# For dat file, use the input file name with the .csv extension
tokens = test.split('.')
testFileRoot = str.join('.', tokens[:-1])
datFileName = testFileRoot + '.csv'
jp_results = []
ps_results = []
fp_results = []
jp_run = []
ps_run = []
fp_run = []
for i in range(tries):
sdg = synthDataGen.run(test, datSize)
d = getData.DataReader(datFileName)
data = d.read()
prob = ProbSpace(data)
lim = 3 # Std's from the mean to test conditionals
numPts = 30 # How many eval points for each conditional
print('Test Limit = ', lim, 'standard deviations from mean')
print('Dimensions = ', dims, '. Conditionals = ', dims - 1)
print('Number of points to test for each conditional = ', numPts)
N = prob.N
evalpts = int(
sqrt(N)
) # How many target points to sample for expected value: E(Z | X=x. Y=y)
print('JPROB points for mean evaluation = ', evalpts)
vars = prob.fieldList
cond = []
# Get the conditional variables
for i in range(len(vars)):
print('Testing: ', test, '--', testDescript)
# For dat file, use the input file name with the .csv extension
tokens = test.split('.')
testFileRoot = str.join('.', tokens[:-1])
datFileName = testFileRoot + '.csv'
jp_results = []
ps_results = []
jp_run = []
ps_run = []
for i in range(tries):
sdg = synthDataGen.run(test, datSize)
d = getData.DataReader(datFileName)
data = d.read()
prob = ProbSpace(data)
lim = 3 # Std's from the mean to test conditionals
numPts = 30 # How many eval points for each conditional
print('Test Limit = ', lim, 'standard deviations from mean')
print('Dimensions = ', dims, '. Conditionals = ', dims - 1)
print('Number of points to test for each conditional = ', numPts)
N = prob.N
evalpts = int(
sqrt(N)
) # How many target points to sample for expected value: E(Z | X=x. Y=y)
print('JPROB points for mean evaluation = ', evalpts)
vars = prob.fieldList
cond = []
# Get the conditional variables
for i in range(len(vars)):
print('Testing: ', test, '--', testDescript)
# For dat file, use the input file name with the .csv extension
tokens = test.split('.')
testFileRoot = str.join('.', tokens[:-1])
datFileName = testFileRoot + '.csv'
jp_results = []
ps_results = []
jp_run = []
ps_run = []
for i in range(tries):
sdg = synthDataGen.run(test, datSize)
d = getData.DataReader(datFileName)
data = d.read()
prob = ProbSpace(data)
lim = 3 # Std's from the mean to test conditionals
numPts = 30 # How many eval points for each conditional
print('Test Limit = ', lim, 'standard deviations from mean')
print('Dimensions = ', dims, '. Conditionals = ', dims - 1)
print('Number of points to test for each conditional = ', numPts)
N = prob.N
cond = 'B'
target = 'A2'
R1 = RKHS(prob.ds, delta=None, includeVars=[target, cond], s=smoothness)
# Do some general assessment of cumulative probabilities
# Do some univariate CDF calculations. F is normal(0,1)
sum2 += t1*t2
if sum1 ==0:
return 0
return sum2/sum1
if __name__ == '__main__':
path = '../models/Cprobdata.csv'
d = getData.DataReader(path)
data = d.read()
X1 = data['X']
Y1 = data['Y']
X = X1[:7500]
Y = Y1[:7500]
ps = ProbSpace(data)
size = len(X)
testPoints = []
testMin = 5
testMax = 7
tp = testMin
numTP = 100
interval = (testMax - testMin) / numTP
sq = []
probpy = []
cond = []
Perror = 0.0
Pdev = []
def condP(self, Vals,K = None):
#Vals is a list of (x1,x2....xn) such that P(X1=x1|X2=x2.....), same UI as rkhsmv
if(K != None):
self.k = K
filter_len = floor((len(self.includeVars)-1)*self.k*0.01)
dims = len(Vals)
print(dims,filter_len)
if(self.rangeFactor==0):
self.rangeFactor = 0.5
if(filter_len !=0):
filter_vars = self.includeVars[-filter_len:]
filter_vals = Vals[-filter_len:]
include_vars = self.includeVars[:-filter_len]
self.minPoints = ceil(self.N**((dims-filter_len)/dims))*self.rangeFactor
