def stackTest(fModels, K, testFile, DIR):
"""
Implementation of algorithm 21. Takes in k classifiers( fModel ), and fin are strings (filename), k is the level of stack
"""
fIn = map(lambda k:DIR+"test%d.megam"%k, range(K))
fOut = map(lambda k: DIR+"Y_test%d.megam"%k, range(K))
data = stackFeatures.initFeatures(citeFile, testFile, fIn[0])
err = [0]*K
for k in range(0,K):
if k!=0: stackFeatures.stackFeatures(data, fIn[k-1], fOut[k-1], fIn[k], k)
# test
print "********** test @ stack %d **********"%k
err[k] = predict(fModels[k],fIn[k],fOut[k])
return fOut, err
def stackTrain(K, contentFile, DIR):
"""
implementation stacking algorithm. Takes in k which is the number of stacks
"""
fIn = map(lambda k: DIR+"train%d.megam"%k, range(K))
fOut = map(lambda k: DIR+"Y_train%d.megam"%k, range(K))
err = [0]*K
classifiers = map(lambda k: DIR+"model%d.megam"%k, range(K)) # name of the classifiers
for k in range(0,K):
if k == 0: data = stackFeatures.initFeatures(citeFile, contentFile, fIn[k])
else: stackFeatures.stackFeatures(data, fIn[k-1],fOut[k-1],fIn[k],k)
# train
print "********** train stack %d **********"%k
trainMegam(fIn[k], classifiers[k])
# test, get \hat Y[k], here don't do stackTest but reuse
err[k] = predict(classifiers[k], fIn[k], fOut[k])
return (classifiers,err)