def train_randomized_forest(training_data):
print "Training Random Forest"
pb = ProgBar()
#m=100,kcandidates=10,nmin=15 -> 53%
forest = treerandom.build_randomized_forest(training_data,m=100,kcandidates=5,nmin=5, callback = pb.callback)
del pb
return forest
def build_random_big_forest(rows,m,kcandidates,nmin,pickcandidate=treerandom.pick_candidate_random,number_of_forests=10,callback=None):
forests = []
for i in range(0,number_of_forests):
pb = ProgBar()
forests.append(treerandom.build_randomized_forest(rows,m,kcandidates,nmin,callback=pb.callback))
del pb
print i
if callback: callback(i,number_of_forests-i)
return forests
def train_randomized_forest(training_data):
print "Training Random Forest"
pb = ProgBar()
#m=100,kcandidates=10,nmin=15 -> 53%
forest = treerandom.build_randomized_forest(training_data,
m=100,
kcandidates=5,
nmin=5,
callback=pb.callback)
del pb
return forest
def do_kcross_validation(fin,finy,kfolds):
print "Starting k=" + str(kfolds)+" validation for random forest"
#there is 2500 tracks
labels = dt.get_lines(finy,int)
pb = ProgBar()
lines = dt.get_lines(fin,float," ", callback = pb.callback)
del pb
#normalize features
lines = dt.transform_features(lines)
data = dt.add_labels_to_lines(lines, labels)
block_size = len(lines)/kfolds
print "chunk size = " + str(block_size)
example_chunks = list(dt.chunks(data, block_size))
#labels_chunks = list(dt.chunks(labels, block_size))
print "number of chunks = " +str(len(example_chunks))
#holds avg accuracy for one forest
accuracy_results = []
#need to add loop here, to loop over configurations of m,n,k
m = [100]
k = [5]
n = [5]
bestm = 0
bestk = 0
bestn = 0
bestaccuracy = 0
for p in range(0,len(m)):
for f in range(0,len(k)):
for g in range(0,len(n)):
for i in range(0,len(example_chunks)):
#we leave set in index i out of train
print "prepare validation set"
validationdata = example_chunks[i]
#extract validation chunk
print "leaving out block " + str(i) + " for validation"
leaveout = i
validationdata = [ exampleentry(validationdata[i][0:len(validationdata[i])-1],validationdata[i][-1]) for i in range(0,len(validationdata)) ]
trainingdata = []
print("merging blocks "),
for j in range(0,len(example_chunks)):
if(j != leaveout):
#print "j="+str(j) + " i="+ str(leaveout)
print(str(j) + ","),
trainingdata = trainingdata + example_chunks[j]
print "\nprepare training set"
print "training on " + str(len(trainingdata))
print "each track has " + str(len(trainingdata[0])) + " features"
pb = ProgBar()
forest = treerandom.build_randomized_forest(trainingdata,m=m[p],kcandidates=k[f],nmin=n[g], callback=pb.callback)
del pb
print "testing on " + str(len(validationdata))
corrects = 0
#classify a set of entries
for example in validationdata:
#print example.features
result = treerandom.classify(example.features,forest)
#print 'expected : ' + str(example.label) + ' result : '+ str(result)
if(result == example.label):
corrects = corrects + 1
#calculate the % of accuracy
accuracy_percentage = (corrects*100)/len(validationdata)
print "accuracy = " + str(accuracy_percentage) + "%"
accuracy_results.append(accuracy_percentage)
avgcc = dt.average(accuracy_results)
print "average accuracy using m="+str(m[p]) + ", k="+str(k[f])+", n="+str(n[g]) + "---> " + str(avgcc) + "%"
if(avgcc > bestaccuracy):
bestm = m[p]
bestk = k[f]
bestn = n[g]
bestaccuracy = avgcc
print "BEST COMBINATION m="+str(bestm) + ", k="+str(bestk)+", n="+str(bestn) + "---> " + str(bestaccuracy) + "%"
def do_kcross_validation(fin, finy, kfolds):
print "Starting k=" + str(kfolds) + " validation for random forest"
#there is 2500 tracks
labels = dt.get_lines(finy, int)
pb = ProgBar()
lines = dt.get_lines(fin, float, " ", callback=pb.callback)
del pb
#normalize features
lines = dt.transform_features(lines)
data = dt.add_labels_to_lines(lines, labels)
block_size = len(lines) / kfolds
print "chunk size = " + str(block_size)
example_chunks = list(dt.chunks(data, block_size))
#labels_chunks = list(dt.chunks(labels, block_size))
print "number of chunks = " + str(len(example_chunks))
#holds avg accuracy for one forest
accuracy_results = []
#need to add loop here, to loop over configurations of m,n,k
m = [100]
k = [5]
n = [5]
bestm = 0
bestk = 0
bestn = 0
bestaccuracy = 0
for p in range(0, len(m)):
for f in range(0, len(k)):
for g in range(0, len(n)):
for i in range(0, len(example_chunks)):
#we leave set in index i out of train
print "prepare validation set"
validationdata = example_chunks[i]
#extract validation chunk
print "leaving out block " + str(i) + " for validation"
leaveout = i
validationdata = [
exampleentry(
validationdata[i][0:len(validationdata[i]) - 1],
validationdata[i][-1])
for i in range(0, len(validationdata))
]
trainingdata = []
print("merging blocks "),
for j in range(0, len(example_chunks)):
if (j != leaveout):
#print "j="+str(j) + " i="+ str(leaveout)
print(str(j) + ","),
trainingdata = trainingdata + example_chunks[j]
print "\nprepare training set"
print "training on " + str(len(trainingdata))
print "each track has " + str(len(
trainingdata[0])) + " features"
pb = ProgBar()
forest = treerandom.build_randomized_forest(
trainingdata,
m=m[p],
kcandidates=k[f],
nmin=n[g],
callback=pb.callback)
del pb
print "testing on " + str(len(validationdata))
corrects = 0
#classify a set of entries
for example in validationdata:
#print example.features
result = treerandom.classify(example.features, forest)
#print 'expected : ' + str(example.label) + ' result : '+ str(result)
if (result == example.label):
corrects = corrects + 1
#calculate the % of accuracy
accuracy_percentage = (corrects *
100) / len(validationdata)
print "accuracy = " + str(accuracy_percentage) + "%"
accuracy_results.append(accuracy_percentage)
avgcc = dt.average(accuracy_results)
print "average accuracy using m=" + str(m[p]) + ", k=" + str(
k[f]) + ", n=" + str(n[g]) + "---> " + str(avgcc) + "%"
if (avgcc > bestaccuracy):
bestm = m[p]
bestk = k[f]
bestn = n[g]
bestaccuracy = avgcc
print "BEST COMBINATION m=" + str(bestm) + ", k=" + str(
bestk) + ", n=" + str(bestn) + "---> " + str(bestaccuracy) + "%"