forked from scikit-learn/scikit-learn
-
Notifications
You must be signed in to change notification settings - Fork 0
/
bench_sgd_regression.py
152 lines (129 loc) · 5.21 KB
/
bench_sgd_regression.py
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
# Author: Peter Prettenhofer <[email protected]>
# License: BSD 3 clause
import numpy as np
import matplotlib.pyplot as plt
import gc
from time import time
from sklearn.linear_model import Ridge, SGDRegressor, ElasticNet
from sklearn.metrics import mean_squared_error
from sklearn.datasets import make_regression
"""
Benchmark for SGD regression
Compares SGD regression against coordinate descent and Ridge
on synthetic data.
"""
print(__doc__)
if __name__ == "__main__":
list_n_samples = np.linspace(100, 10000, 5).astype(int)
list_n_features = [10, 100, 1000]
n_test = 1000
max_iter = 1000
noise = 0.1
alpha = 0.01
sgd_results = np.zeros((len(list_n_samples), len(list_n_features), 2))
elnet_results = np.zeros((len(list_n_samples), len(list_n_features), 2))
ridge_results = np.zeros((len(list_n_samples), len(list_n_features), 2))
asgd_results = np.zeros((len(list_n_samples), len(list_n_features), 2))
for i, n_train in enumerate(list_n_samples):
for j, n_features in enumerate(list_n_features):
X, y, coef = make_regression(
n_samples=n_train + n_test,
n_features=n_features,
noise=noise,
coef=True,
)
X_train = X[:n_train]
y_train = y[:n_train]
X_test = X[n_train:]
y_test = y[n_train:]
print("=======================")
print("Round %d %d" % (i, j))
print("n_features:", n_features)
print("n_samples:", n_train)
# Shuffle data
idx = np.arange(n_train)
np.random.seed(13)
np.random.shuffle(idx)
X_train = X_train[idx]
y_train = y_train[idx]
std = X_train.std(axis=0)
mean = X_train.mean(axis=0)
X_train = (X_train - mean) / std
X_test = (X_test - mean) / std
std = y_train.std(axis=0)
mean = y_train.mean(axis=0)
y_train = (y_train - mean) / std
y_test = (y_test - mean) / std
gc.collect()
print("- benchmarking ElasticNet")
clf = ElasticNet(alpha=alpha, l1_ratio=0.5, fit_intercept=False)
tstart = time()
clf.fit(X_train, y_train)
elnet_results[i, j, 0] = mean_squared_error(clf.predict(X_test), y_test)
elnet_results[i, j, 1] = time() - tstart
gc.collect()
print("- benchmarking SGD")
clf = SGDRegressor(
alpha=alpha / n_train,
fit_intercept=False,
max_iter=max_iter,
learning_rate="invscaling",
eta0=0.01,
power_t=0.25,
tol=1e-3,
)
tstart = time()
clf.fit(X_train, y_train)
sgd_results[i, j, 0] = mean_squared_error(clf.predict(X_test), y_test)
sgd_results[i, j, 1] = time() - tstart
gc.collect()
print("max_iter", max_iter)
print("- benchmarking A-SGD")
clf = SGDRegressor(
alpha=alpha / n_train,
fit_intercept=False,
max_iter=max_iter,
learning_rate="invscaling",
eta0=0.002,
power_t=0.05,
tol=1e-3,
average=(max_iter * n_train // 2),
)
tstart = time()
clf.fit(X_train, y_train)
asgd_results[i, j, 0] = mean_squared_error(clf.predict(X_test), y_test)
asgd_results[i, j, 1] = time() - tstart
gc.collect()
print("- benchmarking RidgeRegression")
clf = Ridge(alpha=alpha, fit_intercept=False)
tstart = time()
clf.fit(X_train, y_train)
ridge_results[i, j, 0] = mean_squared_error(clf.predict(X_test), y_test)
ridge_results[i, j, 1] = time() - tstart
# Plot results
i = 0
m = len(list_n_features)
plt.figure("scikit-learn SGD regression benchmark results", figsize=(5 * 2, 4 * m))
for j in range(m):
plt.subplot(m, 2, i + 1)
plt.plot(list_n_samples, np.sqrt(elnet_results[:, j, 0]), label="ElasticNet")
plt.plot(list_n_samples, np.sqrt(sgd_results[:, j, 0]), label="SGDRegressor")
plt.plot(list_n_samples, np.sqrt(asgd_results[:, j, 0]), label="A-SGDRegressor")
plt.plot(list_n_samples, np.sqrt(ridge_results[:, j, 0]), label="Ridge")
plt.legend(prop={"size": 10})
plt.xlabel("n_train")
plt.ylabel("RMSE")
plt.title("Test error - %d features" % list_n_features[j])
i += 1
plt.subplot(m, 2, i + 1)
plt.plot(list_n_samples, np.sqrt(elnet_results[:, j, 1]), label="ElasticNet")
plt.plot(list_n_samples, np.sqrt(sgd_results[:, j, 1]), label="SGDRegressor")
plt.plot(list_n_samples, np.sqrt(asgd_results[:, j, 1]), label="A-SGDRegressor")
plt.plot(list_n_samples, np.sqrt(ridge_results[:, j, 1]), label="Ridge")
plt.legend(prop={"size": 10})
plt.xlabel("n_train")
plt.ylabel("Time [sec]")
plt.title("Training time - %d features" % list_n_features[j])
i += 1
plt.subplots_adjust(hspace=0.30)
plt.show()