def main():
# -----------------------------------------------------------------------------------
# Adjustable Parameters
parser = argparse.ArgumentParser()
parser.add_argument('--train',
action='store_true',
help='training or scoring')
parser.add_argument('--inputfile', type=str, help='input data file name')
parser.add_argument('--outputfile',
type=str,
help='output prediction file name')
args = parser.parse_args()
# directory for the input data and output prediction:
DATA_DIR = 'data'
OUTPUT_DIR = 'output'
# columns used:
CAT_COLS = [
'Auction', 'Transmission', 'WheelType', 'Nationality', 'Size',
'TopThreeAmericanName', 'IsOnlineSale'
]
NUM_COLS = [
'VehicleAge', 'VehOdo', 'VehBCost', 'WarrantyCost',
'MMRCurrentAuctionAveragePrice', 'MMRAcquisitionAuctionAveragePrice',
'MMRCurrentAuctionCleanPrice', 'MMRAcquisitionAuctionCleanPrice',
'MMRCurrentRetailAveragePrice', 'MMRAcquisitionRetailAveragePrice',
'MMRCurrentRetailCleanPrice', 'MMRAcquisitonRetailCleanPrice'
]
DATE_COLS = ['PurchDate']
LABEL_COL = 'IsBadBuy'
IDS_COL = 'RefId'
# current time for computing recency feature
NOW = '2010-12-31'
# modeling step:
# model checkpoint for future scoring
MODEL_DIR = 'model'
CHECKPOINT_PREPROCESS = os.path.join(MODEL_DIR, 'preprocess.pkl')
CHECKPOINT_XGB = os.path.join(MODEL_DIR, 'xgb.pkl')
# parameter that only relevant for training stage and not scoring
if args.train:
# number of cross validation and hyperparameter settings to try
CV = 10
N_ITER = 5
MODEL_RANDOM_STATE = 4321
# train/validation stratified split
VAL_SIZE = 0.1
TEST_SIZE = 0.1
SPLIT_RANDOM_STATE = 1234
# -----------------------------------------------------------------------------------
logger.info('preprocessing')
input_path = os.path.join(DATA_DIR, args.inputfile)
if args.train:
data = clean(input_path, NOW, CAT_COLS, NUM_COLS, DATE_COLS, IDS_COL,
LABEL_COL)
ids = data[IDS_COL].values
label = data[LABEL_COL].values
data = data.drop([IDS_COL, LABEL_COL], axis=1)
# train/test split twice to achieve train/validaion/test three way split
df_train, df_test, y_train, y_test, ids_train, ids_test = train_test_split(
data,
label,
ids,
test_size=TEST_SIZE,
random_state=SPLIT_RANDOM_STATE,
stratify=label)
df_train, df_val, y_train, y_val, ids_train, ids_val = train_test_split(
df_train,
y_train,
ids_train,
test_size=VAL_SIZE,
random_state=SPLIT_RANDOM_STATE,
stratify=y_train)
# obtain finalized columns
num_cols_cleaned = list(
SortedSet(df_train.columns) - SortedSet(CAT_COLS))
preprocess = Preprocesser(num_cols=num_cols_cleaned, cat_cols=CAT_COLS)
X_train = preprocess.fit_transform(df_train)
X_val = preprocess.transform(df_val)
X_test = preprocess.transform(df_test)
logger.info('modeling')
eval_set = [(X_train, y_train), (X_val, y_val)]
xgb_tuned = build_xgb(N_ITER, CV, MODEL_RANDOM_STATE, eval_set)
xgb_tuned.fit(X_train, y_train)
if not os.path.isdir(MODEL_DIR):
os.mkdir(MODEL_DIR)
dump(preprocess, CHECKPOINT_PREPROCESS)
dump(xgb_tuned, CHECKPOINT_XGB)
# model evaluation metric reporting
y_pred = []
xgb_best = xgb_tuned.best_estimator_
zipped = zip(('train', 'validation', 'test'), (X_train, X_val, X_test),
(y_train, y_val, y_test))
for name, X, y in zipped:
xgb_pred = xgb_best.predict_proba(
X, ntree_limit=xgb_best.best_ntree_limit)[:, 1]
score = round(roc_auc_score(y, xgb_pred), 3)
logger.info('{} AUC: {}'.format(name, score))
y_pred.append(xgb_pred)
ids = np.hstack((ids_train, ids_val, ids_test))
y_pred = np.hstack(y_pred)
else:
data = clean(input_path, NOW, CAT_COLS, NUM_COLS, DATE_COLS, IDS_COL)
ids = data[IDS_COL].values
data = data.drop(IDS_COL, axis=1)
logger.info('scoring')
preprocess = load(CHECKPOINT_PREPROCESS)
xgb_tuned = load(CHECKPOINT_XGB)
X = preprocess.transform(data)
xgb_best = xgb_tuned.best_estimator_
y_pred = xgb_best.predict_proba(
X, ntree_limit=xgb_best.best_ntree_limit)[:, 1]
if not os.path.isdir(OUTPUT_DIR):
os.mkdir(OUTPUT_DIR)
output_path = os.path.join(OUTPUT_DIR, args.outputfile)
write_output(ids, IDS_COL, y_pred, LABEL_COL, output_path)
def main():