self.maxPoints = ceil(self.N**((dims-filter_len)/dims))/self.rangeFactor
else:
filter_vars = []
filter_vals = []
include_vars = self.includeVars
# print("filter vars:",filter_vars)
# print("include vars:",self.includeVars[:-filter_len])
# print(self.includeVars)
#Calculating R1
zDim = floor(self.k * (dims-1)* 0.01)
minminpoints = 10
if(filter_len == (len(self.includeVars)-1)):
for i in range(zDim,0,-1):
print("runing i=",i)
filter_len = floor(i*self.k*0.01)
self.minPoints = ceil(self.N*((dims-i)/dims))*self.rangeFactor
self.maxPoints = ceil(self.N*((dims-i)/dims))/self.rangeFactor
rkhsminpoints = minminpoints*(dims-i)
if(self.minPoints < rkhsminpoints):
print("minpoints < minminpoints",self.minPoints,rkhsminpoints)
continue
P = ProbSpace(self.data)
filter_data = []
for j in range(filter_len):
x = (filter_vars[j],filter_vals[j])
filter_data.append(x)
print("filter metrics = ",filter_data)
FilterData, parentProb, finalQuery = P.filter(filter_data,self.minPoints,self.maxPoints)
if(len(FilterData[self.includeVars[0]])<self.minPoints):
print("not enough filter points for filterlen =",i,len(FilterData[self.includeVars[0]]),self.minPoints,self.maxPoints)
continue
print("filter len",filter_len)
print("filtered datapoints:",len(FilterData['B']))
print("include vars:",self.includeVars[:-filter_len])
self.R1 = RKHS(FilterData,includeVars=self.includeVars[:(dims-filter_len)],delta=self.delta,s=self.s)
self.r1filters = filter_vals
return self.R1.P(Vals[:dims-filter_len])
print("running empty")
self.R1 = RKHS(self.data,includeVars=self.includeVars,delta=self.delta,s=self.s)
p = self.R1.condP(Vals)
if p>0:
return p
else:
return None
elif(filter_len != 0 and self.r1filters != filter_vals):
P = ProbSpace(self.data)
filter_data = []
for i in range(filter_len):
x = (filter_vars[i],filter_vals[i])
filter_data.append(x)
FilterData, parentProb, finalQuery = P.filter(filter_data,self.minPoints,self.maxPoints)
# print("filter len",filter_len)
# print("filtered datapoints:",len(FilterData['B']))
# print("include vars:",self.includeVars[:-filter_len])
self.R1 = RKHS(FilterData,includeVars=self.includeVars[:-filter_len],delta=self.delta,s=self.s)
self.r1filters = filter_vals
elif(self.R1==None):
print("running empty")
self.R1 = RKHS(self.data,includeVars=self.includeVars,delta=self.delta,s=self.s)
elif(self.R1.varNames != include_vars):
self.R1 = RKHS(self.data,includeVars=self.includeVars,delta=self.delta,s=self.s)
if(filter_len != 0):
p = self.R1.condP(Vals[:-filter_len])
if p>0:
return p
else:
return None
else:
p = self.R1.condP(Vals)
if p>0:
return p
else:
return None
def condE(self,target, Vals,K = None):
#Vals is a list of (x1,x2....xn) such that E(Y|X1=x1,X2=x2.....), same UI as rkhsmv
if(K == None):
K = self.k
filter_len = floor((len(self.includeVars)-1)*K*0.01)
#print("filter len",filter_len)
dims = len(Vals) + 1
if(self.rangeFactor == None):
self.rangeFactor = 0.8
minminpoints = 5
if(filter_len !=0):
filter_vars = self.includeVars[-filter_len:]
filter_vals = Vals[-filter_len:]
include_vars = self.includeVars[1:-filter_len]
self.minPoints = self.N**((dims-filter_len)/dims)*self.rangeFactor
self.maxPoints = self.N**((dims-filter_len)/dims)/self.rangeFactor
#print("minpoints,maxpoints=",self.minPoints,self.maxPoints)
else:
filter_vars = []
filter_vals = []
include_vars = self.includeVars
#print("filter vars:",filter_vars)
#print("include vars:",self.includeVars[:-filter_len])
#print("self:",self.R2.varNames,"cond",self.includeVars[:-filter_len])
if(filter_len == (len(self.includeVars)-1) ):
P = ProbSpace(self.data)
filter_vars = self.includeVars[1:]
filter_vals = Vals
filter_data = []
for i in range(filter_len):
x = (filter_vars[i],filter_vals[i])
filter_data.append(x)
