def analyze(pfile_pk, filename, agent_pk, model, db):
pfile = -1
try:
pfile = int(pfile_pk)
except ValueError:
print >> sys.stdout, 'ERROR: Provided pfile_pk is not a number: %r' % line
return -1
path = libfosspython.repMkPath('files', filename)
if (not os.path.exists(path)):
print >> sys.stdout, 'ERROR: File not found. path=%s' % (path)
return -1
text = open(path).read(READMAX)
offsets = library.label_file(text,model)
if len(offsets) == 0:
sql = """INSERT INTO copyright (agent_fk, pfile_fk, copy_startbyte, copy_endbyte, content, hash, type)
VALUES (%d, %d, NULL, NULL, NULL, NULL, 'statement')""" % (agent_pk, pfile)
result = db.access(sql)
if result != 0:
print >> sys.stdout, "ERROR: DB Access error, returned %d.\nERROR: DB STATUS: %s\nERROR: DB ERRMSG: %s\nERROR: sql=%sERROR: filename=%s" % (result, db.status(), db.errmsg(), sql, filename)
return -1
else:
for i in range(len(offsets)):
str = text[offsets[i][0]:offsets[i][1]]
str = str.decode('ascii', 'ignore') # make sure that it is using ascii encoding
pd = library.parsetext(str)
str = re.escape(' '.join([token[1] for token in pd]))
sql = """INSERT INTO copyright (agent_fk, pfile_fk, copy_startbyte, copy_endbyte, content, hash, type)
VALUES (%d, %d, %d, %d, E'%s', E'%s', '%s')""" % (agent_pk, pfile, offsets[i][0], offsets[i][1],
str, hex(abs(hash(str))),
offsets[i][2])
result = db.access(sql)
if result != 0:
print >> sys.stdout, "ERROR: (nonfatal) DB Access error, returned %d.\nERROR: DB STATUS: %s\nERROR: DB ERRMSG: %s\nERROR: sql=%s\nERROR: filename=%s" % (result, db.status(), db.errmsg(), sql, filename)
return 0
def crossvalidation(data_file, folds=10):
"""
Performs cross validation on set of data and returns the results.
This is a randomized algorithms so you should NOT use if to
regression testing.
data_file should be a file with each line containing a triple quoted
canonical string representation, i.e. call repr on the string.
The text to be hi-lighted should be wrapped in <s>...</s> tags.
Outputs a dictionary that holds specific statistics about the performance
of the classifier.
"""
if folds <= 1:
raise Exception("Number of folds is too small. A value greater than 1 is required.")
training_data = [eval(line) for line in open(data_file).readlines()]
N = len(training_data)
if N < folds:
raise Exception("Number of folds is greater than number of data points.")
# The approximate number of data items in each fold
n = int(round(N/float(folds)))
# shuffle the training data so we dont have any funky correlation issues
# with its ordering.
random.shuffle(training_data)
parsed_data = [library.parsetext(text) for text in training_data]
tokens = [[parsed_data[i][j][0] for j in xrange(len(parsed_data[i]))] for i in xrange(N)]
bio_data = [library.tokens_to_BIO(tokens[i]) for i in xrange(N)]
fold_index = []
for i in range(folds):
fold_index.append(range(i*n,min([n+n*i,N])))
accuracy = 0.0
classes = ['B', 'I', 'O']
matrix = {'B':{'B':0.0, 'I':0.0, 'O':0.0}, 'I':{'B':0.0, 'I':0.0, 'O':0.0}, 'O':{'B':0.0, 'I':0.0, 'O':0.0}}
for i in range(folds):
print "Fold %d." % i
testing = fold_index[i]
training = list(set(testing).symmetric_difference(set(range(N))))
testing_data = [bio_data[d] for d in testing]
training_data = [bio_data[d] for d in training]
PFC = library.tuned_model(training_data)
passed = 0
for test in testing_data:
tokens = test[0]
labels = test[1]
out = library.label_nb(PFC, tokens)
if out == labels:
passed += 1
for l in range(len(labels)):
matrix[labels[l]][out[l]] += 1.0
accuracy += passed/float(len(testing))
raw_accuracy = accuracy/float(folds)
total = sum([sum([matrix[c1][c2] for c2 in classes]) for c1 in classes])
for c1 in classes:
s = sum([matrix[c1][c2] for c2 in classes])
for c2 in matrix[c1]:
matrix[c1][c2] = matrix[c1][c2]/s
print "Raw Accuracy: %1.2f" % raw_accuracy
print "------------------------"
print " | B | I | O "
print "------------------------"
print " B | %1.2f | %1.2f | %1.2f " % (matrix['B']['B'], matrix['B']['I'], matrix['B']['O'])
print "------------------------"
print " I | %1.2f | %1.2f | %1.2f " % (matrix['I']['B'], matrix['I']['I'], matrix['I']['O'])
print "------------------------"
print " O | %1.2f | %1.2f | %1.2f " % (matrix['O']['B'], matrix['O']['I'], matrix['O']['O'])
print "------------------------"
return {'raw accuracy':raw_accuracy, 'matrix':matrix}
def crossvalidation(data_file, folds=10):
"""
Performs cross validation on set of data and returns the results.
