from census.utils import cols_name_to_indices
from sklearn.ensemble import RandomForestClassifier
from census.transformer import MyScaler
from sklearn.pipeline import Pipeline
from census.shortcuts import load_datas
print("""
Loading data from disk and transform it with :
- categorical variables are label so that the corresponding
numerical values are growing with the proportion of +50.000$
""")
limit_to_rows = 199523 # corresponds to the whole dataset
X, y, X_test, y_test, columns, cols_interpret = load_datas(
nrows=limit_to_rows,
sort_mapping=True,
booleans=False
# dummify=True
)
# SET MODEL TYPE IN A PIPELINE
scale_cols = ['AGE', 'INSTANCE_WEIGHT', 'WWORKY', 'WPERH',
'CAP_GAINS', 'CAP_LOSSES', 'STOCK_DIV']
idx_scale_cols = cols_name_to_indices(columns, scale_cols)
ms = MyScaler(idx_scale_cols)
rf = RandomForestClassifier(n_jobs=-1)
predictor = Pipeline([('my_scaling', ms), ('rf', rf)])
print("""
---------------------------------
from census.shortcuts import load_datas
print("""
Loading data from disk and transform it with :
- dummification of categorical variables
- removing of some columns (see below)
- add boolean variables when numerical inputs code in fact two variables
eg: WWORKY weeks worked per year : have instance worked in term of weeks ?
if yes, how much ?
""")
limit_to_rows = 199523 # corresponds to the whole dataset
X, y, X_test, y_test, columns, cols_interpret = load_datas(
nrows=limit_to_rows,
remove_cols=['YEAR', 'RESI_REGION', 'RESI_PREV', 'RESI_1YEAR'],
sort_mapping=False,
booleans=True,
dummify=True
)
# SET MODEL TYPE IN A PIPELINE
scale_cols = ['AGE', 'INSTANCE_WEIGHT', 'WWORKY', 'WPERH',
'CAP_GAINS', 'CAP_LOSSES', 'STOCK_DIV']
idx_scale_cols = cols_name_to_indices(columns, scale_cols)
ms = MyScaler(idx_scale_cols)
rf = RandomForestClassifier(n_jobs=-1)
predictor = Pipeline([('my_scaling', ms), ('rf', rf)])
print("""
from census.learn import FindBest, most_influential, report_influential, score_report
from census.utils import cols_name_to_indices
from sklearn.linear_model import LogisticRegression
from census.transformer import MyScaler
from sklearn.pipeline import Pipeline
from census.shortcuts import load_datas
# LOAD DATA
limit_to_rows = 199523 # corresponds to the whole dataset
X, y, X_test, y_test, columns, cols_interpret = load_datas(
nrows=limit_to_rows,
remove_cols=[
"YEAR",
"RESI_REGION",
"RESI_PREV",
"RESI_1YEAR",
# 'MIG_REGION', 'MIG_MOVE', 'MIG_MSA', 'INSTANCE_WEIGHT',
# 'ORIG_MOTHER'
],
dummify=True,
)
# SET MODEL TYPE IN A PIPELINE
scale_cols = ["AGE", "INSTANCE_WEIGHT", "WWORKY", "WPERH", "CAP_GAINS", "CAP_LOSSES", "STOCK_DIV"]
idx_scale_cols = cols_name_to_indices(columns, scale_cols)
ms = MyScaler(idx_scale_cols)
lr = LogisticRegression(penalty="l1")
predictor = Pipeline([("my_scaling", ms), ("lr", lr)])
