def computePerformance(self, train_indexes, val_ind):
'''Check errors of training and validation sets
@param train_indexes Tuple that contains indexes of the first and last elements of the training set.
@param val_ind Tuple that contains indexes of the first and last elements of the validation set.
'''
train_error = 0
train_sampl = train_indexes[1] - train_indexes[0] # Count of training samples
for i in range(train_indexes[0], train_indexes[1]):
train_error = train_error + self.computeMlpError(sample = self.data[i])
self.setTrainError(train_error/train_sampl)
if val_ind:
val_error = 0
val_sampl = val_ind[1] - val_ind[0]
answers = np.ma.zeros(val_sampl)
out = np.ma.zeros(val_sampl)
for i in xrange(val_ind[0], val_ind[1]):
sample = self.data[i]
val_error = val_error + self.computeMlpError(sample = self.data[i])
input = np.hstack( (sample['state'],sample['factors']) )
output = self.getOutput(input)
out[i-val_ind[0]] = self.outCategory(output)
answers[i-val_ind[0]] = self.outCategory(sample['output'])
self.setValError(val_error/val_sampl)
depCoef = DependenceCoef(out, answers, expand=True)
self.valKappa = depCoef.kappa(mode=None)
def computePerformance(self, train_indexes, val_ind):
'''Check errors of training and validation sets
@param train_indexes Tuple that contains indexes of the first and last elements of the training set.
@param val_ind Tuple that contains indexes of the first and last elements of the validation set.
'''
train_error = 0
train_sampl = train_indexes[1] - train_indexes[
0] # Count of training samples
for i in range(train_indexes[0], train_indexes[1]):
train_error = train_error + self.computeMlpError(
sample=self.data[i])
self.setTrainError(train_error / train_sampl)
if val_ind:
val_error = 0
val_sampl = val_ind[1] - val_ind[0]
answers = np.ma.zeros(val_sampl)
out = np.ma.zeros(val_sampl)
for i in xrange(val_ind[0], val_ind[1]):
sample = self.data[i]
val_error = val_error + self.computeMlpError(
sample=self.data[i])
input = np.hstack((sample['state'], sample['factors']))
output = self.getOutput(input)
out[i - val_ind[0]] = self.outCategory(output)
answers[i - val_ind[0]] = self.outCategory(sample['output'])
self.setValError(val_error / val_sampl)
depCoef = DependenceCoef(out, answers, expand=True)
self.valKappa = depCoef.kappa(mode=None)
def train(self):
X = np.column_stack((self.data['state'], self.data['factors']))
Y = self.data['output']
self.labelCodes = np.unique(Y)
self.logreg.fit(X, Y, maxiter=self.maxiter)
out = self.logreg.predict(X)
depCoef = DependenceCoef(np.ma.array(out), np.ma.array(Y), expand=True)
self.Kappa = depCoef.kappa(mode=None)
self.pseudoR = depCoef.correctness(percent=False)
def train(self):
X = np.column_stack( (self.data['state'], self.data['factors']) )
Y = self.data['output']
self.labelCodes = np.unique(Y)
self.logreg.fit(X, Y, maxiter=self.maxiter)
out = self.logreg.predict(X)
depCoef = DependenceCoef(np.ma.array(out), np.ma.array(Y), expand=True)
self.Kappa = depCoef.kappa(mode=None)
self.pseudoR = depCoef.correctness(percent = False)
def test_cramer(self):
dc = DependenceCoef(self.X, self.Y)
self.T_cramer = np.subtract(self.T, self.T_cramer_expect)
self.T_cramer = np.square(self.T_cramer)
