def test_GeometricCorner(display=0):
#Create Test Data: defined on a 3x3 square comprising (0,3) and (0,3)
n = 1000
x = np.random.random_sample((n, 2)) * 3
#Let's create a dividing line on x = 1
y = np.reshape(np.logical_and(x[:, 0] > 2, x[:, 1] > 1), (1, n))
dataSet = np.hstack((x, y.T))
dataLength = len(dataSet)
if display:
plt.scatter(x[:, 0], x[:, 1], c=y)
plt.show()
DT = BranchNode().\
set_maxDepth(2).\
set_regressorIndex(2).\
set_columnTypes([0,0,1])
for i in range(0, 1):
#print DT.findMinimumSplit(dataSet[np.random.choice(range(0,dataLength),size=dataLength*.5,replace=False)],1,2,0,1)
DT = DT.set_dataSet(dataSet[np.random.choice(range(0, dataLength),
size=dataLength * .5,
replace=False)])
print DT.constructTreeFromNode()
def test_FindMinimumSplit_DiscreteContinuous():
n = 100
xValues = [0, 1, 2]
xProbs1 = [0, 0, 1]
xProbs2 = [.5, .5, 0]
l = []
for i in range(0, 20):
x = np.reshape(
np.concatenate((np.random.choice(xValues, size=n * .5, p=xProbs1),
np.random.choice(xValues, size=n * .5,
p=xProbs2))), (1, n))
y = np.reshape(
np.concatenate((
np.random.random(n * .5) * .5 +
10, #Average 10.25 (range 0.5) <- lower variance should be taken
np.random.random(n * .5) * 2 + 20)), #Average 21.00 (range 2)
(1, n))
DT = BranchNode()
dataSet = np.vstack((x, y)).T
DT = DT.findMinimumSplit(dataSet, 0, 1, 1, 0)
splitIndex = DT.decisionPoint
l.append(splitIndex)
print DT
if collections.Counter(l).most_common()[0][0] == 2.0:
print '*\t\tDiscrete Continuous Test passed'
else:
print '**\t\tDiscrete Continuous Test failed'
def test_GiniImpurity():
n = 100
DT = BranchNode().set_impurityType('i')
data = np.arange(1, n)
maxIG = DT.impurityFunction(data)
data = np.ones(n)
minIG = DT.impurityFunction(data)
data = np.concatenate((np.zeros(n - 1), np.ones(1)))
midIG = DT.impurityFunction(data)
if minIG == 0 and (minIG < midIG < maxIG):
print '*\t\tGini Impurity Test passed'
else:
print '**\t\tGini Impurity Test failed'
def test_FindMinimumSplit_DiscreteDiscrete():
n = 50
yValues = [0, 1, 2]
xValues = ['a', 'b', 'c']
xProbs1 = [1, 0, 0]
yProbs1 = [.3, .3, .4]
yProbs2 = [0, 1, 0]
xProbs2 = [0, 1, 0]
y = np.concatenate((np.random.choice(yValues, size=n, p=yProbs1),
np.random.choice(yValues, size=n, p=yProbs2)))
x = np.concatenate((np.random.choice(xValues, size=n, p=xProbs1),
np.random.choice(xValues, size=n, p=xProbs2)))
DT = BranchNode()
dataSet = np.vstack((x, y)).T
DT = DT.findMinimumSplit(dataSet, 0, 1, 1, 1)
if DT.trueVal == '1' and DT.decisionPoint == 'b' and DT.trueValProportions == 1.0:
print '*\t\tDiscrete Discrete Test passed'
else:
print '**\t\tDiscrete Discrete Test failed'
def test_FindMinimumSplit_ContinuousContinuous_1():
l = []
for i in range(0, 100):
n = 100
y = np.arange(0, n)
x = np.concatenate((
np.random.random(n * .5) * 2 + 10, #average 11 (max 12)
np.random.random(n * .5) * 2 + 12)) #average 13 (min 12)
DT = BranchNode()
dataSet = np.vstack((x, y)).T
DT = DT.findMinimumSplit(dataSet, 0, 1, 0, 0)
l.append(DT.decisionPoint)
print DT
decisionPoint = np.mean(l)
if decisionPoint < 12.5 and decisionPoint > 11.5:
print '*\t\tContinuous Continuous Test 1 passed'
else:
print '**\t\tContinuous Continuous Test 1 failed'
print np.mean(l)
print np.std(l)
def test_FindMinimumSplit_ContinuousDiscrete():
n = 50
yValues = [0, 1, 2]
yProbs1 = [.5, .5, 0]
yProbs2 = [0, .1, .9]
x = np.concatenate((
np.random.random(n) * 10 + 10, #average 15 (max 20)
np.random.random(n) * 10 + 20)) #average 25 (min 20)
y = np.concatenate((np.random.choice(yValues, size=n, p=yProbs1),
np.random.choice(yValues, size=n, p=yProbs2)))
DT = BranchNode()
DT = DT.set_depth(5)
DT = DT.set_impurityType('g')
dataSet = np.vstack((x, y)).T
DT = DT.findMinimumSplit(dataSet, 0, 1, 0, 1)
# DT.establishProportions()
print DT
decisionPoint = DT.decisionPoint
if decisionPoint < 21 and decisionPoint > 20:
print '*\t\tContinuous Discrete passed'
else:
print '**\t\tContinuous Discrete failed'