_cause_reassignment(self, df, code_system_id):
"""Replace the actual data cause under certain conditions.
There are instances where a PI has good reason to
believe that a certain group of deaths were assigned
to the wrong cause, and it is known what cause to re-assign
those deaths to. Implement here.
This essentially allows mapping based on not just the cause
and code system but based on other information like
the location, NID, year, etc.
It can also be used (sparingly) for hotfixes like
changing all codes with values 'acause_digest_gastrititis'
to be named 'acause_digest_gastritis'.
Args:
df (DataFrame): data with cause
Returns:
DataFrame: with any modifications
"""
cache_args = {
'force_rerun': False,
'block_rerun': True,
'cache_dir': 'standard',
'cache_results': False
}
# Some SRS codes get redistributed differently than
# other ICD10 datasets
df = add_nid_metadata(
df, 'source', **cache_args
)
if (df['source'] == "India_SRS_states_report").any():
print_log_message("Changing SRS codes to custom garbage groups")
assert (df['source'] == "India_SRS_states_report").all()
df = add_code_metadata(
df, 'value', code_system_id=code_system_id,
**cache_args
)
custom_grbg = pd.read_csv(
self.cg.get_resource("srs_custom_garbage_groups")
)
custom_grbg = custom_grbg.query('active == 1')
custom_grbg['value'] = custom_grbg['srs_custom_garbage_group']
custom_grbg = add_code_metadata(
custom_grbg, 'code_id', code_system_id=code_system_id,
merge_col='value', **cache_args
)
custom_grbg = custom_grbg.rename(
columns={'code_id': 'new_code_id'})
custom_grbg = custom_grbg[['package_id', 'new_code_id']]
gp_dfs = []
for package_id in custom_grbg.package_id.unique():
# THIS QUERIES THE DATABASE - BUT THERE SHOULD NEVER BE A TON
# OF SRS JOBS HAPPENING AT ONCE SO IT SHOULD BE OK
gp_df = get_garbage_from_package(
code_system_id, package_id, package_arg_type="package_id"
)
assert len(gp_df) != 0, \
"Found 0 codes for package {}".format(package_id)
gp_dfs.append(gp_df)
gp_df = pd.concat(gp_dfs, ignore_index=True)
gp_df = gp_df.merge(custom_grbg, how='left')
report_if_merge_fail(gp_df, 'new_code_id', 'package_id')
gp_df = gp_df[['value', 'new_code_id']]
gp_df['value'] = gp_df['value'].str.strip()
df = df.merge(gp_df, how='left', on='value')
df.loc[df['new_code_id'].notnull(), 'code_id'] = df['new_code_id']
df['code_id'] = df['code_id'].astype(int)
df = df.drop(['new_code_id', 'value'], axis=1)
df = df.drop('source', axis=1)
china_cdc_2008 = (df['nid'] == 270005) & (df['extract_type_id'] == 2)
# J96.00 - move five to four digit J96.0 (this should be a rule in formatting, only keep 4 digit detail)
five_dig_code = df['code_id'] == 13243
df.loc[
china_cdc_2008 & five_dig_code,
'code_id'
] = 13242
return df
def collapse_and_sum_by_deaths(self, df):
"""Group by final columns, summing across deaths.
Directly modifies the dataframe, keeping only the columns needed
to move on to the next Claude step. Also includes an assertion
that there are no duplicates.
"""
df = df.groupby(self.id_cols, as_index=False)[self.data_col].sum()
self.assert_unique_cols_unique(df)
return df
def assert_valid_mappings(self, df, code_system_id):
"""Test that the mapping worked.
Runs a suite of assertions to make sure that mapping was successful.
Args:
df (DataFrame): with at least code_id and cause_id
Returns:
None
Raises:
AssertionError: Any condition fails
"""
# add code value from cached code map
print("Adding value")
df = add_code_metadata(
df, ['value'], code_system_id,
force_rerun=False,
block_rerun=True,
cache_dir=self.cache_dir
)
report_if_merge_fail(df, 'value', 'code_id')
# get acause from cached cause hierarchy
print("Adding acause")
df = add_cause_metadata(
df, ['acause'],
cause_set_version_id=self.cause_set_version_id,
force_rerun=False,
block_rerun=True,
cache_dir=self.cache_dir
)
report_if_merge_fail(df, 'acause', 'cause_id')
# Test that all causes starting with 'acause_' are mapped correctly.
# acause_cvd, for example, should be mapped to 'cvd' (not 'cvd_ihd').
# 'acause__gc_X59' should be mapped to '_gc', etc.
print("Checking implied acauses")
check_df = df.loc[df['value'].str.startswith('acause_')]
check_df['implied_acause'] = \
check_df['value'].str.replace('acause_', '', 1)
check_df.loc[
check_df['value'].str.contains("acause__gc"),
'implied_acause'
] = "_gc"
bad_df = check_df.loc[
check_df['acause'] != check_df['implied_acause']
]
if len(bad_df) > 0:
bad_stuff = bad_df[['value', 'acause']].drop_duplicates()
raise AssertionError(
"These code values do not match their acause: "
"\n{}".format(bad_stuff)
)
print("Checking for bad values")
# assert incorrect acauses are gone
bad_acauses = ['acause_digest_gastrititis',
'acause_hiv_tb',
'acause_tb_drug']
bad_df = df.loc[df['value'].isin(bad_acauses)].value.unique()
if len(bad_df) > 0:
raise AssertionError(
"Found these bad code values in the data: {}".format(bad_stuff)
)
def assert_unique_cols_unique(self, df):
"""Test that columns that should uniquely identify the dataframe do."""
assert not df.duplicated(self.unique_cols).any()
class BridgeMapper(CodProcess):
"""Replace acauses with those in the bridge map.
Arguments:
source (str)
cause_set_version_id (int)
code_system (str)
Returns:
df, pandas DataFrame: only change is replacing some cause_ids
diag_df, pandas DataFrame: shows which cause_ids have been changed
"""
id_cols = ['nid', 'extract_type_id', 'location_id', 'year_id',
'age_group_id', 'sex_id', 'cause_id',
'site_id']
val_cols = ['deaths', 'deaths_rd', 'deaths_corr', 'deaths_raw']
# data type id for verbal autopsy
VA = 8
def __init__(self, source, cause_meta_df, code_system):
self.source = source
self.code_system = code_system
self.conf = Configurator("standard")
self.bridge_map_path = Path(self.conf.get_directory('bridge_maps'))
self.cause_meta_df = cause_meta_df
self.cache_options = {
'force_rerun': False,
'block_rerun': True,
'cache_results': False,
'cache_dir': 'standard'
}
def get_computed_dataframe(self, df):
"""Replace acauses with those in the bridge map."""
# VA sources are the only ones where this may not work
# might need to split dataframe by data_type_id for bridge map
df = add_nid_metadata(df, ['data_type_id'], **self.cache_options)
has_verbal_autopsy = self.VA in df['data_type_id'].unique()
df.drop(columns='data_type_id', inplace=True)
if self.needs_bridging(has_verbal_autopsy):
file_name = self.get_file_name(has_verbal_autopsy)
map_df = pd.read_csv(self.bridge_map_path / file_name)
map_df = map_df[['acause', 'bridge_code']]
# add acause column to deaths data
bridge_mapped = add_cause_metadata(
df,
['acause'],
merge_col='cause_id',
cause_meta_df=self.cause_meta_df
)
# hack, this cause_id snuck in somehow...
bridge_mapped.loc[
bridge_mapped['cause_id'] == 606, 'acause'
] = 'gyne_femaleinfert'
report_if_merge_fail(bridge_mapped, 'acause', 'cause_id')
bridge_mapped.drop(['cause_id'], axis=1, inplace=True)
# perform zz bridge code redistribution before other bridge mapping
bridge_mapped = self.redistribute_zz_bridge_codes(bridge_mapped, map_df)
bridge_mapped = bridge_mapped.merge(
map_df, how='left', on='acause'
)
bridge_mapped = self.acause_to_bridge_code(bridge_mapped)
# bring cause_id back
bridge_mapped = add_cause_metadata(
bridge_mapped,
['cause_id'],
merge_col='acause',
cause_meta_df=self.cause_meta_df
)
# hack, this cause_id snuck in
bridge_mapped.loc[
bridge_mapped['acause'] == 'gyne_femaleinfert', 'cause_id'
] = 606
report_if_merge_fail(bridge_mapped, 'cause_id', 'acause')
# output diagnostic dataframe
self.diag_df = bridge_mapped
# drop unnecessary columns
bridge_mapped = self.clean_up(bridge_mapped)
return bridge_mapped
else:
self.diag_df = df
df = self.clean_up(df)
return df
def needs_bridging(self, has_verbal_autopsy):
"""
Check data type and code_system to see if the bridge map is needed.
