def generate_record(self):
retval = SparseVector()
for i in range(0, self.dim):
retval.components[i] = random.choice([-1, 1]) + random.uniform(
-self.noiseEps, +self.noiseEps)
return retval
def testLinearKernel(self):
sigma = 1.0
kernel = LinearKernel()
x1 = SparseVector()
x1[0] = 1
x1[1] = 0
x2 = SparseVector()
x2[0] = 0
x2[1] = 1
val = kernel.compute(x1, x2)
true_val = np.dot(np.array([1, 0]), np.array([0, 1]))
self.assert_(
val == true_val,
"Linear kernel of two vectors is not how I expected it...")
def _generate_example(self):
iNumberOfFeaturesPerStock = 4
iPosCurStock = self.m_aStocks.index(self.m_stActStock)
iTargetStockPos = iNumberOfFeaturesPerStock * iPosCurStock
iFeatureIndex = self.prediction_index - 1
if iFeatureIndex < 0:
iFeatureIndex = len(self.m_lFeatures) - 1
record = SparseVector()
features = self.getFeatures(iFeatureIndex)
label = self.getLabel(self.prediction_index)
for i in xrange(len(features)):
val = features[i]
if math.isnan(val) or math.isinf(val):
print "Error, invalid feature value: " + str(val)
val = 0.0
#the following code ensures that the features of the target stock are always at the beginning of the feature vector
# if i < iTargetStockPos or i >= iTargetStockPos + iNumberOfFeaturesPerStock:
# record[i+iNumberOfFeaturesPerStock] = val
# else:
# record[i-iTargetStockPos] = val
record[i] = val
self.prediction_index += 1
if self.prediction_index >= len(self.m_hPrices[self.m_stActStock]):
self.prediction_index = 0
if self.generated_examples % self.number_of_nodes == 0:
self.drift()
return (record, label)
def generate_example(self):
record = SparseVector()
for i in xrange(len(self.m_aFeatureNames)):
aFeatures = self.getFeatures(self.prediction_index)
record.components[self.m_aFeatureNames[i]] = aFeatures[i]
label = self.m_aAnnotations[self.m_stTargetStock][
self.prediction_index]
if self.m_bDelay == True:
if self.prediction_index % self.m_iAccountingPeriod == 0:
label = []
for t in xrange(
self.prediction_index - self.m_iAccountingPeriod + 1,
self.prediction_index + 1):
if t >= 0:
label.append(
self.m_aAnnotations[self.m_stTargetStock][t])
else:
label = None
self.prediction_index += 1
if self.prediction_index % self.m_iRepetitionInterval == 0:
if self.m_iCurrRepitions < self.m_iRepetitions:
self.prediction_index -= self.m_iRepetitionInterval
self.m_iCurrRepitions += 1
else:
self.m_iCurrRepitions = 0
if self.prediction_index >= len(self.m_aPrices[self.m_stTargetStock]):
if len(self.m_aTargetStocks) > 0:
self.m_stTargetStock = self.m_aTargetStocks.pop(0)
self.m_oCorrel.changeSet([self.m_stTargetStock])
self.prediction_index = 0
return (record, label)
def _get_sparse_rep(self, example_batch):
sv_list = []
for ex in example_batch:
rec = SparseVector()
for i in range(len(ex)):
rec[i] = ex[i]
sv_list.append(rec)
return sv_list
def _generate_example(self):
self._try_drift()
label = self.label_distribution.rvs()
hidden_values = []
record = set()
for i in xrange(len(self.hidden_layer)):
hidden_values.append(self.hidden_layer[i].rvs(label))
for i in xrange(self.dim):
if self.output_layer[i].rvs(hidden_values[i % len(self.hidden_layer)]) == 1:
record.add(i)
if label == 0: label = -1
record_vector = SparseVector()
for i in xrange(self.dim):
if i in record:
record_vector.components[i] = 1.0
else:
record_vector.components[i] = -1.0
return (record_vector, label)
def testKernelModelAveraging(self):
kernel = LinearKernel() #GaussianKernel(sigma)
truncOp = NoCompression()
x1 = SparseVector()
x1[0] = 1
x1[1] = 0
x1[2] = 0
sv1 = SupportVector()
sv1.record = x1
sv1.weight = 0.1
x2 = SparseVector()
x2[0] = 0
x2[1] = 1
x2[2] = 0
sv2 = SupportVector()
sv2.record = x2
sv2.weight = 0.2
x3 = SparseVector()
x3[0] = 0
x3[1] = 1
x3[2] = 0
sv3 = SupportVector()
sv3.record = x2
sv3.weight = 0.3
model1 = KernelClassificationModel(kernel, truncOp)
model2 = KernelClassificationModel(kernel, truncOp)
model1.supportVectors.append(sv1)
model1.supportVectors.append(sv3)
model2.supportVectors.append(sv2)
model3 = aritmethic_mean([model1, model2])
self.assert_(model3.supportVectors[0].weight == sv1.weight / 2.0,
"Averaging of models failed.")
self.assert_(
model3.supportVectors[1].weight == (sv2.weight + sv3.weight) / 2.0,
"Averaging of models failed.")
def testGaussKernel(self):
sigma = 1.0
kernel = GaussianKernel(sigma)
x1 = SparseVector()
x1[0] = 1
x1[1] = 0
x2 = SparseVector()
x2[0] = 0
x2[1] = 1
val1 = kernel.compute(x1, x1)
self.assert_(
val1 == 1.0,
"Gaussian kernel of vector with itself is not equal to 1.")
val2 = kernel.compute(x1, x2)
true_norm = np.linalg.norm(np.array([1, 0]) - np.array([0, 1]))
true_exp = -1 * (true_norm**2)
true_exp = true_exp / (2 * (sigma**2))
true_val = np.exp(true_exp)
self.assert_(
val2 == true_val,
"Gaussian kernel of two vectors is not how I expected it...")
def _generate_example(self):
record = SparseVector()
features = self.getFeatures(self.prediction_index - 1)
label = self.getLabel(self.prediction_index)
if math.isnan(label) or math.isinf(label):
print "Error, invalid label: " + str(label)
label = 0.0
for i, val in enumerate(features):
if math.isnan(val) or math.isinf(val):
print "Error, invalid feature value: " + str(val)
val = 0.0
record[i] = val
self.prediction_index += 1
return (record, label)
def generate_record(self):
return SparseVector()
def _generate_example(self):
record = SparseVector(0.0)
for i in xrange(self.dim):
record.components[i] = 0.0
return (record, 0.0)
def generate_record(self):
retval = SparseVector()
for i in range(0, self.dim):
retval.components[i] = random.choice([-1, 1])
return retval
def generate_record(self):
record = SparseVector()
for key in self.singletons:
val = uniform(self.minVal, self.maxVal)
record.__setitem__(key, val)
return record