#print("minpoints,maxpoints:",self.minPoints,self.maxPoints)
FilterData, parentProb, finalQuery = P.filter(filter_data,self.minPoints,self.maxPoints)
X = FilterData[self.includeVars[0]]
if(len(X)<self.minPoints):
newk = ceil((((K*(dims-1)*0.01)-1)/(dims-1))*100) # update K =100 to K = 80
#newk = ceil(K - ((filter_len-1)/filter_len)*100) #update k = 100 to K = 20
print("not enough datapoints, newk=",newk)
return self.condE(target, Vals, newk)
#print(len(X))
if(len(X)!=0):
return sum(X)/len(X)
else:
return 0
elif(filter_len != 0 and self.r2filters != filter_vals):
P = ProbSpace(self.data)
filter_data = []
for i in range(filter_len):
x = (filter_vars[i],filter_vals[i])
#print(x)
filter_data.append(x)
FilterData, parentProb, finalQuery = P.filter(filter_data,self.minPoints,self.maxPoints)
print("filter len",filter_len)
print("filtered datapoints:",len(FilterData['B']))
print("include vars:",self.includeVars[:-filter_len])
X = FilterData[self.includeVars[0]]
if(len(X)<self.minPoints or len(X)<=minminpoints):
newk = ceil((((K*(dims-1)*0.01)-1)/(dims-1)) * 100)
#newk = ceil(((filter_len+1)/filter_len)*K)
print("not enough datapoints, newk=",newk)
return self.condE(target, Vals, newk)
self.R2 = RKHS(FilterData,includeVars=self.includeVars[1:-filter_len],delta=self.delta,s=self.s)
self.r2filters = filter_vals
elif(self.R2==None):
self.R2 = RKHS(self.data,includeVars=self.includeVars[1:],delta=self.delta,s=self.s)
elif(self.R2.varNames != include_vars):
self.R2 = RKHS(self.data,includeVars=self.includeVars[1:],delta=self.delta,s=self.s)
if(filter_len !=0):
return self.R2.condE(target,Vals[:-filter_len])
else:
return self.R2.condE(target, Vals)
else:
v3 = 'C'
f = open(test, 'r')
exec(f.read(), globals())
print('Testing: ', test, '--', testDescript)
# For dat file, use the input file name with the .csv extension
tokens = test.split('.')
testFileRoot = str.join('.', tokens[:-1])
datFileName = testFileRoot + '.csv'
d = getData.DataReader(datFileName)
data = d.read()
prob = ProbSpace(data)
traces = {}
traces['X'] = []
traces['Y'] = []
traces['Z'] = []
v1distr = prob.distr(v1)
v2distr = prob.distr(v2)
v3distr = prob.distr(v3)
v1mean = v1distr.E()
v2mean = v2distr.E()
v3mean = v3distr.E()
#ymin = targdistr.minVal()
#ymax = targdistr.maxVal()
#xmin = conddistr.minVal()
#xmax = conddistr.maxVal()
lim = 3
def run(filename):
r = getData.DataReader(filename)
dat = r.read()
ps = ProbSpace(dat, density=1, power=1)
start = time.time()
print()
print('Testing probability module.')
print()
print('Testing basic statistics for various types of distribution:')
print('stats(A) = ', ps.fieldStats('A'))
print('stats(C) = ', ps.fieldStats('C'))
a = ps.distr('A')
mean = a.mean()
std = a.stDev()
print('stats(dice1): mean, std, skew, kurtosis, median, mode = ', mean,
std, a.skew(), a.kurtosis(), ' Exp: (3.5, ?, 0, ?)')
c = ps.distr('C')
print('stats(d1 + d2): mean, std, skew, kurtosis, median, mode = ', c.E(),
c.stDev(), c.skew(), c.kurtosis(), c.median(), c.mode(),
' Exp: (7, ?, 0, ?, 7, 7)')
d = ps.distr('EXP')
print('stats(Exponential): mean, std, skew, kurtosis = ', d.E(), d.stDev(),
d.skew(), d.kurtosis(), ' Exp: (1, 1, 2, 6)')
d = ps.distr('IVB')
print('stats(Logistic): mean, std, skew, kurtosis = ', d.E(), d.stDev(),
d.skew(), d.kurtosis(), ' Exp: (0, 1.8138, 0, 1.2)')
d = ps.distr('N')
print('stats(Normal): mean, std, skew, kurtosis, median = ', d.E(),
d.stDev(), d.skew(), d.kurtosis(), d.median(), 'Exp: (0, 1, 0, 0)')
d = ps.distr('N2')
print('stats(N2: sum of normals): mean, std, skew, kurtosis = ', d.E(),
d.stDev(), d.skew(), d.kurtosis(), 'Exp: (1, 1.414, 0, 0)')
print()
print(
'Testing discrete deterministic probabilities (2-dice -- ala Craps):')
print('A is Die #1. B is Die #2. C is the total of the 2 dice.')
