def main_with_validation(fileName, intercept=False, evalFunc = Util.brierScore):
trainX, trainY, testX, testY = Util.readData(fileName, intercept=intercept)
clf = GLS()
regs = np.logspace(-3, 3, 7)
xis = np.linspace(-1., 1.5, 26)
bestScore, bestXi, bestReg = 1e10, None, None
for xi in xis:
clf.setXi(xi)
print("Current xi = ", xi)
score, reg = Util.crossValidate(clf, trainX, trainY, \
evalFunc, 5, "Regular", regs)
if score < bestScore:
bestScore, bestXi, bestReg = score, xi, reg
print("bestScore, bestXi, bestReg = ", bestScore, bestXi, bestReg)
clf.setXi(bestXi)
clf.setRegular(bestReg)
clf.fit(trainX, trainY)
testScore = evalFunc(clf.predict(testX), testY)
with open("log/GLS_final_log.txt", 'a') as f:
log = ','.join([dt.now().strftime("%Y/%m/%d %H:%M"), str(fileName), \
str(bestReg), str(bestXi), str(intercept), \
evalFunc.__name__, str(bestScore), str(testScore)])
f.write(log + '\n')
def main_with_validation(fileName, preproc=False, evalFunc = Util.brierScore):
validateX, validateY, \
trainX, trainY, \
testX, testY = Util.readData(fileName, True, preproc)
clf = GLS()
xis = np.linspace(-1., 1.5, 26)
bestScore, bestXi = Util.crossValidate(clf, validateX, validateY, \
evalFunc, 5, "Xi", xis)
print("bestScore, bestXi = ", bestScore, bestXi)
with open("log/GLS_validation_log.txt", 'a') as f:
log = ','.join([dt.now().strftime("%Y/%m/%d %H:%M"), str(fileName), \
str(bestXi), str(preproc), evalFunc.__name__])
f.write(log + '\n')
def worker(chosenNum):
print("Current chosen number: ", chosenNum)
# global validateX, validateY, trainX, trainY, regs, xis, lock, clfs
clf = GLS()
hda = HighDimAlg(clf, "convolve")
hda.setWidth(28)
hda.setKernels([(2,2), (4,4), (7,7)])
target = validateY.copy()
target[target == chosenNum] = 999
target[target != 999] = 0
target[target == 999] = 1
bestScore, bestXi, bestReg = 1e10, None, None
for xi in xis:
hda.setXi(xi)
score, reg = Util.crossValidate(hda, validateX, target, \
Util.brierScore, 2, "Regular", regs)
if score < bestScore:
bestScore, bestXi, bestReg = score, xi, reg
print("bestScore, bestXi, bestReg = ", bestScore, bestXi, bestReg)
hda.setXi(bestXi)
hda.setRegular(bestReg)
target = trainY.copy()
target[target == chosenNum] = 999
target[target != 999] = 0
target[target == 999] = 1
hda.fit(trainX, target)
res = {}
res['num'] = chosenNum
res['clf'] = hda
res['validateScore'] = bestScore
lock.acquire()
try:
clfs.append(res)
finally:
lock.release()
scores = np.zeros(10 * 10).reshape(10, 10)
for i in range(10):
trainX, trainY, testX, testY = Util.readData(source=os.path.join(path, f))
est = MAPGEV.MAPGEV(iterations=20)
est.fit(trainX, trainY)
beta, xi = est.beta.reshape(-1, 1), est.xi
clf = GLS.GLS()
# clf.manualInitial = True
if xi < -1:
clf.setXi(-1)
elif xi > 50:
clf.setXi(50)
else:
clf.setXi(xi)
print("xi = %f clf.xi = %f" % (xi, clf.xi))
bestScore, bestReg = Util.crossValidate(clf, trainX, trainY, Util.brierScore, 5, "Regular", regs)
# bestScore, bestReg = 1., 0.
# for reg in regs:
# clf.setRegular(reg)
# clf.setBeta(beta)
# clf.fit(trainX, trainY)
# score = Util.brierScore(clf.predict(testX), testY)
# if score < bestScore:
# bestScore = score
# bestReg = reg
print("bestScore = %f bestReg = %f" % (bestScore, bestReg))
clf.setRegular(bestReg)
clf.setBeta(beta)
clf.manualInitial = True
clf.fit(trainX, trainY)
scores[i] = Util.evaluate(clf.predict(testX), testY)
