def computeIdealPenalty(args):
"""
Find the complete penalty.
"""
(X, y, fullX, C, gamma, gridPoints, pdfX, pdfY1X, pdfYminus1X) = args
svm = LibSVM('gaussian', gamma, C)
svm.learnModel(X, y)
predY = svm.predict(X)
predFullY, decisionsY = svm.predict(fullX, True)
decisionGrid = numpy.reshape(decisionsY, (gridPoints.shape[0], gridPoints.shape[0]), order="F")
trueError = ModelSelectUtils.bayesError(gridPoints, decisionGrid, pdfX, pdfY1X, pdfYminus1X)
idealPenalty = trueError - Evaluator.binaryError(predY, y)
return idealPenalty
def testGetWeights(self):
try:
import sklearn
except ImportError as error:
return
numExamples = 6
X = numpy.array([[-3], [-2], [-1], [1], [2], [3]], numpy.float64)
#X = numpy.random.rand(numExamples, 10)
y = numpy.array([[-1], [-1], [-1], [1], [1], [1]])
svm = LibSVM()
svm.learnModel(X, y.ravel())
weights, b = svm.getWeights()
#Let's see if we can compute the decision values
y, decisions = svm.predict(X, True)
decisions2 = numpy.zeros(numExamples)
decisions2 = numpy.dot(X, weights) - b
self.assertTrue((decisions == decisions2).all())
predY = numpy.sign(decisions2)
self.assertTrue((y.ravel() == predY).all())
#Do the same test on a random datasets
numExamples = 50
numFeatures = 10
X = numpy.random.rand(numExamples, numFeatures)
y = numpy.sign(numpy.random.rand(numExamples) - 0.5)
svm = LibSVM()
svm.learnModel(X, y.ravel())
weights, b = svm.getWeights()
#Let's see if we can compute the decision values
y, decisions = svm.predict(X, True)
decisions2 = numpy.dot(X, weights) + b
tol = 10**-6
self.assertTrue(numpy.linalg.norm(decisions - decisions2) < tol)
predY = numpy.sign(decisions2)
self.assertTrue((y.ravel() == predY).all())
def testSetSvmType(self):
try:
import sklearn
except ImportError as error:
return
numExamples = 100
numFeatures = 10
X = numpy.random.randn(numExamples, numFeatures)
X = Standardiser().standardiseArray(X)
c = numpy.random.randn(numFeatures)
y = numpy.dot(X, numpy.array([c]).T).ravel() + 1
y2 = numpy.array(y > 0, numpy.int32) * 2 - 1
svm = LibSVM()
svm.setSvmType("Epsilon_SVR")
self.assertEquals(svm.getType(), "Epsilon_SVR")
#Try to get a good error
Cs = 2**numpy.arange(-6, 4, dtype=numpy.float)
epsilons = 2**numpy.arange(-6, 4, dtype=numpy.float)
bestError = 10
for C in Cs:
for epsilon in epsilons:
svm.setEpsilon(epsilon)
svm.setC(C)
svm.learnModel(X, y)
yp = svm.predict(X)
if Evaluator.rootMeanSqError(y, yp) < bestError:
bestError = Evaluator.rootMeanSqError(y, yp)
self.assertTrue(
bestError < Evaluator.rootMeanSqError(y, numpy.zeros(y.shape[0])))
svm.setSvmType("C_SVC")
svm.learnModel(X, y2)
yp2 = svm.predict(X)
self.assertTrue(0 <= Evaluator.binaryError(y2, yp2) <= 1)
def testGetWeights(self):
try:
import sklearn
except ImportError as error:
return
numExamples = 6
X = numpy.array([[-3], [-2], [-1], [1], [2] ,[3]], numpy.float64)
#X = numpy.random.rand(numExamples, 10)
y = numpy.array([[-1], [-1], [-1], [1], [1] ,[1]])
svm = LibSVM()
svm.learnModel(X, y.ravel())
weights, b = svm.getWeights()
#Let's see if we can compute the decision values
y, decisions = svm.predict(X, True)
decisions2 = numpy.zeros(numExamples)
decisions2 = numpy.dot(X, weights) - b
self.assertTrue((decisions == decisions2).all())
predY = numpy.sign(decisions2)
self.assertTrue((y.ravel() == predY).all())
#Do the same test on a random datasets
numExamples = 50
numFeatures = 10
X = numpy.random.rand(numExamples, numFeatures)
y = numpy.sign(numpy.random.rand(numExamples)-0.5)
svm = LibSVM()
svm.learnModel(X, y.ravel())
weights, b = svm.getWeights()
#Let's see if we can compute the decision values
y, decisions = svm.predict(X, True)
decisions2 = numpy.dot(X, weights) + b
tol = 10**-6
self.assertTrue(numpy.linalg.norm(decisions - decisions2) < tol)
predY = numpy.sign(decisions2)
self.assertTrue((y.ravel() == predY).all())
def testSetSvmType(self):
try:
import sklearn
except ImportError as error:
return
numExamples = 100
numFeatures = 10
X = numpy.random.randn(numExamples, numFeatures)
X = Standardiser().standardiseArray(X)
c = numpy.random.randn(numFeatures)
y = numpy.dot(X, numpy.array([c]).T).ravel() + 1
y2 = numpy.array(y > 0, numpy.int32)*2 -1