self.x2 = np.sum(np.divide(self.T_cramer, self.T_cramer_expect))
self.cramer = math.sqrt(self.x2 / (self.total * min(self.r-1,self.s-1)))
self.assertEqual(dc.cramer(), self.cramer, 'cramer coeff failed')
dc = DependenceCoef(self.X, self.X)
self.assertEqual(dc.cramer(), 1.0, 'cramer coeff failed')
def test_correlation(self):
dc = DependenceCoef(self.X, self.Y)
n = len(np.ma.compressed(self.X))
mean_x = np.ma.mean(self.X)
mean_y = np.ma.mean(self.Y)
self.cov = np.ma.sum(np.multiply(np.subtract(self.X, mean_x), np.subtract(self.Y, mean_y)))/n
self.S_x = np.std(self.X)
self.S_y = np.std(self.Y)
self.R = self.cov / (self.S_x * self.S_y)
self.assertEqual(dc.correlation(), self.R,'correlation failed')
dc = DependenceCoef(self.X, self.X)
self.assertEqual(dc.correlation(), 1.0,'correlation failed')
def test_kappa(self):
dc = DependenceCoef(self.Y, self.Y1)
#~ table = np.array([
#~ [1, 2, 1],
#~ [0, 1, 0],
#~ [2, 0, 1],
#~ ])
Pa = 3.0/8
Pe = 21.0/64
Pmax = 6.0/8
answer = (Pa - Pe)/(1 - Pe)
self.assertEqual(dc.kappa(), answer)
answer = (Pa - Pe)/(Pmax - Pe)
self.assertEqual(dc.kappa(mode='loc'), answer)
answer = (Pmax - Pe)/(1 - Pe)
self.assertEqual(dc.kappa(mode='histo'), answer)
def test_kappa(self):
dc = DependenceCoef(self.Y, self.Y1)
#~ table = np.array([
#~ [1, 2, 1],
#~ [0, 1, 0],
#~ [2, 0, 1],
#~ ])
Pa = 3.0 / 8
Pe = 21.0 / 64
Pmax = 6.0 / 8
answer = (Pa - Pe) / (1 - Pe)
self.assertEqual(dc.kappa(), answer)
answer = (Pa - Pe) / (Pmax - Pe)
self.assertEqual(dc.kappa(mode='loc'), answer)
answer = (Pmax - Pe) / (1 - Pe)
self.assertEqual(dc.kappa(mode='histo'), answer)
def test_cramer(self):
dc = DependenceCoef(self.X, self.Y)
self.T_cramer = np.subtract(self.T, self.T_cramer_expect)
self.T_cramer = np.square(self.T_cramer)
self.x2 = np.sum(np.divide(self.T_cramer, self.T_cramer_expect))
self.cramer = math.sqrt(self.x2 /
(self.total * min(self.r - 1, self.s - 1)))
self.assertEqual(dc.cramer(), self.cramer, 'cramer coeff failed')
dc = DependenceCoef(self.X, self.X)
self.assertEqual(dc.cramer(), 1.0, 'cramer coeff failed')
def test_correlation(self):
dc = DependenceCoef(self.X, self.Y)
n = len(np.ma.compressed(self.X))
mean_x = np.ma.mean(self.X)
mean_y = np.ma.mean(self.Y)
self.cov = np.ma.sum(
np.multiply(np.subtract(self.X, mean_x), np.subtract(
self.Y, mean_y))) / n
self.S_x = np.std(self.X)
self.S_y = np.std(self.Y)
self.R = self.cov / (self.S_x * self.S_y)
self.assertEqual(dc.correlation(), self.R, 'correlation failed')
dc = DependenceCoef(self.X, self.X)
self.assertEqual(dc.correlation(), 1.0, 'correlation failed')
def test_jiu(self):
dc = DependenceCoef(self.X, self.Y)
self.assertAlmostEqual(dc.jiu(), 0.385101639127, 9, 'joint coeff failed')
dc = DependenceCoef(self.X, self.X)
self.assertEqual(dc.jiu(), 1.0, 'joint coeff failed')
def test_correctness(self):
dc = DependenceCoef(self.X, self.X)
self.assertEqual(dc.correctness(), 100.0, 'correctness % failed')
def test_jiu(self):
dc = DependenceCoef(self.X, self.Y)
self.assertAlmostEqual(dc.jiu(), 0.385101639127, 9,
'joint coeff failed')
dc = DependenceCoef(self.X, self.X)
self.assertEqual(dc.jiu(), 1.0, 'joint coeff failed')
def test_correctness(self):
dc = DependenceCoef(self.X, self.X)
self.assertEqual(dc.correctness(), 100.0, 'correctness % failed')