-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrun-nextract-svm-obv.py
executable file
·220 lines (179 loc) · 8.35 KB
/
run-nextract-svm-obv.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
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
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
#!/global/scratch/sness/openmir/tools/python/bin/python
#
#
#
import sys
import os
import datetime
import commands
import re
import time
import simplejson as json
import random
TOOLSDIR = os.getenv('TOOLSDIR', '/global/scratch/sness/openmir/tools')
TMPDIR = os.getenv('TMPDIR', '/scratch/')
nextractPath = os.path.join(TOOLSDIR, "marsyas/release/bin/nextract")
liblinearTrainPath = os.path.join(TOOLSDIR, "liblinear-1.93/train")
liblinearPredictPath = os.path.join(TOOLSDIR, "liblinear-1.93/predict")
libsvmTrainPath = os.path.join(TOOLSDIR, "libsvm-3.17/svm-train")
libsvmPredictPath = os.path.join(TOOLSDIR, "libsvm-3.17/svm-predict")
libsvmScalePath = os.path.join(TOOLSDIR, "libsvm-3.17/svm-scale")
wekaPath = "java -classpath %s" % (os.path.join(TOOLSDIR, "weka/weka.jar"))
#DEBUG = False
DEBUG = True
def parseInput(inFilename):
data = []
with open(inFilename) as f:
for line in f:
item = json.loads(line)
data.append(item)
return data
def generateFilenames(runs):
outputDir = 'output/features'
for run in runs:
run['randHash'] = "%032x" % random.getrandbits(128)
# TODO(sness) - Add parameters to nextract and svm to baseFilename
run['baseFilename'] = os.path.join(TMPDIR,'nextract-svm-%s' % (run['randHash']))
run['extractTrainFilename'] = '%s.train.features' % (run['baseFilename'])
run['extractTestFilename'] = '%s.test.features' % (run['baseFilename'])
run['arffTrainFilename'] = '%s.train.arff' % (run['baseFilename'])
run['arffTestFilename'] = '%s.test.arff' % (run['baseFilename'])
run['scaleTrainFilename'] = '%s.train.scaled' % (run['baseFilename'])
run['scaleTestFilename'] = '%s.test.scaled' % (run['baseFilename'])
run['scaleParamsFilename'] = '%s.params' % (run['baseFilename'])
run['modelFilename'] = '%s.model' % (run['baseFilename'])
run['predictionFilename'] = '%s.prediction' % (run['baseFilename'])
return runs
def runExtract(run,inTrainCollection,inTestCollection):
""" Extract audio features. """
run['extractTrainCommand'] = "%s %s %s -w %s -o %s" % (
nextractPath, run['extractOptions'], inTrainCollection, run['arffTrainFilename'], run['extractTrainFilename'])
startTime = time.time()
run['extractTrainOutput'] = commands.getoutput(run['extractTrainCommand'])
run['extractTrainTime'] = time.time() - startTime
run['extractTestCommand'] = "%s %s %s -w %s -o %s" % (
nextractPath, run['extractOptions'], inTestCollection, run['arffTestFilename'], run['extractTestFilename'])
startTime = time.time()
run['extractTestOutput'] = commands.getoutput(run['extractTestCommand'])
run['extractTestTime'] = time.time() - startTime
if DEBUG:
print "extractTrainCommand=%s" % (run['extractTrainCommand'])
print "extractTrainOutput=%s" % (run['extractTrainOutput'])
print "extractTrainTime=%s" % (run['extractTrainTime'])
print "extractTestCommand=%s" % (run['extractTestCommand'])
print "extractTestOutput=%s" % (run['extractTestOutput'])
print "extractTestTime=%s" % (run['extractTestTime'])
def runScale(run):
""" Scale the data with libsvm/scale. """
if DEBUG:
print "runScale"
if run['scale'] == 'false':
if DEBUG:
print "Not scaling data"
run['scaleTrainFilename'] = run['extractTrainFilename']
run['scaleTestFilename'] = run['extractTestFilename']
return
run['scaleTrainCommand'] = "%s -s %s %s > %s" % (libsvmScalePath, run['scaleParamsFilename'], run['extractTrainFilename'], run['scaleTrainFilename'])
startTime = time.time()
run['scaleTrainOutput'] = commands.getoutput(run['scaleTrainCommand'])
run['scaleTrainTime'] = time.time() - startTime
