def includepath(self,feat):
if len(SimEngine.paths_to_include)>0:
return getpathtype(feat) in SimEngine.paths_to_include
elif len(SimEngine.blacklist)>0:
return getpathtype(feat) not in SimEngine.blacklist
else:
return True
def computepathtotals(self):
self.pathtotals={}
for feature in self.features.keys():
pathtype=getpathtype(feature)
sofar=self.pathtotals.get(pathtype,0.0)
self.pathtotals[pathtype]=sofar+float(self.features[feature])
def showsuffix(self, path, minorder=1, maxorder=1):
feats = {}
for feat in self.features.keys():
thispath = getpathtype(feat)
order = getorder(feat)
if thispath.endswith(path) and order >= minorder and order <= maxorder:
feats[feat] = self.features[feat]
values = sorted(feats.items(), key=itemgetter(1), reverse=True)
print values
def compute_typetotals(self,type,cds=False):
#compute totals for different paths over all entries (using column totals given in feattots)
if not self.coltots_loaded[type]:
self.load_coltotals(type,cds)
print "Computing path totals C<*,t,*>"
self.typetots={}
for feature in self.feattots.keys():
pathtype=getpathtype(feature)
sofar=self.typetots.get(pathtype,0.0)
self.typetots[pathtype]=sofar+float(self.feattots[feature])
def reweight(self,weighting,feattots,typetots,grandtot=0,ppmithreshold=0,saliency=0):
self.featureweights={}
self.lgth=-1
self.wdth=-1
for feature in self.features.keys():
freq=float(self.features[feature]) # C<w1,p,w2>
try:
total=float(self.pathtotals[getpathtype(feature)]) # C<w1,p,*>
except:
total=0.0001
print "Warning: no path total for %s: %s"%(feature,getpathtype(feature))
feattot=float(feattots[feature]) #C<*,p,w2>
typetot=float(typetots[getpathtype(feature)]) #C<*,p,*>
entrytotal=float(self.total) # C<w1,*,*>
if "ttest" in weighting:
expected = (total*feattot)/(typetot*typetot) #incorrect! this should be the type total for the entry not the total
obs=freq/typetot
score= (obs-expected)/math.pow(expected,0.5)
if score>ppmithreshold:
self.featureweights[feature]=score
else:
try:
if "gof_ppmi" in weighting:
pmi=math.log10((freq*grandtot)/(feattot*entrytotal))
else:
pmi=math.log10((freq*typetot)/(feattot*total))
except:
pmi=0
shifted_pmi=pmi-ppmithreshold
if shifted_pmi>0:
if "pnppmi" in weighting:
shifted_pmi=shifted_pmi * total/entrytotal
if "plmi" in weighting:
shifted_pmi=shifted_pmi * freq/typetot
self.featureweights[feature]=shifted_pmi
self.reducesaliency(saliency)
def reducesaliency(self,saliency,saliencyperpath=False):
if saliency==0:
return
else:
#print "Carrying out saliency reduction / context selection to top",str(saliency)
feats=sorted(self.featureweights.items(),key=itemgetter(1),reverse=True)
self.featureweights={}
donetypes={}
all=0
for tuple in feats:
feature=tuple[0]
pathtype=getpathtype(feature)
done=donetypes.get(pathtype,0)
if (saliencyperpath and done<saliency)or(not saliencyperpath and all<saliency):
self.featureweights[feature]=tuple[1]
donetypes[pathtype]=done+1
all+=1
def profile(self, minorder=0, maxorder=10):
paths = {}
totalweight = 0
thisorderweight = 0
for feat in self.features.keys():
path = getpathtype(feat)
order = getorder(feat)
weight = self.features[feat]
sofar = paths.get(path, 0)
if order >= minorder and order <= maxorder:
paths[path] = sofar + weight
thisorderweight += weight
totalweight += weight
print "total weight of features", totalweight
print "total weight of required order features", thisorderweight
profile = sorted(paths.items(), key=itemgetter(1), reverse=True)
print profile