"""
code_systems_to_bridge_map = [
"ICD9_detail", "ICD9_BTL", "ICD10_tabulated",
"ICD8_detail", "ICD8A",
"China_1991_2002", "India_SCD_states_rural", "India_MCCD_states_ICD10",
"India_MCCD_states_ICD9", "India_SRS_states_report",
"Russia_FMD_1989_1998", "ICD9_USSR_Tabulation", "INDEPTH_ICD10_VA",
"India_Maharashtra_SCD", "India_CRS", "PHL_VSR_1999_2005"
]
special_sources_to_bridge_map = [
"Russia_FMD_ICD9",
"India_SRS_states_report", "India_MCCD_Orissa_ICD10"
]
# not all VA sources use a bridge map... something to think about
# in the future, but not necessary right now
if has_verbal_autopsy | \
(self.code_system in code_systems_to_bridge_map) | \
(self.source in special_sources_to_bridge_map):
# we need to use the bridge map!
return True
else:
# we do not need to use the bridge map
return False
def get_file_name(self, has_verbal_autopsy):
"""Determine the file name needed based on the source or code system.
Note: The default file name will be the name of the code system,
with some exceptions. For some sources we have specified specific
files to bridge map with, all other sources will use the file
that matches its code_system.
"""
source_to_sheet = {
"India_MCCD_Orissa_ICD10": "India_MCCD_states_ICD10",
"India_MCCD_Delhi_ICD10": "India_MCCD_states_ICD10",
"Thailand_Public_Health_Statistics": "ICD10_tabulated",
"India_SRS_states_report": "India_SRS_states_report",
"UKR_databank_ICD10_tab": "ICD10_tabulated",
"Russia_FMD_ICD9": "Russia_FMD_1989_1998",
}
if has_verbal_autopsy and (self.source != 'India_SRS_states_report'):
file_name = 'INDEPTH_ICD10_VA'
else:
file_name = source_to_sheet.get(self.source, self.code_system)
return file_name + '.csv'
def redistribute_zz_bridge_codes(self, df, map_df):
"""
A mini-redistribution, but only redistributes causes bridge mapped to zz codes
"""
grouping_cols = list(set(self.id_cols) - {'cause_id'})
start_deaths = {col: df.groupby(grouping_cols)[col].sum() for col in self.val_cols}
zz_code_idxs = map_df['bridge_code'].str.startswith('ZZ-')
# get the order to do the zz code redistribution in:
# start on lowest level of hierarchy and work our way up
zz_code_targets = (map_df
.loc[zz_code_idxs, ['bridge_code']]
.drop_duplicates()
.assign(acause=lambda d: d['bridge_code'].str.replace('ZZ-', '_'))
.merge(self.cause_meta_df, on='acause')
.sort_values(['level', 'acause'], ascending=False)
.loc[:, 'bridge_code']
.tolist()
)
# don't distribute onto anything that maps to a zz code
all_causes_to_zz_codes = set(map_df.loc[zz_code_idxs, 'acause'])
for zz_code in zz_code_targets:
child_cause_ids = get_all_related_causes(zz_code.strip().replace('ZZ-', '_'),
self.cause_meta_df)
child_causes = self.cause_meta_df.loc[
self.cause_meta_df['cause_id'].isin(child_cause_ids),
'acause'].tolist()
acauses_to_redistribute = map_df.loc[map_df['bridge_code'] == zz_code, 'acause']
to_redistribute = df['acause'].isin(acauses_to_redistribute)
valid_child_causes = set(child_causes) - all_causes_to_zz_codes
print_log_message('Found ZZ code: {}, deaths: {}'
.format(zz_code, df.loc[to_redistribute, 'deaths'].sum()))
# distribute onto at least all combinations of these
# this is to ensure everything in df[to_redistribute]
# get weights
values_to_include = {
'acause': valid_child_causes,
}
for col in grouping_cols:
values_to_include[col] = df.loc[to_redistribute, col].unique()
distributed = distribute(df[to_redistribute],
based_on=df[df['acause'].isin(valid_child_causes)],
distribute_over='acause',
within=grouping_cols,
value_col='deaths',
values_to_include=values_to_include,
base_value=0.001, # this is mostly arbitrary
)
report_if_merge_fail(distributed, check_col='acause', merge_cols=grouping_cols)
# what follows is an unfortunate side effect of having multiple value columns
# in the data -- it makes the merging somewhat more involved than simply
# appending distributed data to existing data
# TODO: refactor this into a generic method in redistribution_utils
df = df.merge(distributed[grouping_cols + ['acause', 'deaths']],
how='outer',
on=grouping_cols + ['acause'],
suffixes=('', '_new'),
)
# default to 0 deaths in all values where new variables / IDs (i.e. new causes)
# are in the distributed data (right only)
# and where distributed does not have data (i.e. other causes in original
# data that weren't distributed onto) (left only)
df[self.val_cols + ['deaths_new']] = df[self.val_cols + ['deaths_new']].fillna(0)
# Set values that were distributed away from their cause to 0.
# This has the effect of moving deaths away from one cause to another.
df.loc[df['acause'].isin(acauses_to_redistribute), 'deaths'] = 0
# now add distributed data to old
df['deaths'] += df['deaths_new']
df.drop(columns='deaths_new', inplace=True)
# make sure deaths didn't move out of a nid-etid-site-location-year-sex-age group
for col in self.val_cols:
end_deaths = df.groupby(grouping_cols)[col].sum()
assert np.allclose(start_deaths[col], end_deaths), \
"Dropped/added deaths during ZZ code redistribution: " + \
"start {}: {}, end {}: {}".format(col, start_deaths[col], col, end_deaths)
return df
def acause_to_bridge_code(self, df):
"""Replace the acause with the bridge code."""
# there might still be zz codes in the data because we aren't
# performing zz code redistribution on the other value columns,
# so if something is coded to i.e. _neo in the raw data, then
# we keep it as _neo.
df['acause'].update(df['bridge_code'].str.replace('ZZ-', '_'))
return df
def get_diagnostic_dataframe(self):
"""Return a diagnostic dataframe.
Diagnostic dataframe shows all changes made due to bridge mapping.
Maybe change this later to there is some sort of output.
"""
if self.diag_df is None:
print("No run of get computed dataframe yet")
else:
return self.diag_df
def clean_up(self, df):
"""Group rogue duplicates."""
df = df.groupby(self.id_cols, as_index=False)[self.val_cols].sum()
return df
inplace=True)
code_system_type_dict = {1: "ICD10", 6: "ICD9"}
df["code_system"] = code_system_type_dict[code_system_id]
df["garbage_level"] = ""
df["package_description"] = df["package_description"].str.lower()
df = df[list(mcod_map)]
return df
if __name__ == '__main__':
mcod_map = pd.read_excel("{}/mcause_map.xlsx".format(
CONF.get_directory("process_inputs")))
# archive the current map
mcod_map.to_excel(
f"{CONF.get_directory('process_inputs')}/_archive/mcause_map_2020_07_04.xlsx",
index=False)
# ger rid of current inj mapping
mcod_map = mcod_map.loc[~mcod_map["package_description"].str.contains(
"y34|x59|ncode|nn", flags=re.IGNORECASE, regex=True)]
dfs = pd.DataFrame()
for code_system_id in [1, 6]:
df = prep_all_inj_codes(code_system_id, mcod_map)
dfs = dfs.append(df, ignore_index=True)
class Recoder(CodProcess):
"""Move deaths from one thing to another based on expert opinon."""
id_cols = [
'nid', 'extract_type_id', 'location_id', 'year_id', 'age_group_id',
'sex_id', 'cause_id', 'site_id'
]
val_cols = ['deaths', 'deaths_rd', 'deaths_corr', 'deaths_raw']
def __init__(self, cause_meta_df, source, code_system_id, data_type_id):
self.source = source
self.code_system_id = code_system_id
self.data_type_id = data_type_id
self.cause_meta_df = cause_meta_df
self.conf = Configurator("standard")
self.vr_indicators_path = self.conf.get_resource('vr_indicators')
self.cache_options = {
'force_rerun': False,
'block_rerun': True,
'cache_results': False,
'cache_dir': self.conf.get_directory('db_cache')
}
def get_computed_dataframe(self, df):
"""Return computations."""
