def selectParamKNN(XTrain, yTrain, peopleTrain):
KRange = np.arange(3, 11 + 1, 2)
bestK = 0
bestPerformance = 0
for k in KRange:
clf = KNeighborsClassifier(k)
performance = util.cvPerformance(clf, XTrain, yTrain, peopleTrain)
print "k = " + str(k) + ", accuracy = " + str(performance)
if performance > bestPerformance:
bestPerformance = performance
bestK = k
return bestK
def selectParamLinear(XTrain, yTrain, peopleTrain):
CRange = 10.0**np.arange(-4, 0)
bestC = 0
bestPerformance = 0
for c in CRange:
clf = SVC(kernel='linear', C=c)
performance = util.cvPerformance(clf, XTrain, yTrain, peopleTrain)
print "C = " + str(c) + ", accuracy = " + str(performance)
if performance > bestPerformance:
bestPerformance = performance
bestC = c
return bestC
def selectParamLogReg(XTrain, yTrain, peopleTrain):
CRange = 10.0**np.arange(-4, 1)
bestC = 0
bestPerformance = 0
for c in CRange:
clf = LogisticRegression(C=c,
multi_class="multinomial",
solver="sag",
max_iter=1000)
performance = util.cvPerformance(clf, XTrain, yTrain, peopleTrain)
print "C = " + str(c) + ", accuracy = " + str(performance)
if performance > bestPerformance:
bestPerformance = performance
bestC = c
return bestC
def selectParamRBF(XTrain, yTrain, peopleTrain):
bestPerformance = 0
bestTuple = (0, 0)
gammaRange = 10.0**np.arange(-4, -1)
CRange = 10.0**np.arange(0, 4)
for gamma in gammaRange:
for c in CRange:
clf = SVC(kernel='rbf', C=c, gamma=gamma)
score = util.cvPerformance(clf, XTrain, yTrain, peopleTrain)
print "gamma = " + str(gamma) + ", C = " + str(
c) + ", accuracy = " + str(score)
if score > bestPerformance:
bestPerformance = score
bestTuple = (gamma, c)
return bestTuple