svm = LibSVM()
svm.setSvmType("Epsilon_SVR")
self.assertEquals(svm.getType(), "Epsilon_SVR")
#Try to get a good error
Cs = 2**numpy.arange(-6, 4, dtype=numpy.float)
epsilons = 2**numpy.arange(-6, 4, dtype=numpy.float)
bestError = 10
for C in Cs:
for epsilon in epsilons:
svm.setEpsilon(epsilon)
svm.setC(C)
svm.learnModel(X, y)
yp = svm.predict(X)
if Evaluator.rootMeanSqError(y, yp) < bestError:
bestError = Evaluator.rootMeanSqError(y, yp)
self.assertTrue(bestError < Evaluator.rootMeanSqError(y, numpy.zeros(y.shape[0])))
svm.setSvmType("C_SVC")
svm.learnModel(X, y2)
yp2 = svm.predict(X)
self.assertTrue(0 <= Evaluator.binaryError(y2, yp2) <= 1)
def testPredict(self):
try:
import sklearn
except ImportError as error:
return
numExamples = 100
numFeatures = 10
X = numpy.random.randn(numExamples, numFeatures)
c = numpy.random.randn(numFeatures)
y = numpy.dot(X, numpy.array([c]).T).ravel()
y = numpy.array(y > 0, numpy.int32) * 2 - 1
svm = LibSVM()
svm.learnModel(X, y)
y2, d = svm.predict(X, True)
def testComputeTestError(self):
C = 10.0
gamma = 0.5
numTrainExamples = self.X.shape[0]*0.5
trainX, trainY = self.X[0:numTrainExamples, :], self.y[0:numTrainExamples]
testX, testY = self.X[numTrainExamples:, :], self.y[numTrainExamples:]
svm = LibSVM('gaussian', gamma, C)
args = (trainX, trainY, testX, testY, svm)
error = computeTestError(args)
svm = LibSVM('gaussian', gamma, C)
svm.learnModel(trainX, trainY)
predY = svm.predict(testX)
self.assertEquals(Evaluator.binaryError(predY, testY), error)
def testPredict(self):
try:
import sklearn
except ImportError as error:
return
numExamples = 100
numFeatures = 10
X = numpy.random.randn(numExamples, numFeatures)
c = numpy.random.randn(numFeatures)
y = numpy.dot(X, numpy.array([c]).T).ravel()
y = numpy.array(y > 0, numpy.int32)*2 -1
svm = LibSVM()
svm.learnModel(X, y)
y2, d = svm.predict(X, True)
def testComputeTestError(self):
C = 10.0
gamma = 0.5
numTrainExamples = self.X.shape[0] * 0.5
trainX, trainY = self.X[
0:numTrainExamples, :], self.y[0:numTrainExamples]
testX, testY = self.X[numTrainExamples:, :], self.y[numTrainExamples:]
svm = LibSVM('gaussian', gamma, C)
args = (trainX, trainY, testX, testY, svm)
error = computeTestError(args)
svm = LibSVM('gaussian', gamma, C)
svm.learnModel(trainX, trainY)
predY = svm.predict(testX)
self.assertEquals(Evaluator.binaryError(predY, testY), error)
"""
#Figure out why the penalty is increasing
X = trainX
y = trainY
for i in range(foldsSet.shape[0]):
folds = foldsSet[i]
idx = Sampling.crossValidation(folds, validX.shape[0])
penalty = 0
fullError = 0
trainError = 0
learner.learnModel(validX, validY)
predY = learner.predict(X)
predValidY = learner.predict(validX)
idealPenalty = Evaluator.rootMeanSqError(predY, y) - Evaluator.rootMeanSqError(predValidY, validY)
for trainInds, testInds in idx:
trainX = validX[trainInds, :]
trainY = validY[trainInds]
#learner.setGamma(gamma)
#learner.setC(C)
learner.learnModel(trainX, trainY)
predY = learner.predict(validX)
predTrainY = learner.predict(trainX)
fullError += Evaluator.rootMeanSqError(predY, validY)
trainError += Evaluator.rootMeanSqError(predTrainY, trainY)
penalty += Evaluator.rootMeanSqError(predY, validY) - Evaluator.rootMeanSqError(predTrainY, trainY)
validX = trainX[trainInds,:]
validY = trainY[trainInds]
#errors = learner.parallelPenaltyGrid(validX, validY, testX, testY, paramDict, computeTestError)
#errors = numpy.squeeze(errors)
errors = numpy.zeros((Cs.shape[0], gammas.shape[0]))
norms = numpy.zeros((Cs.shape[0], gammas.shape[0]))
for i, C in enumerate(Cs):
for j, gamma in enumerate(gammas):
learner.setEpsilon(epsilons[0])
learner.setC(C)
learner.setGamma(gamma)
learner.learnModel(validX, validY)
predY = learner.predict(testX)
errors[i, j] = Evaluator.meanAbsError(predY, testY)
norms[i, j] = learner.weightNorm()
for i in range(gammas.shape[0]):
plt.figure(i)
plt.plot(numpy.log(Cs), errors[:, i], label=str(sampleSize))
plt.legend(loc="upper left")
plt.xlabel("C")
plt.ylabel("Error")
plt.figure(i+gammas.shape[0])
plt.plot(norms[:, i], errors[:, i], label=str(sampleSize))
plt.legend(loc="upper left")
plt.xlabel("Norm")