def PreprocessOriginalData(train, test, filename):
# preprocessing training/testing
print("preprocessing on whole training/testing set")
# deepcopy the train/test set to preprocess
from copy import deepcopy
train = deepcopy(train)
test = deepcopy(test)
test.X = pd.DataFrame.copy(test.X)
# feature engineering
train.X = firststeps(train.X)
test.X = firststeps(test.X)
# fillNA
train.X, NAmethod = PP.fillNA(train.X)
test.X, _ = PP.fillNA(test.X, NAmethod)
# Empirical Bayesian Encoding
encoding_lst = [col for col in train.X.columns if col.endswith('cat')
and len(pd.concat([train.X, test.X])[col].unique()) >= 13]
train.X, Encodingmethod = PP.encoding(train, encoding_lst)
test.X, _ = PP.encoding(test, encoding_lst, param = Encodingmethod)
if True:
# set down which list need 2b dummied
mydummylist = PP.dummylist(
pd.concat([train.X, test.X], axis=0))
myheader = PP.makehead(
pd.concat([train.X, test.X], axis=0), mydummylist)
train.X = PP.dummy(
PP.addhead(train.X, myheader), mydummylist)
train.X = PP.rmhead(train.X)
test.X = PP.dummy(
PP.addhead(test.X, myheader), mydummylist)
test.X = PP.rmhead(test.X)
train.X, STDmethod = PP.standardize(train.X)
test.X, _ = PP.standardize(test.X, STDmethod)
print("saving %strain_X.csv (%d row * %d col)"%(filename, train.X.shape[0], train.X.shape[1]))
train.X.to_csv(filename + "train_X.csv",
index=False, float_format="%.5f")
train.y.to_csv(filename + "train_y.csv",
index=False, float_format="%.5f")
print("saving %stest_X.csv (%d row * %d col)"%(filename, test.X.shape[0], test.X.shape[1]))
test.X.to_csv(filename + "test_X.csv",
index=False, float_format="%.5f")
test.y.to_csv(filename + "test_y.csv",
index=False, float_format="%.5f")
return mydummylist, myheader, encoding_lst
def KFoldsPreprocess(train, test, mydummylist, myheader, encoding_lst, KFOLDS, filename):
folds = list(StratifiedKFold(n_splits=KFOLDS, shuffle=True,
random_state=10086).split(train.X, train.y))
# do methods in preprocessing in each fold
for i, (train_idx, valid_idx) in enumerate(folds):
print("preprocessing on cv #%d" % i)
train_this_cut = dataset(
train.X.iloc[train_idx], train.y.iloc[train_idx])
valid_this_cut = dataset(
train.X.iloc[valid_idx], train.y.iloc[valid_idx])
# feature engineering
train_this_cut.X = firststeps(train_this_cut.X)
valid_this_cut.X = firststeps(valid_this_cut.X)
# fillNA
train_this_cut.X, NAmethod = PP.fillNA(train_this_cut.X)
# Empirical Bayesian Encoding
train_this_cut.X, Encodingmethod = PP.encoding(train_this_cut, encoding_lst)
train_this_cut.X = PP.dummy(
PP.addhead(train_this_cut.X, myheader), mydummylist)
train_this_cut.X = PP.rmhead(train_this_cut.X)
train_this_cut.X, STDmethod = PP.standardize(
train_this_cut.X)
valid_this_cut.X, _ = PP.fillNA(
valid_this_cut.X, NAmethod)
valid_this_cut.X, _ = PP.encoding(valid_this_cut, encoding_lst, param = Encodingmethod)
valid_this_cut.X = PP.dummy(
PP.addhead(valid_this_cut.X, myheader), mydummylist)
valid_this_cut.X = PP.rmhead(valid_this_cut.X)
valid_this_cut.X, _ = PP.standardize(
valid_this_cut.X, STDmethod)
print("saving %strain_X_%d.csv (%d row * %d col)"%(filename, i, train_this_cut.X.shape[0], train_this_cut.X.shape[1]))
train_this_cut.X.to_csv(filename + "train_X_%d.csv" %
i, index=False, float_format="%.5f")
train_this_cut.y.to_csv(filename + "train_y_%d.csv" %
i, index=False, float_format="%.5f")
print("saving %svalid_X_%d.csv (%d row * %d col)"%(filename, i, valid_this_cut.X.shape[0], valid_this_cut.X.shape[1]))
valid_this_cut.X.to_csv(filename + "valid_X_%d.csv" %
i, index=False, float_format="%.5f")
valid_this_cut.y.to_csv(filename + "valid_y_%d.csv" %
i, index=False, float_format="%.5f")