print('E(B) = ', ps.distr('B').E(), ' Exp: 3.5')
print('P(B=0) = ', ps.P(('B', 0)), ' Exp: 0')
print('P(B=1) = ', ps.P(('B', 1)), ' Exp: 1/6 = .166...')
print('P(B=2) = ', ps.P(('B', 2)), ' Exp: 1/6 = .166...')
print('P(B >= 0) = ', ps.P(('B', 0, None)), ' Exp: 1.0')
print('P(B < 0) = ', ps.P(('B', None, 0)), ' Exp: 0.0')
print('P(-inf <= B > inf) = ', ps.P(('B', None, None)), ' Exp: 1.0')
print('P(-1 <= B < 3) = ', ps.P(('B', -1, 3)), ' Exp: 1/3')
print('P(C = 2) =', ps.P(('C', 2)), ' Exp: 1/36 = .0277...')
print('P(C = 3) =', ps.P(('C', 3)), ' Exp: 1/18 = .055...')
print('P( 2 <= C < 4) = ', ps.P(('C', 2, 4)), ' Exp: 3/36 = .0833...')
print('P( 2 <= C < 4 | A = 1) = ', ps.P(('C', 2, 4), ('B', 1)),
' Exp: 1/3')
print('P( C = 7) = ', ps.P(('C', 7)), ' Exp: 1/6 = .166...')
print('P( C = 7 | A = 1, B = 6) = ', ps.P(('C', 7), [('A', 1), ('B', 6)]),
' Exp: 1.0')
print('P( C = 7 | A >= 2, B < 5) = ',
ps.P(('C', 7), [('A', 2, None), ('B', None, 5)]), ' Exp: 1/5 = .2')
print('P(-inf <= A < inf | B >= 1) = ',
ps.P(('A', None, None), ('B', 1, None)), ' Exp: 1.0')
print('P( A >= 3, B >= 3) = ', ps.P([('A', 3, None), ('B', 3, None)]),
'Exp: 4/9 (.444...)')
print('P( C = 7, A = 5) = ', ps.P([('C', 7), ('A', 5)]),
' Exp: 1/36 (.0277...)')
print('P( C = 7, A >= 5) = ', ps.P([('C', 7), ('A', 5, None)]),
' Exp: 1/18 (.0555...)')
print('P( A = 2 | B = 5, C= 7) = ', ps.P(('A', 2), [('B', 5), ('C', 7)]),
' Exp: 1.0')
print('P( B = 5, C= 7) = ', ps.P(('B', 5), ('C', 7)),
' Exp: 1/6 (.166...)')
print('P( A = 2, B = 5) = ', ps.P([('A', 2), ('B', 5)]),
' Exp: 1/36 (.0277...)')
print('P( A = 2, B = 5 | C = 7) = ', ps.P([('A', 2), ('B', 5)], ('C', 7)),
' Exp: 1/6 (.166...)')
print('P( A = 2, B = 5, N < 0| C = 7) = ',
ps.P([('A', 2), ('B', 5), ('N', None, 0)], ('C', 7)),
' Exp: 1/12 (.08333...)')
print('E( C | A = 1, B = 6) = ',
ps.distr('C', [('A', 1), ('B', 6)]).E(), ' Exp: 7')
print('E( C | A = 1, B >= 5) = ',
ps.distr('C', [('A', 1), ('B', 5, None)]).E(), ' Exp: 6')
print()
print('Testing continuous distributions. Using N = normal(0, 1)')
n = ps.distr('N')
mu1 = n.mean()
mu2 = n.stDev()
print('stats(N): mean, std, skew, kurtosis = ', mu1, mu2, n.skew(),
n.kurtosis(), 'Exp: (0, 1, 0, 0)')
print('P( -1 >= N > 1) = ', n.P((-1, 1)), 'Exp: .682')
print('P( -2 >= N > 2) = ', n.P((-2, 2)), 'Exp: .954')
print('P( -3 >= N > 3) = ', n.P((-3, 3)), 'Exp: .997')
print('P( -inf >= N > 0) = ', n.P((None, 0)), 'Exp: .5')
print('P( 0 >= N > inf) = ', n.P((0, None)), 'Exp: .5')
print('P( -inf >= N > inf) = ', n.P((None, None)), 'Exp: 1.0')
print('E( N2 | N = 1) = ', ps.distr('N2', ('N', 1)).E(), ' Exp: 2.0')
print('E( N2 | 1 <= N < 2) = ', ps.distr('N2', ('N', 1, 2)).E())
print()
print('Dependence testing. Note: values < .5 are considered independent')
print('A _||_ B = ', ps.dependence('A', 'B'), ' Exp: < .5')
print('A _||_ C = ', ps.dependence('A', 'C'), ' Exp: > .5')
print('B _||_ C = ', ps.dependence('B', 'C'), ' Exp: > .5')