# this method is de-activated until we establish how data drops
# will be executed (new preference is through not uploading them or
# running them through noise reduction)
# df = self.drop_low_quality_data(self.df)
if 'data_type_id' not in df.columns:
df = add_nid_metadata(df, "data_type_id", **self.cache_options)
df = self.recode(df)
df = self.conform_secret_causes(df)
df = self.clean_up(df)
return df
def get_diagnostic_dataframe(self):
"""Return diagnostics."""
pass
def recode_sids(self, df):
# SIDS in under 4 star locations needs to be recoded to neonatal 02/26/18
path_to_4_stars_sheet = self.conf.get_resource("four_star_locations")
four_five_star_locs = pd.read_csv(path_to_4_stars_sheet)
four_five_star_locs = four_five_star_locs[['location_id']]
four_five_star_locs = four_five_star_locs.location_id.unique()
less_than_four_star = ~df['location_id'].isin(four_five_star_locs)
is_sids = df['cause_id'] == 686
df.loc[is_sids & less_than_four_star, 'cause_id'] = 380
return df
def clean_up(self, df):
"""Group rogue duplicates."""
df = df.groupby(self.id_cols, as_index=False)[self.val_cols].sum()
return df
def conform_secret_causes(self, df):
"""Remove secret causes and conform to reporting cause hierarchy."""
# replace parent_id = 723 if cause is "inj_suicide_pesti",
# "inj_suicide_fire", "inj_suicide_hang")
df = add_cause_metadata(df,
add_cols=['secret_cause', 'parent_id'],
cause_meta_df=self.cause_meta_df,
**self.cache_options)
injuries_replace_parents = [722, 720, 719]
replaced_injuries = df['cause_id'].isin(injuries_replace_parents)
df.loc[replaced_injuries, 'parent_id'] = 723
secret_causes = df['secret_cause'] == 1
not_cc_code = df['cause_id'] != 919
len_before = len(df)
if df['parent_id'].isnull().values.any():
raise AssertionError('There are missing parent cause_ids')
df.loc[secret_causes & not_cc_code, 'cause_id'] = df['parent_id']
len_after = len(df)
if len_before != len_after:
raise AssertionError(
'The length of the dataframe has changed from {} to {}'.format(
len_before, len_after))
df.drop(['parent_id', 'secret_cause'], axis=1, inplace=True)
return df
def drop_leukemia_subtypes(self, df):
"""Remove leukemia subtypes deaths created by redistribution.
Deaths that are created in redistribution for leukemia subtypes should
be recoded to the parent leukemia.
"""
leuk_subtypes = get_all_related_causes('neo_leukemia',
self.cause_meta_df)
# remove parent leukemia cause_id
leuk_subtypes.remove(487)
df.loc[(df['cause_id'].isin(leuk_subtypes)) & (df['deaths_rd'] > 0) &
(df['deaths_raw'] <= 0), 'cause_id'] = 487
return df
def recode(self, df):
"""Recode based on expert judgement.
"""
cause_metadata_df = self.cause_meta_df
cause_metadata_df = cause_metadata_df[[
"cause_id", "path_to_top_parent", "acause"
]]
# recode ckd except for ckd_other to cong_other in neonates
ckd_cause_ids = get_all_related_causes('ckd', cause_metadata_df)
ckd_cause_ids.remove(593)
ckd_less_other = df['cause_id'].isin(ckd_cause_ids)
neonate = df['age_group_id'].isin([2, 3])
df.loc[ckd_less_other & neonate, 'cause_id'] = 652
# recode resp_copd, resp_asthma, resp_other, resp_interstitial to lri
# in neonates
resp_ids = [509, 515, 516, 520]
is_cert_resp_causes = df['cause_id'].isin(resp_ids)
# neonate already defined
df.loc[is_cert_resp_causes & neonate, 'cause_id'] = 322
# recode resp_asthma to lri in perinates
is_asthma = df['cause_id'] == 515
df.loc[is_asthma & (df['age_group_id'] == 4), 'cause_id'] = 322
# Drop any maternal cause below age 10 and above age 55
# (recode to cc_code)
maternal_cause_ids = get_all_related_causes(366, cause_metadata_df)
maternal_cause_ids = df['cause_id'].isin(maternal_cause_ids)
# ages not in the maternal age range
non_maternal_ages = np.logical_not(df['age_group_id'].isin(
[7, 8, 9, 10, 11, 12, 13, 14, 15, 22]))
df.loc[maternal_cause_ids & non_maternal_ages, 'cause_id'] = 919
# Drop alzheimers below age 40 to (recode to cc_code)
# dementia cause_id = 543
alzheimers = df['cause_id'] == 543
under_40 = df['age_group_id'].isin(range(1, 13, 1))
df.loc[alzheimers & under_40, 'cause_id'] = 919
# Recode congenital causes to cc_code in ages over 70
# (stata: substr(acause, 1, 4) == "cong")
cong_causes = get_all_related_causes('cong', cause_metadata_df)
congenital = df['cause_id'].isin(cong_causes)
over_70 = df['age_group_id'].isin([19, 20, 30, 31, 32, 235])
df.loc[congenital & over_70, "cause_id"] = 919
# Recode neonatal-aged hepatitis
# (and all sub-causes) to neonatal_hemolytic
# except ICD9_USSR_Tabulated and ICD10_tabulated
# Recode neonatal-aged hepatitis (and all sub-causes) to neonatal
# if source is ICD9_USSR_Tabulated or ICD10_tabulated
hepatitis = get_all_related_causes(400, cause_metadata_df)
hepatitis = df['cause_id'].isin(hepatitis)
if self.code_system_id in [7, 9]:
df.loc[hepatitis & neonate, "cause_id"] = 380
else:
df.loc[hepatitis & neonate, "cause_id"] = 384
# inj_disaster_light to inj_othunintent 2/07/18
inj_disaster_light = df['cause_id'] == 984
df.loc[inj_disaster_light, 'cause_id'] = 716
# ckd diabetes type to ckd all but icd10 2/07/18
# added ICD9_detail to exception 5/15/18
if self.code_system_id not in [1, 6]:
ckd_diabetes = df['cause_id'].isin([997, 998])
df.loc[ckd_diabetes, 'cause_id'] = 589
# Removing diabetes remap 7/2/2019 - want to use the results of the new
# unspecified diabetes regression for everything
# # diabetes subtypes to parent all but icd10 2/07/18
# # added ICD9_detail, ICD10_tab to exception 5/15/18
# if self.code_system_id not in [1, 6, 9]:
# diabetes_subtypes = df['cause_id'].isin([975, 976])
# df.loc[diabetes_subtypes, 'cause_id'] = 587
# diabetes to type 1 under 15 everywhere 2/07/18
diabetes_type_2 = df['cause_id'] == 976
under_15 = df['age_group_id'] < 8
df.loc[diabetes_type_2 & under_15, 'cause_id'] = 975
# nutrition iron and iodine to zz every data 2/07/18
iron_or_iodine = df['cause_id'].isin([388, 390])
df.loc[iron_or_iodine, 'cause_id'] = 919
# cvd_ihd move to cong_heart in under one year 2/07/18
under_1 = df['age_group_id'] < 5
cvd_ihd = df['cause_id'] == 493
df.loc[cvd_ihd & under_1, 'cause_id'] = 643
if 686 in df.cause_id.unique():
df = self.recode_sids(df)
# Need to map _neo, _mental, _infect
# etc to cc code 2/07/18
df.loc[df.cause_id.isin([344, 409, 410, 542, 558, 669, 680, 961]),
'cause_id'] = 919
# usually we also have to map _inj to cc_code, but in some VA we have
# other sources for splitting _inj we do not move to cc_code 3/26/2018
if self.data_type_id not in [6, 7, 8]:
df.loc[df['cause_id'] == 687, 'cause_id'] = 919
# cvd_ihd to cvd_other in under age one to 14 years 2/07/18 bridge map
one_to_14 = df['age_group_id'].isin([5, 6, 7])