This is a randomized algorithms so you should NOT use if to
regression testing.
data_file should be a file with each line containing a triple quoted
canonical string representation, i.e. call repr on the string.
The text to be hi-lighted should be wrapped in <s>...</s> tags.
Outputs a dictionary that holds specific statistics about the performance
of the classifier.
"""
if folds <= 1:
raise Exception(
"Number of folds is too small. A value greater than 1 is required."
)
training_data = [eval(line) for line in open(data_file).readlines()]
N = len(training_data)
if N < folds:
raise Exception(
"Number of folds is greater than number of data points.")
# The approximate number of data items in each fold
n = int(round(N / float(folds)))
# shuffle the training data so we dont have any funky correlation issues
# with its ordering.
random.shuffle(training_data)
parsed_data = [library.parsetext(text) for text in training_data]
tokens = [[parsed_data[i][j][0] for j in xrange(len(parsed_data[i]))]
for i in xrange(N)]
bio_data = [library.tokens_to_BIO(tokens[i]) for i in xrange(N)]
fold_index = []
for i in range(folds):
fold_index.append(range(i * n, min([n + n * i, N])))
accuracy = 0.0
classes = ['B', 'I', 'O']
matrix = {
'B': {
'B': 0.0,
'I': 0.0,
'O': 0.0
},
'I': {
'B': 0.0,
'I': 0.0,
'O': 0.0
},
'O': {
'B': 0.0,
'I': 0.0,
'O': 0.0
}
}
for i in range(folds):
print "Fold %d." % i
testing = fold_index[i]
training = list(set(testing).symmetric_difference(set(range(N))))
testing_data = [bio_data[d] for d in testing]
training_data = [bio_data[d] for d in training]
PFC = library.tuned_model(training_data)
passed = 0
for test in testing_data:
tokens = test[0]
labels = test[1]
out = library.label_nb(PFC, tokens)
if out == labels:
passed += 1
for l in range(len(labels)):
matrix[labels[l]][out[l]] += 1.0
accuracy += passed / float(len(testing))
raw_accuracy = accuracy / float(folds)
total = sum([sum([matrix[c1][c2] for c2 in classes]) for c1 in classes])
for c1 in classes:
s = sum([matrix[c1][c2] for c2 in classes])
for c2 in matrix[c1]:
matrix[c1][c2] = matrix[c1][c2] / s
print "Raw Accuracy: %1.2f" % raw_accuracy
print "------------------------"
print " | B | I | O "
print "------------------------"
print " B | %1.2f | %1.2f | %1.2f " % (matrix['B']['B'], matrix['B']['I'],
matrix['B']['O'])
print "------------------------"
print " I | %1.2f | %1.2f | %1.2f " % (matrix['I']['B'], matrix['I']['I'],
matrix['I']['O'])
print "------------------------"
print " O | %1.2f | %1.2f | %1.2f " % (matrix['O']['B'], matrix['O']['I'],
matrix['O']['O'])
print "------------------------"
return {'raw accuracy': raw_accuracy, 'matrix': matrix}