1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
|
"""
Demo for defining a custom regression objective and metric
==========================================================
Demo for defining customized metric and objective. Notice that for simplicity reason
weight is not used in following example. In this script, we implement the Squared Log
Error (SLE) objective and RMSLE metric as customized functions, then compare it with
native implementation in XGBoost.
See :doc:`/tutorials/custom_metric_obj` for a step by step walkthrough, with other
details.
The `SLE` objective reduces impact of outliers in training dataset, hence here we also
compare its performance with standard squared error.
"""
import argparse
from time import time
from typing import Dict, List, Tuple
import matplotlib
import numpy as np
from matplotlib import pyplot as plt
import xgboost as xgb
# shape of generated data.
kRows = 4096
kCols = 16
kOutlier = 10000 # mean of generated outliers
kNumberOfOutliers = 64
kRatio = 0.7
kSeed = 1994
kBoostRound = 20
np.random.seed(seed=kSeed)
def generate_data() -> Tuple[xgb.DMatrix, xgb.DMatrix]:
'''Generate data containing outliers.'''
x = np.random.randn(kRows, kCols)
y = np.random.randn(kRows)
y += np.abs(np.min(y))
# Create outliers
for i in range(0, kNumberOfOutliers):
ind = np.random.randint(0, len(y)-1)
y[ind] += np.random.randint(0, kOutlier)
train_portion = int(kRows * kRatio)
# rmsle requires all label be greater than -1.
assert np.all(y > -1.0)
train_x: np.ndarray = x[: train_portion]
train_y: np.ndarray = y[: train_portion]
dtrain = xgb.DMatrix(train_x, label=train_y)
test_x = x[train_portion:]
test_y = y[train_portion:]
dtest = xgb.DMatrix(test_x, label=test_y)
return dtrain, dtest
def native_rmse(dtrain: xgb.DMatrix,
dtest: xgb.DMatrix) -> Dict[str, Dict[str, List[float]]]:
'''Train using native implementation of Root Mean Squared Loss.'''
print('Squared Error')
squared_error = {
'objective': 'reg:squarederror',
'eval_metric': 'rmse',
'tree_method': 'hist',
'seed': kSeed
}
start = time()
results: Dict[str, Dict[str, List[float]]] = {}
xgb.train(squared_error,
dtrain=dtrain,
num_boost_round=kBoostRound,
evals=[(dtrain, 'dtrain'), (dtest, 'dtest')],
evals_result=results)
print('Finished Squared Error in:', time() - start, '\n')
return results
def native_rmsle(dtrain: xgb.DMatrix,
dtest: xgb.DMatrix) -> Dict[str, Dict[str, List[float]]]:
'''Train using native implementation of Squared Log Error.'''
print('Squared Log Error')
results: Dict[str, Dict[str, List[float]]] = {}
squared_log_error = {
'objective': 'reg:squaredlogerror',
'eval_metric': 'rmsle',
'tree_method': 'hist',
'seed': kSeed
}
start = time()
xgb.train(squared_log_error,
dtrain=dtrain,
num_boost_round=kBoostRound,
evals=[(dtrain, 'dtrain'), (dtest, 'dtest')],
evals_result=results)
print('Finished Squared Log Error in:', time() - start)
return results
def py_rmsle(dtrain: xgb.DMatrix, dtest: xgb.DMatrix) -> Dict:
'''Train using Python implementation of Squared Log Error.'''
def gradient(predt: np.ndarray, dtrain: xgb.DMatrix) -> np.ndarray:
'''Compute the gradient squared log error.'''
y = dtrain.get_label()
return (np.log1p(predt) - np.log1p(y)) / (predt + 1)
def hessian(predt: np.ndarray, dtrain: xgb.DMatrix) -> np.ndarray:
'''Compute the hessian for squared log error.'''
y = dtrain.get_label()
return ((-np.log1p(predt) + np.log1p(y) + 1) /
np.power(predt + 1, 2))
def squared_log(predt: np.ndarray,
dtrain: xgb.DMatrix) -> Tuple[np.ndarray, np.ndarray]:
'''Squared Log Error objective. A simplified version for RMSLE used as
objective function.
:math:`\frac{1}{2}[log(pred + 1) - log(label + 1)]^2`
'''
predt[predt < -1] = -1 + 1e-6
grad = gradient(predt, dtrain)
hess = hessian(predt, dtrain)
return grad, hess
def rmsle(predt: np.ndarray, dtrain: xgb.DMatrix) -> Tuple[str, float]:
''' Root mean squared log error metric.
:math:`\sqrt{\frac{1}{N}[log(pred + 1) - log(label + 1)]^2}`
'''
y = dtrain.get_label()
predt[predt < -1] = -1 + 1e-6
elements = np.power(np.log1p(y) - np.log1p(predt), 2)
return 'PyRMSLE', float(np.sqrt(np.sum(elements) / len(y)))
results: Dict[str, Dict[str, List[float]]] = {}
xgb.train({'tree_method': 'hist', 'seed': kSeed,
'disable_default_eval_metric': 1},
dtrain=dtrain,
num_boost_round=kBoostRound,
obj=squared_log,
custom_metric=rmsle,
evals=[(dtrain, 'dtrain'), (dtest, 'dtest')],
evals_result=results)
return results
def plot_history(rmse_evals, rmsle_evals, py_rmsle_evals):
fig, axs = plt.subplots(3, 1)
ax0: matplotlib.axes.Axes = axs[0]
ax1: matplotlib.axes.Axes = axs[1]
ax2: matplotlib.axes.Axes = axs[2]
x = np.arange(0, kBoostRound, 1)
ax0.plot(x, rmse_evals['dtrain']['rmse'], label='train-RMSE')
ax0.plot(x, rmse_evals['dtest']['rmse'], label='test-RMSE')
ax0.legend()
ax1.plot(x, rmsle_evals['dtrain']['rmsle'], label='train-native-RMSLE')
ax1.plot(x, rmsle_evals['dtest']['rmsle'], label='test-native-RMSLE')
ax1.legend()
ax2.plot(x, py_rmsle_evals['dtrain']['PyRMSLE'], label='train-PyRMSLE')
ax2.plot(x, py_rmsle_evals['dtest']['PyRMSLE'], label='test-PyRMSLE')
ax2.legend()
def main(args):
dtrain, dtest = generate_data()
rmse_evals = native_rmse(dtrain, dtest)
rmsle_evals = native_rmsle(dtrain, dtest)
py_rmsle_evals = py_rmsle(dtrain, dtest)
if args.plot != 0:
plot_history(rmse_evals, rmsle_evals, py_rmsle_evals)
plt.show()
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description='Arguments for custom RMSLE objective function demo.')
parser.add_argument(
'--plot',
type=int,
default=1,
help='Set to 0 to disable plotting the evaluation history.')
args = parser.parse_args()
main(args)
|