cvd_ihd = df['cause_id'] == 493
df.loc[cvd_ihd & one_to_14, 'cause_id'] = 507
# TODO test if the distinction between this and the above is necessary,
# e.g. would the bridge map already map neonatal_hemolytic to neonatal?
# Do shared cancer recodes (previously in cancer_recodes.do)
cancer_recodes = get_all_related_causes([
411, 414, 423, 426, 429, 432, 435, 438, 441, 444, 450, 453, 456,
459, 462, 465, 468, 474, 486, 483
], cause_metadata_df)
cancer_recodes = df['cause_id'].isin(cancer_recodes)
cancer_ages = df['age_group_id'].isin(range(2, 8, 1))
df.loc[cancer_recodes & cancer_ages, "cause_id"] = 489
not_icd10 = self.code_system_id != 1
neo_meso = df['cause_id'] == 483
df.loc[neo_meso & not_icd10, "cause_id"] = 489
# Recode digest_hernia to cc_code if source is Ethiopia_AAMSP
# added Ethiopia_subnational_AAMSP in GBD2017
if self.source.endswith("AAMSP"):
digest_hernia = df['cause_id'].isin([531])
df.loc[digest_hernia, "cause_id"] = 919
# in these years we split a garbage of homicide/suicide to
# their causes proportionally, now we want to recode the years
# that we don't want to use in the homicide/suicide model.
if self.source == "Iran_Mohsen_special_ICD10":
homicide_and_suicide = df['cause_id'].isin(
[724, 725, 726, 727, 941, 718, 719, 720, 721, 722, 723])
bad_years = df['year_id'].isin(range(2007, 2015))
# _unintent
df.loc[bad_years & homicide_and_suicide, "cause_id"] = 919
# Recode war subcauses to inj_homicide in Jamaica 2005 VR
inj_war = get_all_related_causes(945, cause_metadata_df)
is_inj_war = df['cause_id'].isin(inj_war)
jamaica = df['location_id'] == 115
year_2005 = df['year_id'] == 2005
vr = df['data_type_id'] == 9
df.loc[is_inj_war & jamaica & year_2005 & vr, 'cause_id'] = 724
# Recode inj_mech_gun to inj_homicide for Jamaica 2006 VR
# "In ICD10 2005 there a large number of deaths due to
# homicides, but in 2006 many of these deaths have moved to
# unintentional firearms.
# 2006 is missing homicides deaths. USERNAME wants to move deaths from
# unintentional firearms to homicides."
inj_mech_gun = df['cause_id'] == 705
year_2006 = df['year_id'] == 2006
df.loc[inj_mech_gun & year_2006 & jamaica & vr, 'cause_id'] = 724
# Recode digest_ibd to digest for Suriname 2005-2012 ICD10
# "Because NR has a very bad effect on IBD in Surinam please recode all
# of data from 1995-2012 (ICD10 ) for "digest_ibd" to "digest" in
# Suriname and keep them in recoding list for every upload"
# TODO should this be more years than just 2012? like all of ICD10?
if self.source == "ICD10":
digest_ibd = df['cause_id'] == 532
suriname = df['location_id'] == 118
year_1995_2012 = df['year_id'].isin(range(1995, 2013, 1))
df.loc[digest_ibd & suriname & year_1995_2012, 'cause_id'] = 526
# Recode endo_procedural to inj_homicide, writ-large
# "GBD2013 HACK: USERNAME and USERNAME want Endo-procedural
# to go to inj_medical just for this round.
# In GBD2014 it will go to endo"
endo_prodcedural = df['cause_id'] == 624
df.loc[endo_prodcedural, 'cause_id'] = 708
# Recode Schizophrenia to cc_code in Tibet - USERNAME's reason:
# "Because have very bad effect in Noise Reduction"
schizo = df['cause_id'] == 559
tibet = df['location_id'] == 518
df.loc[schizo & tibet, 'cause_id'] = 919
# Recode HIV and all sub-causes before 1980 to cc_code, writ-large
hiv = get_all_related_causes(298, cause_metadata_df)
hiv = df['cause_id'].isin(hiv)
pre_1980 = df['year_id'] < 1980
df.loc[hiv & pre_1980, 'cause_id'] = 919
# Recode diabetes and all sub-causes to neonatal, if age is neonatal
# "2-Any death assigned to Diabetes in neonatal period (age 0-28 days)
# in all data format (Except ICD9 and ICD10 detail) including all MCCD,
# DSP , Russia format, VA have to recode to the neonatal death" -USERNAME
# TODO this should be an age restriction for GBD not a recode
# TODO implement
diabetes_causes = get_all_related_causes(587, cause_metadata_df)
diabetes = df['cause_id'].isin(diabetes_causes)
df.loc[neonate & diabetes, 'cause_id'] = 380
# Recode cvd_stroke and all subcauses to cvd
# in Verbal Autopsy under 20 years
# "Any death in VA and SCD that assigned to the Stroke
# in under age 20 years have to recode to all CVD"
# Not done in bridge map; stata code does this for all VA
# despite SCD comment.
under_20 = df['age_group_id'].isin(range(0, 8, 1))
stroke = get_all_related_causes('cvd_stroke', cause_metadata_df)
stroke_deaths = df['cause_id'].isin(stroke)
