'main': all_df.AMT_CREDIT,
'prev': prev_df.AMT_CREDIT.sum(),
'buro1': buro_df.AMT_CREDIT_SUM.sum(),
'buro2': buro_df.AMT_CREDIT_SUM_DEBT.sum(),
'buro3': buro_df.AMT_CREDIT_SUM_LIMIT.sum(),
'income': all_df.AMT_INCOME_TOTAL
}).fillna(0)
for i in range(1, 4):
df[f'sub_credit{i}'] = df['prev'] + df[f'buro{i}']
df[f'all_credit{i}'] = df[f'sub_credit{i}'] + df['main']
df[f'sub_credit{i}_to_income'] = df[f'sub_credit{i}'] / df['income']
df[f'all_credit{i}_to_income'] = df[f'all_credit{i}'] / df['income']
self.df = df.filter(regex='(^sub_|^all_)')
if __name__ == '__main__':
args = get_arguments('main')
with timer('load dataset'):
train = pd.read_feather(TRAIN)
test = pd.read_feather(TEST)
prev = pd.read_feather(PREV)
buro = pd.read_feather(BURO)
with timer('preprocessing'):
prev = prev.query("NAME_CONTRACT_TYPE != 'XNA'")
with timer('create dataset'):
generate_features(globals(), args.force)
# trn = train.merge(prev_df, left_on='SK_ID_CURR', right_index=True, how='left')
# tst = test.merge(prev_df, left_on='SK_ID_CURR', right_index=True, how='left')
# for m, p in itertoolsd.product(main_cols, prev_cols):
# self.train[f'{m}_sub_{p}_max'] = trn[m] - trn[p + '_max']
# self.train[f'{m}_sub_{p}_mean'] = trn[m] - trn[p + '_mean']
# self.test[f'{m}_sub_{p}_max'] = tst[m] - tst[p + '_max']
# self.test[f'{m}_sub_{p}_mean'] = tst[m] - tst[p + '_mean']
#
# self.train[f'{m}_div_{p}_max'] = trn[m] / trn[p + '_max']
# self.train[f'{m}_div_{p}_mean'] = trn[m] / trn[p + '_mean']
# self.test[f'{m}_div_{p}_max'] = tst[m] / tst[p + '_max']
# self.test[f'{m}_div_{p}_mean'] = tst[m] / tst[p + '_mean']
if __name__ == '__main__':
args = get_arguments(Path(__file__).stem)
with timer('load dataset'):
train = pd.read_feather(TRAIN)
test = pd.read_feather(TEST)
prev = pd.read_feather(PREV)
with timer('preprocessing'):
prev.drop(['SK_ID_PREV', 'RATE_INTEREST_PRIMARY', 'RATE_INTEREST_PRIVILEGED'], axis=1, inplace=True)
prev = prev.sort_values(['SK_ID_CURR', 'DAYS_DECISION']).reset_index(drop=True)
prev.loc[:, prev.columns.str.startswith('DAYS_')] = prev.filter(regex='^DAYS_').replace({365243: np.nan})
prev.AMT_DOWN_PAYMENT.fillna(0)
prev.RATE_DOWN_PAYMENT.fillna(0)
cat_cols = prev.select_dtypes(['object']).columns
prev[cat_cols] = prev[cat_cols].fillna('NaN')
with timer('create dataset'):