run['scaleTestCommand'] = "%s -r %s %s > %s" % (libsvmScalePath, run['scaleParamsFilename'], run['extractTestFilename'], run['scaleTestFilename'])
startTime = time.time()
run['scaleTestOutput'] = commands.getoutput(run['scaleTestCommand'])
run['scaleTestTime'] = time.time() - startTime
if DEBUG:
print "scaleTrainCommand=%s" % (run['scaleTrainCommand'])
print "scaleTrainOutput=%s" % (run['scaleTrainOutput'])
print "scaleTrainTime=%s" % (run['scaleTrainTime'])
print "scaleTestCommand=%s" % (run['scaleTestCommand'])
print "scaleTestOutput=%s" % (run['scaleTestOutput'])
print "scaleTestTime=%s" % (run['scaleTestTime'])
def runTrain(run):
""" Train a model with a classifier. """
if DEBUG:
print "runTrain"
# TODO(sness) - Change to allow libsvm or liblinear to be used
if run['svm'] == 'libsvm':
trainPath = libsvmTrainPath
elif run['svm'] == 'weka':
run['trainTime'] = 0.0
return
else:
trainPath = liblinearTrainPath
run['trainCommand'] = "%s %s %s %s" % (trainPath, run['svmOptions'], run['scaleTrainFilename'], run['modelFilename'])
startTime = time.time()
run['trainOutput'] = commands.getoutput(run['trainCommand'])
run['trainTime'] = time.time() - startTime
if DEBUG:
print "trainCommand=%s" % (run['trainCommand'])
print "trainOutput=%s" % (run['trainOutput'])
print "trainTime=%s" % (run['trainTime'])
def runPredict(run):
""" Predict input data with a trained model. """
if DEBUG:
print "runPredict"
weka = False
if run['svm'] == 'libsvm':
predictPath = libsvmPredictPath
elif run['svm'] == 'weka':
predictPath = wekaPath
weka = True
else:
predictPath = liblinearPredictPath
if weka == True:
run['predictCommand'] = "%s %s -t %s -T %s" % (predictPath, run['svmOptions'], run['arffTrainFilename'], run['arffTestFilename'])
else:
run['predictCommand'] = "%s %s %s %s" % (predictPath, run['scaleTestFilename'], run['modelFilename'], run['predictionFilename'])
startTime = time.time()
run['predictOutput'] = commands.getoutput(run['predictCommand'])
run['predictTime'] = time.time() - startTime
if weka == True:
m = re.search('=== Error on test data ===\s+Correctly Classified Instances\s+[0-9]+\s+([0-9.]+)', run['predictOutput'])
if m is not None:
run['predictAccuracy'] = float(m.group(1))
else:
run['predictAccuracy'] = -1.
else:
m = re.search('Accuracy = ([0-9.]+)', run['predictOutput'])
if m is not None:
run['predictAccuracy'] = float(m.group(1))
else:
run['predictAccuracy'] = -1.
if DEBUG:
print "predictCommand=%s" % (run['predictCommand'])
print "predictOutput=%s" % (run['predictOutput'])
print "predictTime=%s" % (run['predictTime'])
def removeTmpFiles(run):
if os.path.exists(run['extractTrainFilename']):
os.remove(run['extractTrainFilename'])
if os.path.exists(run['extractTestFilename']):
os.remove(run['extractTestFilename'])
if os.path.exists(run['scaleTrainFilename']):
os.remove(run['scaleTrainFilename'])
if os.path.exists(run['scaleTestFilename']):
os.remove(run['scaleTestFilename'])
if os.path.exists(run['modelFilename']):
os.remove(run['modelFilename'])
if os.path.exists(run['predictionFilename']):
os.remove(run['predictionFilename'])
def run(runs,inTrainCollection,inTestCollection):
if DEBUG:
print "TOOLSDIR=%s" % (TOOLSDIR)
print "TMPDIR=%s" % (TMPDIR)
for run in runs:
runExtract(run,inTrainCollection,inTestCollection)
runScale(run)
runTrain(run)
runPredict(run)
print "|obv-%s|%s|%s|%.2f|%.2f|%.2f|%s|" % (run['table'], run['extractOptions'], run['svmOptions'], run['extractTrainTime'] + run['extractTestTime'], run['trainTime'], run['predictTime'], run['predictAccuracy'])
#removeTmpFiles(run)
if __name__ == "__main__":
if len(sys.argv) < 3:
print "Usage: run-nextract-svm-train-test.py input/runsvm.test.json input/train.mf input/test.mf"
sys.exit(1)
inCommandFilename = sys.argv[1]
inTrainCollection = sys.argv[2]
inTestCollection = sys.argv[3]
runs = parseInput(inCommandFilename)
runs = generateFilenames(runs)
run(runs,inTrainCollection,inTestCollection)