va = df['data_type_id'] == 8
# cvd cause_id is 491
df.loc[under_20 & stroke_deaths & va, 'cause_id'] = 491
# Recode inj_trans_road_pedal to cc_code if age over 95, for everything
# USERNAME request 1/20/2017 "remove inj_trans_road_pedal for over
# 95 in all countries and years"
# TODO should this be an age restriction? questionable...
over_95 = df['age_group_id'] == 235
inj_trans_road_pedal = df['cause_id'] == 691
df.loc[over_95 & inj_trans_road_pedal, 'cause_id'] = 919
# Recode mental_schizo to _mental everywhere
# "USERNAME request 1/31/2017 to get rid of all mental_schizo as a cause
# of death and map to _mental"
# TODO implement
# TODO should this be yld_only, then? questionable...
# TODO if maintaining this, don't need restriction restricting
# mental_schizo to cc_code in Tibet
df.loc[schizo, 'cause_id'] = 919
# Recode msk and all sub-causes to cc_code in all VA
# "USERNAME and USERNAME request 2/14/2017 "msk recode to cc_code for all
# VA and SRS"
# this is in the bridge map already
# Recode cvd_pvd to cvd in Russia_FMD_1999_2011
# Russia 1999 2011 has a weird outlier for pvd, should be cvd according
# to USERNAME 02/13/2017
# TODO implement
if self.source == "Russia_FMD_1999_2011":
cvd_pvd = df['cause_id'] == 502
df.loc[cvd_pvd, 'cause_id'] = 491
# USERNAME said to remove this following recode 2/26/2018
# # In all VR USERNAME wants to move mental_drug deaths in under 15
# # to unintentional poisoning. -USERNAME 7/8/2015
# # cause_id 562 (mental_drug_opioids) has different age restrictions,
# # so recode it separately
# mental_causes_no_op = df['cause_id'].isin(
# [560, 561, 563, 564, 565, 566]
# )
# mental_no_op_ages = df['age_group_id'].isin(range(2, 8, 1))
# df.loc[mental_causes_no_op & mental_no_op_ages & vr, 'cause_id'] = 700
# mental_op = df['cause_id'] == 562
# mental_op_ages = df['age_group_id'].isin([4, 5, 6, 7])
# df.loc[mental_op & mental_op_ages & vr, 'cause_id'] = 700
# Temp fix for self imposed redistribution error
# move suicide and homicide in these years to cc_code
if self.source == "Iran_Mohsen_special_ICD10":
sui_homi_causes = [
717, 718, 719, 720, 721, 722, 723, 724, 725, 726, 727, 941
]
sui_homi = df['cause_id'].isin(sui_homi_causes)
bad_years = df['year_id'].isin(range(2007, 2015))
df.loc[sui_homi & bad_years, 'cause_id'] = 919
# In India MCCD neonatal sepsis should only be in under 1 month
if "India_MCCD" in self.source:
non_neonates = np.logical_not(df['age_group_id'].isin([2, 3]))
neonatal_sepsis = df['cause_id'].isin([])
df.loc[non_neonates & neonatal_sepsis, 'cause_id'] = 380
# In India_SCD_states_rural we are trying to get rid of all the
# redistribution artifacts
if self.source == "India_SCD_states_rural":
warnings.warn("Implement SCD rd artifact recode")
# Recoding state actor violence to war for proper schocks tracking
# in ICD9btl & icd10 inj_war_execution > inj_war_war in Ecuador '80-'90
inj_war_execution = df['cause_id'] == 854
if self.source == "ICD9_BTL":
ecuador = df['location_id'] == 122
year_1980_1990 = df['year_id'].isin(range(1980, 1991, 1))
df.loc[inj_war_execution & ecuador & year_1980_1990,
'cause_id'] = 855
# inj_war_execution > inj_war_war for BIH from 1985-91
bih = df['location_id'] == 44
year_1985_1991 = df['year_id'].isin(
[1985, 1986, 1987, 1988, 1989, 1990, 1991])
df.loc[inj_war_execution & bih & year_1985_1991, 'cause_id'] = 855
# in icd9_btl there are cancer recodes to be implemented here
warnings.warn("BTL cancer recode needed")
if self.source == "ICD10":
irq = df['location_id'] == 143
year_2008 = df['year_id'] == 2008
df.loc[inj_war_execution & year_2008 & irq, 'cause_id'] = 855
# USERNAME said cirrhosis and hepatitis in India SRS did not go very well (5/26/19)
# "Move any death from SRS in the final stage due to cirrhosis to hepatitis in under 15
# Move 30% death from SRS in the final stage due to cirrhosis to hepatitis in between 15-24"
if self.source == "India_SRS_states_report":
# There should be no cirrhosis subtypes in SRS, but include them in case things change
cirrhosis_ids = [521, 522, 523, 524, 971, 525]
hepatitis_id = 400
# Under 15
under_15 = df['age_group_id'] < 8
cirrhosis = df['cause_id'].isin(cirrhosis_ids)
df.loc[under_15 & cirrhosis, 'cause_id'] = hepatitis_id
# 15-24
start_deaths = df[self.val_cols].sum(axis=0)
# Create proportions to split
split_df = pd.DataFrame()
for age_group_id in [8, 9]:
for cirrhosis_id in cirrhosis_ids:
small_df = pd.DataFrame({
'new_cause_id': [cirrhosis_id, hepatitis_id],
'pct': [0.70, 0.30]
})
small_df['cause_id'] = cirrhosis_id
small_df['age_group_id'] = age_group_id
split_df = split_df.append(small_df, sort=True)
# Merge in the proportions and split
# Do not apply the split retroactively - can't take away deaths from
# cirrhosis in earlier phases if they aren't there yet
df = df.merge(split_df,
how='left',
on=['age_group_id', 'cause_id'])
matches = df.new_cause_id.notnull()
df.loc[matches, 'cause_id'] = df['new_cause_id']
df.loc[matches, 'deaths'] = df['deaths'] * df['pct']
for col in ['deaths_raw', 'deaths_corr', 'deaths_rd']:
df.loc[matches & (df['new_cause_id'] == hepatitis_id), col] = 0
df.drop(["new_cause_id", "pct"], axis='columns', inplace=True)
assert np.allclose(start_deaths, df[self.val_cols].sum(axis=0))
assert df.notnull().values.all()
# USERNAMEFm says we should not have congenital in older age groups
# in this study. USERNAME says that since congenital is created by the
# redistribution of sepsis for this study: "Result of redistrbution on sepsis
# have to be very low, if the problem is just this one drop result of redistribution
# due to sepsis"
# The larger question is if/when we should create causes in VA
malawi_va_study = df['nid'] == 413649
congenital = df.cause_id.isin(
get_all_related_causes('cong', cause_metadata_df))
df.loc[malawi_va_study & congenital, 'cause_id'] = 919
if self.source == "ICD9_detail":
if ((df['location_id'] == 43) & (df['year_id'] == 1997)).any():
warnings.warn("Albania homicide recode needed")
if self.source == "ICD9_USSR_Tabulated":
warnings.warn("Missing some homicide fixes for TJK, ARM here.")
df = self.drop_leukemia_subtypes(df)
# mortuary, burial, self-reported COD, census/survey,
# and tabulated hospital data should be reduced down to just
# injuries, maternal, and cc_code
if self.data_type_id in [1, 3, 5, 7]:
maternal_causes = get_all_related_causes('maternal',
cause_metadata_df)
injury_causes = get_all_related_causes('_inj', cause_metadata_df)
maternal = df['cause_id'].isin(maternal_causes)
inj = df['cause_id'].isin(injury_causes)
df.loc[~(maternal | inj), 'cause_id'] = 919
# for sibling history, we only want maternal and cc_code
if self.data_type_id == 5:
df.loc[~maternal, 'cause_id'] = 919
return df
def run_phase(df, cause_set_version_id, location_set_version_id, data_type_id,
env_run_id, source, nid, extract_type_id, remove_decimal,
code_map_version_id, iso3):
"""Run the full pipeline, chaining together CodProcesses."""
configurator = Configurator('standard')
cache_dir = configurator.get_directory('db_cache')
cache_options = {
'block_rerun': True,
'cache_dir': cache_dir,
'force_rerun': False,
'cache_results': False
}
# get cause hierarchy
cause_meta_df = get_current_cause_hierarchy(
cause_set_version_id=cause_set_version_id, **cache_options)
# get location hierarchy
location_meta_df = get_current_location_hierarchy(
location_set_version_id=location_set_version_id, **cache_options)
# get envelope
env_meta_df = get_env(env_run_id=env_run_id, **cache_options)
# get env with HIV
env_hiv_meta_df = get_env(env_run_id=env_run_id,
with_hiv=True,
**cache_options)
# get age groups
age_meta_df = get_ages(**cache_options)
code_system_id = int(
get_value_from_nid(nid,
'code_system_id',
extract_type_id=extract_type_id))
cause_map = get_cause_map(code_map_version_id=code_map_version_id,
**cache_options)
package_map = get_package_map(code_system_id=code_system_id,
**cache_options)
disagg_df = get_phase_output("disaggregation", nid, extract_type_id)
misdc_df = get_phase_output("misdiagnosiscorrection", nid, extract_type_id)
cause_package_hierarchy = get_cause_package_hierarchy(
code_system_id, **cache_options)
if source == "Cancer_Registry":
df = prune_cancer_registry_data(df, location_meta_df)
# aggregate location
# defaults to simple location -> national aggregation
# running full aggregation for India Survey data
print_log_message("Aggregating location to country level")
location_aggregator = LocationAggregator(df, location_meta_df)
if (data_type_id == 7) & (iso3 == 'IND'):
df = location_aggregator.get_computed_dataframe('full')
else:
df = location_aggregator.get_computed_dataframe()
if data_type_id in POLICE_SURVEY_DATA_TYPE:
# special step to remove HIV from maternal data
print_log_message("Removing HIV from cc_code for maternal data.")
