-
Notifications
You must be signed in to change notification settings - Fork 3
/
fit_lasso.py
137 lines (122 loc) · 3.82 KB
/
fit_lasso.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
import sys
import os
import random
import argparse
import logging
import pickle
import subprocess
import pandas as pd
from scipy.stats import pearsonr
import numpy as np
from common import *
from sklearn.model_selection import GridSearchCV
from sklearn.linear_model import Lasso
from sklearn.linear_model import LogisticRegression
def parse_args():
''' parse command line arguments '''
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('--seed',
type=int,
default=1)
parser.add_argument('--kfold',
type=int,
default=3)
parser.add_argument('--data-index-file',
type=str,
default=None)
parser.add_argument('--data-X-file',
type=str)
parser.add_argument('--data-y-file',
type=str)
parser.add_argument('--data-file',
type=str,
default="_output/data.pkl")
parser.add_argument('--log-file',
type=str,
default="_output/lasso.log")
parser.add_argument('--out-file',
type=str,
default="_output/out.csv")
parser.add_argument('--data-classes',
type=int,
default=0)
parser.add_argument('--scratch',
type=str,
default="_output/scratch")
args = parser.parse_args()
return args
def fit_lasso(
args,
train_file,
test_file):
cmd = [
'Rscript',
'../R/fit_lasso.R',
train_file,
test_file,
args.out_file,
args.data_classes,
args.seed,
args.kfold,
]
cmd = list(map(str, cmd))
print("Calling:", " ".join(cmd))
res = subprocess.call(cmd)
# Check that process complete successfully
assert res == 0
def main(args=sys.argv[1:]):
args = parse_args()
logging.basicConfig(format="%(message)s", filename=args.log_file, level=logging.DEBUG)
#randnum = random.randint(1, 100000)
#train_file_name = os.path.join(args.scratch, "train%d" % randnum)
#test_file_name = os.path.join(args.scratch, "test%d" % randnum)
#write_data_pkl_to_csv(
# None,
# train_file_name,
# test_file_name)
#fit_lasso(args, train_file_name, test_file_name)
#
# os.remove(train_file_name)
# os.remove(test_file_name)
train_data, test_data = read_input_data(args)
if args.data_classes == 0:
lasso = Lasso(
random_state=args.seed, max_iter=10000)
alphas = np.power(10., np.arange(2, -4, -0.1))
tuned_parameters = [{'alpha': alphas}]
elif args.data_classes == 1:
lasso = LogisticRegression(
random_state=args.seed,
max_iter=10000,
penalty='l1')
Cs = np.power(10., np.arange(4, -3, -0.1))
tuned_parameters = [{'C': Cs}]
else:
raise ValueError("huh?")
clf = GridSearchCV(
lasso,
tuned_parameters,
cv=args.kfold,
refit=True,
n_jobs=10)
clf.fit(train_data.x, train_data.y.ravel())
logging.info(clf.cv_results_["mean_test_score"])
logging.info(clf.cv_results_["params"])
logging.info(clf.best_params_)
y_pred = clf.predict(test_data.x)
num_nonzero_inputs = np.sum(np.abs(clf.best_estimator_.coef_) > THRES)
if args.data_classes == 0:
test_loss = get_regress_loss(y_pred, test_data.y_true)
logging.info("pearsonr %s", pearsonr(y_pred.ravel(), test_data.y_true.ravel()))
logistic_loss = 0
else:
test_loss = 1 - get_classification_accuracy(y_pred, test_data.y_true)
logistic_loss = get_logistic_loss(y_pred, test_data.y_true)
result = pd.DataFrame({
"test_loss": [test_loss],
"logistic_loss": [logistic_loss],
"num_nonzero": [num_nonzero_inputs]})
result.T.to_csv(args.out_file)
logging.info("DONE!")
if __name__ == "__main__":
main(sys.argv[1:])