def test_overlap(self):
negatives = [-8,-5,-3,2,21]
positives = [-1,6,12,15,20]
result = Separator.separate(positives,negatives)
#threshold needs to be 2<=t<6
self.assertLess(result,6)
self.assertGreaterEqual(result,2)
def test_overlap(self):
negatives = [-8, -5, -3, 2, 21]
positives = [-1, 6, 12, 15, 20]
result = Separator.separate(positives, negatives)
#threshold needs to be 2<=t<6
self.assertLess(result, 6)
self.assertGreaterEqual(result, 2)
def trainFreqThresh1(self, split):
print "Training split " + str(split)
positives = []
negatives = []
done = 0
for [word1, word2, result] in self.pairmatrix[self.cv_idx != split]:
#if word1 in self.entrydict.keys():
# if word2 in self.entrydict.keys():
if EntailClassifier.do_ratio:
diff = float(self.entrydict[word2].freq) / float(
self.entrydict[word1].freq)
else:
diff = float(self.entrydict[word2].freq) - float(
self.entrydict[word1].freq)
#else:
# print "Error: no frequency information for "+word2
#else:
#print "Error: no frequency information for "+word1
if int(result) == 1:
positives.append(diff)
else:
negatives.append(diff)
done += 1
if self.verbose and done % 1000 == 0:
print "Trained on " + str(done)
print len(positives), len(negatives)
threshold = Separator.separate(positives, negatives, trials=1000000)
return [threshold]
def trainFreqThresh1(self,split):
print"Training split "+str(split)
positives=[]
negatives=[]
done=0
for [word1,word2,result] in self.pairmatrix[self.cv_idx!=split]:
#if word1 in self.entrydict.keys():
# if word2 in self.entrydict.keys():
if EntailClassifier.do_ratio:
diff = float(self.entrydict[word2].freq)/float(self.entrydict[word1].freq)
else:
diff = float(self.entrydict[word2].freq)-float(self.entrydict[word1].freq)
#else:
# print "Error: no frequency information for "+word2
#else:
#print "Error: no frequency information for "+word1
if int(result)==1:
positives.append(diff)
else:
negatives.append(diff)
done+=1
if self.verbose and done%1000==0:
print "Trained on "+str(done)
print len(positives),len(negatives)
threshold = Separator.separate(positives,negatives,trials=1000000)
return [threshold]
def trainCRThresh(self, split, method):
print "Training split " + str(split)
positives = []
negatives = []
done = 0
for [word1, word2, result] in self.pairmatrix[self.cv_idx != split]:
#diff = float(self.entrydict[word2].freq)-float(self.entrydict[word1].freq)
if method == "CR_thresh":
precision = float(self.entrydict[word1].precision(
self.entrydict[word2]))
recall = float(self.entrydict[word2].precision(
self.entrydict[word1]))
elif method == "clarke_thresh":
precision = float(self.entrydict[word1].min_precision(
self.entrydict[word2]))
recall = float(self.entrydict[word2].min_precision(
self.entrydict[word1]))
elif method == "invCL":
precision = float(self.entrydict[word1].invCL(
self.entrydict[word2]))
recall = 1 #invCL uses 1-recall anyway so threshold just on precision value
else:
print "Unknown CR method " + method
exit(1)
if recall == 0:
ratio = 0
#hm=0
else:
if EntailClassifier.do_ratio:
ratio = precision / recall
else:
ratio = precision - recall
#hm=2*precision*recall/(precision+recall)
if int(result) == 1:
positives.append(ratio)
else:
negatives.append(ratio)
done += 1
if self.verbose and done % 1000 == 0:
print "Trained on " + str(done)
print len(positives), len(negatives)
threshold = Separator.separate(positives,
negatives,
trials=1000000,
integer=False)
return [threshold]
def trainCRThresh(self,split,method):
print"Training split "+str(split)
positives=[]
negatives=[]
done=0
for [word1,word2,result] in self.pairmatrix[self.cv_idx!=split]:
#diff = float(self.entrydict[word2].freq)-float(self.entrydict[word1].freq)
if method=="CR_thresh":
precision = float(self.entrydict[word1].precision(self.entrydict[word2]))
recall=float(self.entrydict[word2].precision(self.entrydict[word1]))
elif method=="clarke_thresh":
precision = float(self.entrydict[word1].min_precision(self.entrydict[word2]))
recall=float(self.entrydict[word2].min_precision(self.entrydict[word1]))
elif method=="invCL":
precision = float(self.entrydict[word1].invCL(self.entrydict[word2]))
recall=1 #invCL uses 1-recall anyway so threshold just on precision value
else:
print "Unknown CR method "+method
exit(1)
if recall == 0:
ratio=0
#hm=0
else:
if EntailClassifier.do_ratio:
ratio=precision/recall
else:
ratio=precision-recall
#hm=2*precision*recall/(precision+recall)
if int(result)==1:
positives.append(ratio)
else:
negatives.append(ratio)
done+=1
if self.verbose and done%1000==0:
print "Trained on "+str(done)
print len(positives),len(negatives)
threshold = Separator.separate(positives,negatives,trials=1000000,integer=False)
return [threshold]
def test_separated_lists(self):
negatives = [-4, 2, 3]
positives = [5, 10, 12]
result = Separator.separate(positives, negatives)
self.assertEqual(result, 4)
def test_separated_lists(self):
negatives=[-4,2,3]
positives=[5,10,12]
result = Separator.separate(positives,negatives)
self.assertEqual(result,4)