maternal_hiv_remover = MaternalHIVRemover(df, env_meta_df,
env_hiv_meta_df, source, nid)
df = maternal_hiv_remover.get_computed_dataframe()
print_log_message("Calculating sample size")
df = calc_sample_size(df)
print_log_message(log_statistic(df))
print_log_message("Converting to cause fractions")
df = df.loc[df['sample_size'] > 0]
df = convert_to_cause_fractions(
df, ['deaths', 'deaths_rd', 'deaths_corr', 'deaths_raw'])
print_log_message(log_statistic(df))
if data_type_id == VA_DATA_TYPE:
# run VA anemia adjusment
print_log_message("Running VA Anemia adjustment")
va_anemia_adjuster = AnemiaAdjuster()
df = va_anemia_adjuster.get_computed_dataframe(df)
if data_type_id == POLICE_DATA_TYPE:
if source == 'Various_RTI':
rti_adjuster = RTIAdjuster(df, cause_meta_df, age_meta_df,
location_meta_df)
df = rti_adjuster.get_computed_dataframe()
if data_type_id in POLICE_SURVEY_DATA_TYPE:
# issue: rows with > 0 sample size are dropped
# most common in maternal data, but relevant anywhere
# we have only cc_code and one other cause and there
# are 0 deaths for the other cause for a given age/sex
cause_list = df.cause_id.unique()
square_me = (len(cause_list) == 2) & (CC_CODE in cause_list)
if (source in MATERNAL_SQUARED) or square_me:
print_log_message("Squaring maternal data")
df = square_maternal_sources(df, cause_meta_df, age_meta_df)
print_log_message("Dropping cc code")
df = drop_cc_code(df)
print_log_message(log_statistic(df))
print_log_message("Splitting locations.")
env_loc_splitter = EnvelopeLocationSplitter(df, env_meta_df, source)
df = env_loc_splitter.get_computed_dataframe()
print_log_message(log_statistic(df))
# aggregate causes
print_log_message("Aggregating causes")
cause_aggregator = CauseAggregator(df, cause_meta_df, source)
df = cause_aggregator.get_computed_dataframe()
print_log_message(log_statistic(df))
print_log_message("Adding parnt-mapped garbage to aggregated causes")
parent_gbg_adder = ParentMappedAggregatedGarbageAdder(
nid, extract_type_id, source, cause_package_hierarchy, cause_meta_df,
package_map, cause_map, remove_decimal, disagg_df, misdc_df)
df = parent_gbg_adder.get_computed_dataframe(df)
print_log_message("Applying hiv-prevalance in pregnancy adjustment to "
"maternal deaths")
hmp = HIVMatPAFs()
df = hmp.get_computed_dataframe(df, cause_meta_df, location_meta_df)
print_log_message(log_statistic(df))
# TO DO
# ** In the recode step for BTL some cancer deaths were moved to the
# cancer parent. The squaring step created 0's. Get rid of the 0's in
# country-years the recode was previously applied to.
print_log_message(
"Removing HIV and shocks from cause fraction denominator")
hiv_shock_remover = SampleSizeCauseRemover(cause_meta_df)
df = hiv_shock_remover.get_computed_dataframe(df)
print_log_message(log_statistic(df))
# not sure why we do this, but could use a comment of some kind.
df = conform_one_like_cf_to_one(df)
print_log_message("Verifying cause fractions not null between 0 and 1")
assert_valid_cause_fractions(df)
if dataset_has_redistribution_variance(data_type_id, source):
# Determine the redistribution variance
rdvar = RedistributionVarianceEstimator(
nid,
extract_type_id,
cause_meta_df,
remove_decimal,
code_system_id,
cause_map,
package_map,
code_map_version_id=code_map_version_id)
df = rdvar.get_computed_dataframe(df, **cache_options)
return df
get_current_location_hierarchy
)
from cod_prep.utils import (
print_log_message,
report_duplicates,
report_if_merge_fail,
cod_timestamp
)
from cod_prep.claude.claude_io import get_claude_data, makedirs_safely
from cod_prep.claude.configurator import Configurator
from save_proportions_for_tableau import SharedPackage
CONF = Configurator()
MODEL_DATA_CODE_SYSTEMS = [1, 6]
RDP_REG_DIR = CONF.get_directory('rdp_regressions')
def get_package_code_ids(regression_specification, code_system_id):
"""Returns code_ids for garbage codes in package for given code system"""
package_description = regression_specification[
'package_descriptions'
][code_system_id]
packages = get_package_list(code_system_id)
package_id = packages.loc[
packages['package_description'] == package_description,
'package_id'
]
assert len(package_id) == 1
package_id = package_id.iloc[0]
write to file for uploading
"""
CONF = Configurator('standard')
# sources containing maternal deaths that are noise reduced
MATERNAL_NR_SOURCES = [
"Mexico_BIRMM",
"Maternal_report",
"SUSENAS",
"China_MMS",
"China_Child",
]
NR_DIR = CONF.get_directory('nr_process_data')
def get_malaria_noise_reduction_model_result(malaria_model_group,
launch_set_id):
"""Read in the csv with saved malaria model result."""
malaria_dfs = []
for model_group in malaria_model_group:
if model_group != "NO_NR":
malaria_filepath = "FILEPATH".format(nr=NR_DIR,
model_group=model_group,
lsid=launch_set_id)
df = just_keep_trying(pd.read_csv,
args=[malaria_filepath],
max_tries=100,
seconds_between_tries=6,
class Recoder(CodProcess):
id_cols = [
'nid', 'extract_type_id', 'location_id', 'year_id', 'age_group_id',
'sex_id', 'cause_id', 'site_id'
]
val_cols = ['deaths', 'deaths_rd', 'deaths_corr', 'deaths_raw']
def __init__(self, cause_meta_df, source, code_system_id, data_type_id):
self.source = source
self.code_system_id = code_system_id
self.data_type_id = data_type_id
self.cause_meta_df = cause_meta_df
self.conf = Configurator("standard")
self.vr_indicators_path = self.conf.get_resource('vr_indicators')
self.cache_options = {
'force_rerun': False,
'block_rerun': True,
'cache_results': False,
'cache_dir': self.conf.get_directory('db_cache')
}
def get_computed_dataframe(self, df):
if 'data_type_id' not in df.columns:
df = add_nid_metadata(df, "data_type_id", **self.cache_options)
df = self.recode(df)
df = self.conform_secret_causes(df)
df = self.clean_up(df)
return df
def get_diagnostic_dataframe(self):
"""Return diagnostics."""
pass
def recode_sids(self, df):
path_to_4_stars_sheet = self.conf.get_resource("four_star_locations")
four_five_star_locs = pd.read_csv(path_to_4_stars_sheet)
four_five_star_locs = four_five_star_locs[['location_id']]
four_five_star_locs = four_five_star_locs.location_id.unique()
less_than_four_star = ~df['location_id'].isin(four_five_star_locs)
is_sids = df['cause_id'] == 686
df.loc[is_sids & less_than_four_star, 'cause_id'] = 380
return df
def clean_up(self, df):
"""Group rogue duplicates."""