# -----------------------------------------------------------------------------------
# Adjustable Parameters
parser = argparse.ArgumentParser()
parser.add_argument(
'--train', action = 'store_true', help = 'training or scoring')
parser.add_argument(
'--inputfile', type = str, help = 'input data file name')
parser.add_argument(
'--outputfile', type = str, help = 'output prediction file name')
args = parser.parse_args()
# directory for the input data and output prediction:
DATA_DIR = 'data'
OUTPUT_DIR = 'output'
# columns used:
CAT_COLS = ['Auction', 'Transmission', 'WheelType', 'Nationality',
'Size', 'TopThreeAmericanName', 'IsOnlineSale']
NUM_COLS = ['VehicleAge', 'VehOdo', 'VehBCost', 'WarrantyCost',
'MMRCurrentAuctionAveragePrice', 'MMRAcquisitionAuctionAveragePrice',
'MMRCurrentAuctionCleanPrice', 'MMRAcquisitionAuctionCleanPrice',
'MMRCurrentRetailAveragePrice', 'MMRAcquisitionRetailAveragePrice',
'MMRCurrentRetailCleanPrice', 'MMRAcquisitonRetailCleanPrice']
DATE_COLS = ['PurchDate']
LABEL_COL = 'IsBadBuy'
IDS_COL = 'RefId'
# current time for computing recency feature
NOW = '2010-12-31'
# modeling step:
# model checkpoint for future scoring
MODEL_DIR = 'model'
CHECKPOINT_XGB = 'xgb.pkl'
CHECKPOINT_PREPROCESS = 'preprocess.pkl'
# parameter that only relevant for training stage and not scoring
if args.train:
# number of cross validation and hyperparameter settings to try
CV = 10
N_ITER = 5
MODEL_RANDOM_STATE = 4321
# train/validation stratified split
VAL_SIZE = 0.1
TEST_SIZE = 0.1
SPLIT_RANDOM_STATE = 1234
# -----------------------------------------------------------------------------------
logger.info('preprocessing')
checkpoint_preprocess = os.path.join(MODEL_DIR, CHECKPOINT_PREPROCESS)
checkpoint_xgb = os.path.join(MODEL_DIR, CHECKPOINT_XGB)
input_path = os.path.join(DATA_DIR, args.inputfile)
if args.train:
data = clean(input_path, NOW, CAT_COLS, NUM_COLS, DATE_COLS, IDS_COL, LABEL_COL)
ids = data[IDS_COL].values
label = data[LABEL_COL].values
data = data.drop([IDS_COL, LABEL_COL], axis = 1)
# train/test split twice to achieve train/validation/test three way split
df_train, df_test, y_train, y_test, ids_train, ids_test = train_test_split(
data, label, ids, test_size = TEST_SIZE,
random_state = SPLIT_RANDOM_STATE, stratify = label)
df_train, df_val, y_train, y_val, ids_train, ids_val = train_test_split(
df_train, y_train, ids_train, test_size = VAL_SIZE,
random_state = SPLIT_RANDOM_STATE, stratify = y_train)
# obtain finalized columns
num_cols_cleaned = list(SortedSet(df_train.columns) - SortedSet(CAT_COLS))
preprocess = Preprocesser(num_cols = num_cols_cleaned, cat_cols = CAT_COLS)
X_train = preprocess.fit_transform(df_train)
X_val = preprocess.transform(df_val)
X_test = preprocess.transform(df_test)
logger.info('modeling')
eval_set = [(X_train, y_train), (X_val, y_val)]
xgb_tuned = build_xgb(N_ITER, CV, MODEL_RANDOM_STATE, eval_set)
xgb_tuned.fit(X_train, y_train)
if not os.path.isdir(MODEL_DIR):
os.mkdir(MODEL_DIR)
dump(preprocess, checkpoint_preprocess)
dump(xgb_tuned, checkpoint_xgb)
# model evaluation metric reporting
y_pred = []
xgb_best = xgb_tuned.best_estimator_
zipped = zip(
('train', 'validation', 'test'),
(X_train, X_val, X_test),
(y_train, y_val, y_test))
for name, X, y in zipped:
xgb_pred = xgb_best.predict_proba(
X, ntree_limit = xgb_best.best_ntree_limit)[:, 1]
score = round(roc_auc_score(y, xgb_pred), 3)
logger.info('{} AUC: {}'.format(name, score))
y_pred.append(xgb_pred)
ids = np.hstack((ids_train, ids_val, ids_test))
y_pred = np.hstack(y_pred)
else:
data = clean(input_path, NOW, CAT_COLS, NUM_COLS, DATE_COLS, IDS_COL)
ids = data[IDS_COL].values
data = data.drop(IDS_COL, axis = 1)
logger.info('scoring')
preprocess = load(checkpoint_preprocess)
xgb_tuned = load(checkpoint_xgb)
X = preprocess.transform(data)
xgb_best = xgb_tuned.best_estimator_
y_pred = xgb_best.predict_proba(
X, ntree_limit = xgb_best.best_ntree_limit)[:, 1]
if not os.path.isdir(OUTPUT_DIR):
os.mkdir(OUTPUT_DIR)
output_path = os.path.join(OUTPUT_DIR, args.outputfile)
write_output(ids, IDS_COL, y_pred, LABEL_COL, output_path)