def readandfixRound1(taglength=2):
sentences1, pretagged1 = readconll()
sentences2, pretagged2 = tb.readtb()
tagset1, contexts1 = tagsandstuff(pretagged1)
tagset2, contexts2 = tagsandstuff(pretagged2)
tags = alltags([tagset1, tagset2])
pretagged = pretagged1 + pretagged2
names = reallynames(tags)
fixnames(pretagged, names)
fixdates(pretagged, tags)
fixlower(pretagged)
return pretagged, [contexts1, contexts2], tags, [tagset1, tagset2], names
def comp34411(metering=0, brilltagger=False, sentences=False, pretagged=False, qwindow=2, stackwindow=False, folds=5, tagsize=2, atagsize=False, forced=False, precision=0.97, threshold=500):
print "reading data"
tb.METERING = metering
if not atagsize:
atagsize = tagsize
if stackwindow == False:
stackwindow = qwindow
if not sentences:
sentences, pretagged = tb.readtb()
N = 10000
K = 10000
print "initial values for pretagged[:5], sentences[0]"
print pretagged[:5]
sentences[0].showDTree()
tbl.METERING = metering
tbl.tag.METERING = metering
if not brilltagger:
print "making tagger"
t0 = datetime.datetime.now()
brilltagger = tbl.trainwithmxlandbrill(pretagged=pretagged, N=N, K=K, mxltagsize=tagsize, tagsize=tagsize, atagsize=atagsize)
print "training the tagger took %s seconds"%((datetime.datetime.now()-t0).seconds)
print "pretagged[:5], sentences[0] after training the tagger"
print pretagged[:5]
sentences[0].showDTree()
fp.METERING = metering
if not forced:
print "forceparse"
t0 = datetime.datetime.now()
fp.forceparseall(sentences, stackwindow=stackwindow, qwindow=qwindow,tagsize=tagsize)
print "pretagged[:5], sentences[0] after forceparseall"
print "forced parsing took %s seconds"%((datetime.datetime.now()-t0).seconds)
print "nfold %s (precision = %s)"%(folds, precision)
a, ca, c, parser, training, testing = fp.nfold(sentences, n=folds, stackwindow=stackwindow, qwindow=qwindow, tagsize=tagsize, precision=precision, threshold=threshold)
print "average accuracy over all folds (precision = %s): %.3f\naverage classifier accuracy %.3f"%(precision, float(a), float(ca))
print """
return the parser from the last fold & the training and testing sets
for that parser so we can do subsequent experiments soundly
"""
return sentences, pretagged, brilltagger, parser, training, testing