df = df.groupby(self.id_cols, as_index=False)[self.val_cols].sum()
return df
def conform_secret_causes(self, df):
df = add_cause_metadata(df,
add_cols=['secret_cause', 'parent_id'],
cause_meta_df=self.cause_meta_df,
**self.cache_options)
injuries_replace_parents = [722, 720, 719]
replaced_injuries = df['cause_id'].isin(injuries_replace_parents)
df.loc[replaced_injuries, 'parent_id'] = 723
secret_causes = df['secret_cause'] == 1
not_cc_code = df['cause_id'] != 919
len_before = len(df)
if df['parent_id'].isnull().values.any():
raise AssertionError('There are missing parent cause_ids')
df.loc[secret_causes & not_cc_code, 'cause_id'] = df['parent_id']
len_after = len(df)
if len_before != len_after:
raise AssertionError(
'The length of the dataframe has changed from {} to {}'.format(
len_before, len_after))
df.drop(['parent_id', 'secret_cause'], axis=1, inplace=True)
return df
def drop_leukemia_subtypes(self, df):
leuk_subtypes = get_all_related_causes('neo_leukemia',
self.cause_meta_df)
leuk_subtypes.remove(487)
df.loc[(df['cause_id'].isin(leuk_subtypes)) & (df['deaths_rd'] > 0) &
(df['deaths_raw'] <= 0), 'cause_id'] = 487
return df
def recode(self, df):
cause_metadata_df = self.cause_meta_df
cause_metadata_df = cause_metadata_df[[
"cause_id", "path_to_top_parent", "acause"
]]
ckd_cause_ids = get_all_related_causes('ckd', cause_metadata_df)
ckd_cause_ids.remove(593)
ckd_less_other = df['cause_id'].isin(ckd_cause_ids)
neonate = df['age_group_id'].isin([2, 3])
df.loc[ckd_less_other & neonate, 'cause_id'] = 652
resp_ids = [509, 515, 516, 520]
is_cert_resp_causes = df['cause_id'].isin(resp_ids)
df.loc[is_cert_resp_causes & neonate, 'cause_id'] = 322
is_asthma = df['cause_id'] == 515
df.loc[is_asthma & (df['age_group_id'] == 4), 'cause_id'] = 322
maternal_cause_ids = get_all_related_causes(366, cause_metadata_df)
maternal_cause_ids = df['cause_id'].isin(maternal_cause_ids)
non_maternal_ages = np.logical_not(df['age_group_id'].isin(
[7, 8, 9, 10, 11, 12, 13, 14, 15, 22]))
df.loc[maternal_cause_ids & non_maternal_ages, 'cause_id'] = 919
alzheimers = df['cause_id'] == 543
under_40 = df['age_group_id'].isin(range(1, 13, 1))
df.loc[alzheimers & under_40, 'cause_id'] = 919
cong_causes = get_all_related_causes('cong', cause_metadata_df)
congenital = df['cause_id'].isin(cong_causes)
over_70 = df['age_group_id'].isin([19, 20, 30, 31, 32, 235])
df.loc[congenital & over_70, "cause_id"] = 919
hepatitis = get_all_related_causes(400, cause_metadata_df)
hepatitis = df['cause_id'].isin(hepatitis)
if self.code_system_id in [7, 9]:
df.loc[hepatitis & neonate, "cause_id"] = 380
else:
df.loc[hepatitis & neonate, "cause_id"] = 384
inj_disaster_light = df['cause_id'] == 984
df.loc[inj_disaster_light, 'cause_id'] = 716
if self.code_system_id not in [1, 6]:
ckd_diabetes = df['cause_id'].isin([997, 998])
df.loc[ckd_diabetes, 'cause_id'] = 589
if self.code_system_id not in [1, 6, 9]:
diabetes_subtypes = df['cause_id'].isin([975, 976])
df.loc[diabetes_subtypes, 'cause_id'] = 587
diabetes_type_2 = df['cause_id'] == 976
under_15 = df['age_group_id'] < 8
df.loc[diabetes_type_2 & under_15, 'cause_id'] = 975
iron_or_iodine = df['cause_id'].isin([388, 390])
df.loc[iron_or_iodine, 'cause_id'] = 919
under_1 = df['age_group_id'] < 5
cvd_ihd = df['cause_id'] == 493
df.loc[cvd_ihd & under_1, 'cause_id'] = 643
if 686 in df.cause_id.unique():
df = self.recode_sids(df)
df.loc[df.cause_id.isin([344, 409, 410, 542, 558, 669, 680, 961]),
'cause_id'] = 919
if self.data_type_id not in [6, 7, 8]:
df.loc[df['cause_id'] == 687, 'cause_id'] = 919
one_to_14 = df['age_group_id'].isin([5, 6, 7])
cvd_ihd = df['cause_id'] == 493
df.loc[cvd_ihd & one_to_14, 'cause_id'] = 507
cancer_recodes = get_all_related_causes([
411, 414, 423, 426, 429, 432, 435, 438, 441, 444, 450, 453, 456,
459, 462, 465, 468, 474, 486, 483
], cause_metadata_df)
cancer_recodes = df['cause_id'].isin(cancer_recodes)
cancer_ages = df['age_group_id'].isin(range(2, 8, 1))
df.loc[cancer_recodes & cancer_ages, "cause_id"] = 489
not_icd10 = self.code_system_id != 1
neo_meso = df['cause_id'] == 483
df.loc[neo_meso & not_icd10, "cause_id"] = 489
if self.source.endswith("AAMSP"):
digest_hernia = df['cause_id'].isin([531])
df.loc[digest_hernia, "cause_id"] = 919
if self.source == "":
homicide_and_suicide = df['cause_id'].isin(
[724, 725, 726, 727, 941, 718, 719, 720, 721, 722, 723])
bad_years = df['year_id'].isin(range(2007, 2015))
# _unintent
df.loc[bad_years & homicide_and_suicide, "cause_id"] = 919
inj_war = get_all_related_causes(945, cause_metadata_df)
is_inj_war = df['cause_id'].isin(inj_war)
jamaica = df['location_id'] == 115
year_2005 = df['year_id'] == 2005
vr = df['data_type_id'] == 9
df.loc[is_inj_war & jamaica & year_2005 & vr, 'cause_id'] = 724
inj_mech_gun = df['cause_id'] == 705
year_2006 = df['year_id'] == 2006
df.loc[inj_mech_gun & year_2006 & jamaica & vr, 'cause_id'] = 724
if self.source == "ICD10":
digest_ibd = df['cause_id'] == 532
suriname = df['location_id'] == 118
year_1995_2012 = df['year_id'].isin(range(1995, 2013, 1))
df.loc[digest_ibd & suriname & year_1995_2012, 'cause_id'] = 526
endo_prodcedural = df['cause_id'] == 624
df.loc[endo_prodcedural, 'cause_id'] = 708
schizo = df['cause_id'] == 559
tibet = df['location_id'] == 518
df.loc[schizo & tibet, 'cause_id'] = 919
hiv = get_all_related_causes(298, cause_metadata_df)
hiv = df['cause_id'].isin(hiv)
pre_1980 = df['year_id'] < 1980
df.loc[hiv & pre_1980, 'cause_id'] = 919
diabetes_causes = get_all_related_causes(587, cause_metadata_df)
diabetes = df['cause_id'].isin(diabetes_causes)
df.loc[neonate & diabetes, 'cause_id'] = 380
under_20 = df['age_group_id'].isin(range(0, 8, 1))
stroke = get_all_related_causes('cvd_stroke', cause_metadata_df)
stroke_deaths = df['cause_id'].isin(stroke)
va = df['data_type_id'] == 8
df.loc[under_20 & stroke_deaths & va, 'cause_id'] = 491
over_95 = df['age_group_id'] == 235
inj_trans_road_pedal = df['cause_id'] == 691
df.loc[over_95 & inj_trans_road_pedal, 'cause_id'] = 919
df.loc[schizo, 'cause_id'] = 919
if self.source == "Russia_FMD_1999_2011":
cvd_pvd = df['cause_id'] == 502
df.loc[cvd_pvd, 'cause_id'] = 491
if self.source == "":
sui_homi_causes = [
717, 718, 719, 720, 721, 722, 723, 724, 725, 726, 727, 941
]
sui_homi = df['cause_id'].isin(sui_homi_causes)
bad_years = df['year_id'].isin(range(2007, 2015))
df.loc[sui_homi & bad_years, 'cause_id'] = 919
if "India_MCCD" in self.source:
non_neonates = np.logical_not(df['age_group_id'].isin([2, 3]))
neonatal_sepsis = df['cause_id'].isin([])
df.loc[non_neonates & neonatal_sepsis, 'cause_id'] = 380
if self.source == "India_SCD_states_rural":
warnings.warn("Implement SCD rd artifact recode")
inj_war_execution = df['cause_id'] == 854
if self.source == "ICD9_BTL":
ecuador = df['location_id'] == 122
year_1980_1990 = df['year_id'].isin(range(1980, 1991, 1))
df.loc[inj_war_execution & ecuador & year_1980_1990,
'cause_id'] = 855
bih = df['location_id'] == 44
year_1985_1991 = df['year_id'].isin(
[1985, 1986, 1987, 1988, 1989, 1990, 1991])
df.loc[inj_war_execution & bih & year_1985_1991, 'cause_id'] = 855
warnings.warn("BTL cancer recode needed")
if self.source == "ICD10":
irq = df['location_id'] == 143
year_2008 = df['year_id'] == 2008
df.loc[inj_war_execution & year_2008 & irq, 'cause_id'] = 855
if self.source == "ICD9_detail":
if ((df['location_id'] == 43) & (df['year_id'] == 1997)).any():
warnings.warn("Albania homicide recode needed")
if self.source == "ICD9_USSR_Tabulated":
warnings.warn("Missing some homicide fixes for TJK, ARM here.")
df = self.drop_leukemia_subtypes(df)
if self.data_type_id in [1, 3, 5, 7]:
maternal_causes = get_all_related_causes('maternal',
cause_metadata_df)
injury_causes = get_all_related_causes('_inj', cause_metadata_df)
maternal = df['cause_id'].isin(maternal_causes)
inj = df['cause_id'].isin(injury_causes)
df.loc[~(maternal | inj), 'cause_id'] = 919
if self.data_type_id == 5:
df.loc[~maternal, 'cause_id'] = 919
return df
class HIVMatPAFs(CodProcess):
calc_cf_col = 'cf'
all_cf_cols = ['cf', 'cf_raw', 'cf_corr', 'cf_rd']
def __init__(self):
self.configurator = Configurator('standard')
self.cache_dir = self.configurator.get_directory('db_cache')
self.maternal_hiv_props_path = \
self.configurator.get_directory('maternal_hiv_props')
def get_computed_dataframe(self, df, cause_meta_df, location_meta_df):
restricted_maternal_df = \
self.restrict_to_maternal_data(df, cause_meta_df)
if restricted_maternal_df is None:
return df
appended_pafs = self.append_maternal_pafs(
restricted_maternal_df.year_id.unique())
merged_data = \
self.merge_data_and_proportions(restricted_maternal_df,
appended_pafs)
percent_maternal = self.generate_percentages(merged_data)
split_maternal = self.generate_splits(percent_maternal)
hiv_cfs = self.create_maternal_hiv_cfs(split_maternal)
cleaned = self.clean_adjusted_data(hiv_cfs)
final = \
self.append_adjusted_orig(df, restricted_maternal_df, cleaned)
group_cols = [
col for col in final.columns
if col not in self.all_cf_cols and col not in ['sample_size']
]
final = final.groupby(group_cols, as_index=False).agg({
'sample_size': 'mean',
'cf': 'sum',
'cf_raw': 'sum',
'cf_corr': 'sum',
'cf_rd': 'sum'
})
return final
def restrict_to_maternal_data(self, df, cause_meta_df):
"""Restrict incoming dataframe to only maternal data."""
df = df.copy()
# get age start and age end for maternal ages
maternal_metadata = cause_meta_df.loc[cause_meta_df['cause_id'] == 366]
age_start = maternal_metadata['yll_age_start']
assert len(age_start) == 1
age_start = age_start.iloc[0]
age_end = maternal_metadata.yll_age_end
assert len(age_end) == 1
age_end = age_end.iloc[0]
data = add_age_metadata(df,
add_cols=['simple_age'],
merge_col='age_group_id',
force_rerun=False,
block_rerun=True,
cache_results=False,
cache_dir=self.cache_dir)
data.rename(columns={'simple_age': 'age'}, inplace=True)
maternal_data = data.loc[(df['cause_id'] == 366)
& (data['age'] >= age_start) &
(data['age'] <= age_end) &
(data['sex_id'] == 2) &
(data['year_id'] >= 1980)]
maternal_data.drop('age', axis=1, inplace=True)
if len(maternal_data) == 0:
return None
else:
return maternal_data
def append_maternal_pafs(self, years):
"""Read in proportions."""
props = pd.DataFrame()
for year in years:
year = int(year)
if os.path.isfile("{}maternal_hiv_props_{}.csv".format(
self.maternal_hiv_props_path, year)):
data = pd.read_csv("FILEPATH".format(year))
props = props.append(data)
props = props.rename(columns={'year': 'year_id'})
return props
def duplicate_national_props(self, props_df, loc_df):
subnational = loc_df.loc[
loc_df['level'] > 3,
['location_id', 'parent_id', 'level', 'path_to_top_parent']]
# Russia sub nationals are level 5 while other countries are level 4
subnational.loc[
subnational['level'] == 5, 'parent_id'] = \
subnational['path_to_top_parent'].str.split(',').str[3].astype(int)
# only keep rows with the needed sub national locations
subnational = subnational.loc[subnational['parent_id'].isin(
self.need_subnational_props)]
# drop level 4 sub national location_ids for Russia
subnational = subnational.loc[~((subnational['parent_id'] == 62) &
(subnational['level'] == 4))]
subnational = subnational[['location_id', 'parent_id']]
subnational.rename(columns={
'location_id': 'child_location_id',
'parent_id': 'location_id'
},
inplace=True)
# create sub national maternal_hiv proportions from national
subnational = props_df.merge(subnational, on='location_id')
subnational.drop('location_id', axis=1, inplace=True)
subnational.rename(columns={'child_location_id': 'location_id'},
inplace=True)
props_df = pd.concat([props_df, subnational])
assert not props_df.duplicated().any(), 'please check maternal'\
' proportions, there are duplicates'
return props_df
def merge_data_and_proportions(self, data, props):
"""Merge restricted maternal data and proportions."""
merged_data = data.merge(props,
on=['location_id', 'age_group_id', 'year_id'],
how='left')
assert merged_data.notnull().values.any(), 'maternal proportions '\
'were not successfully merged with incoming data'
return merged_data
def generate_percentages(self, df):
df['pct_maternal'] = 1 - df['pct_hiv'] - df['pct_maternal_hiv']
df.loc[df['pct_maternal'].isnull(), 'pct_maternal'] = 1
df.loc[df['pct_hiv'].isnull(), 'pct_hiv'] = 0
df.loc[df['pct_maternal_hiv'].isnull(), 'pct_maternal_hiv'] = 0
assert all(x > 0 for x in df['pct_maternal'])
assert df[['pct_maternal', 'pct_hiv', 'pct_maternal_hiv'
]].notnull().values.any(), 'there are missing percentages'
assert all(
abs(df['pct_maternal'] + df['pct_hiv'] + df['pct_maternal_hiv']) -
1) < .0001
assert (df['pct_maternal_hiv_vr'] <= .13).all()
assert not (df['cause_id'] == 741).any()
return df
''' '''
def generate_splits(self, df):
df = add_nid_metadata(
df,
add_cols='data_type_id',
block_rerun=True,
cache_dir=self.cache_dir,
force_rerun=False,
)
df.loc[df['data_type_id'].isin([7, 5]), 'split_maternal'] = 1
df.loc[df['split_maternal'].isnull(), 'split_maternal'] = 0
df.loc[df['split_maternal'] == 0, 'pct_maternal'] = 1
df.loc[df['split_maternal'] == 0,
'pct_maternal_hiv'] = df['pct_maternal_hiv_vr']
df.loc[df['split_maternal'] == 0, 'pct_hiv'] = 0
df.drop('pct_maternal_hiv_vr', axis=1, inplace=True)
return df
def create_maternal_hiv_cfs(self, df):
df = df.copy()
maternal_hiv_df = df.copy()
maternal_hiv_df['cf'] = maternal_hiv_df['cf'] * \
maternal_hiv_df['pct_maternal_hiv']
maternal_hiv_df['cause_id'] = 741
maternal_hiv_df['cf_raw'] = 0
maternal_hiv_df['cf_corr'] = 0
maternal_hiv_df['cf_rd'] = 0
maternal_df = df.copy()
maternal_df['cf'] = maternal_df['cf'] * maternal_df['pct_maternal']
maternal_df['cause_id'] = 366
df = pd.concat([maternal_hiv_df, maternal_df], ignore_index=True)
return df
def clean_adjusted_data(self, df):
va_vr = df.loc[df['split_maternal'] == 0]
if len(va_vr) > 0:
assert set([741, 366]) == set(va_vr.cause_id.unique())
va_vr = va_vr.loc[va_vr['cause_id'] != 366]
va_vr['cause_id'] = 366
df = pd.concat([df, va_vr], ignore_index=True)
df = df.groupby([
'nid', 'extract_type_id', 'location_id', 'year_id', 'site_id',
'age_group_id', 'sex_id', 'sample_size', 'cause_id'
],
as_index=False)[self.all_cf_cols].sum()
assert (df['cf'] < 1.1).all()
df.loc[df['cf'] > 1, 'cf'] = 1
return df
def append_adjusted_orig(self, orig, maternal_data, adjusted):
"""Remove original maternal data and append on adjusted."""
data = orig.merge(maternal_data, how='left', indicator=True)
data = data.loc[data['_merge'] != 'both']
data.drop('_merge', axis=1, inplace=True)
data = data.append(adjusted, ignore_index=True)
return data