print('N _||_ N2 = ', ps.dependence('N', 'N2'), ' Exp: > .5')
print('N _||_ C = ', ps.dependence('N', 'C'), ' Exp: < .5')
print('C _||_ N = ', ps.dependence('C', 'N'), ' Exp: < .5')
print('A _||_ B | C >= 8 = ', ps.dependence('A', 'B', [('C', 8, None)]),
' Exp: > .5')
print('A _||_ B | C < 7 = ', ps.dependence('A', 'B', [('C', None, 7)]),
' Exp: > .5')
print('A _||_ B | C = 7 = ', ps.dependence('A', 'B', [('C', 7)]),
' Exp: > .5')
print('A _||_ B | C = 6 = ', ps.dependence('A', 'B', [('C', 6)]),
' Exp: > .5')
print('A _||_ B | C = 5 = ', ps.dependence('A', 'B', [('C', 5)]),
' Exp: > .5')
print('A _||_ B | C = 4 = ', ps.dependence('A', 'B', [('C', 4)]),
' Exp: > .5')
print('A _||_ B | C = 3 = ', ps.dependence('A', 'B', [('C', 3)]),
' Exp: > .5')
print('A _||_ B | C = 2 = ', ps.dependence('A', 'B', [('C', 2)]),
' Exp: < .5')
print('A _||_ B | C = 12 = ', ps.dependence('A', 'B', [('C', 12)]),
' Exp: < .5')
print('A _||_ B | C = ', ps.dependence('A', 'B', ['C']), ' Exp: > .5')
print()
print('Independence testing (values > .5 are considered independent):')
print('A _||_ B = ', ps.independence('A', 'B'), ps.isIndependent('A', 'B'),
' Exp: > .5, True')
print('A _||_ C = ', ps.independence('A', 'C'), ps.isIndependent('A', 'C'),
' Exp: < .5, False')
print('A _||_ B | C = ', ps.independence('A', 'B', 'C'),
ps.isIndependent('A', 'B', 'C'), ' Exp: < .5, False')
print('A _||_ N = ', ps.independence('A', 'N'), ps.isIndependent('A', 'N'),
' Exp: > .5, True')
print()
print('Testing Conditionalization:')
ivaDist = ps.distr('IVA')
ivaMean = ivaDist.E()
ivaStd = ivaDist.stDev()
upper = ivaMean + .5 * ivaStd
lower = ivaMean - .5 * ivaStd
diff = upper - lower
pwr = 2
print('test interval = ', upper - lower)
ivcGupper = ps.E('IVC', ('IVA', upper), power=pwr)
print('E( IVC | IVA = upper)', ivcGupper)
ivcGlower = ps.E('IVC', ('IVA', lower), power=pwr)
print('E( IVC | IVA = upper)', ivcGupper)
print('E( IVC | IVA = lower)', ivcGlower)
ivcGupper = ps.E('IVC', [('IVA', upper), 'IVB'], power=pwr)
print('E( IVC | IVA = upper, IVB)', ivcGupper)
ivcGlower = ps.E('IVC', [('IVA', lower), 'IVB'], power=pwr)
print('E( IVC | IVA = lower, IVB)', ivcGlower)
print('ACE(A,C) = ', (ivcGupper - ivcGlower) / diff, ' Exp: ~ 0')
print()
print('Testing continuous causal dependence:')
print('IVB _||_ IVA = ', ps.dependence('IVB', 'IVA'), ' Exp: > .5')
print('IVA _||_ IVB = ', ps.dependence('IVA', 'IVB'), ' Exp: > .5')
print('IVB _||_ IVC = ', ps.dependence('IVB', 'IVC'), ' Exp: > .5')
print('IVA _||_ IVC = ', ps.dependence('IVA', 'IVC'), ' Exp: > .5')
print('IVA _||_ IVC | IVB = ', ps.dependence('IVA', 'IVC', 'IVB'),
' Exp: < .5')
print('IVA _||_ IVC | IVB, N = ',
ps.dependence('IVA', 'IVC', ['IVB', 'N']), ' Exp: < .5')
print()
print('Testing Bayesian Relationships:')
# P(C=7 | A=5) = P(A=5|C=7) * P(A=5) / P(C=7)
pA_C = ps.P(('A', 5), ('C', 7))
pA = ps.P(('A', 5))
pC = ps.P(('C', 7))
pC_A = ps.P(('C', 7), ('A', 5))
invpC_A = pA_C * pA / pC
err = abs(invpC_A - pC_A)
print(
'Inverse P(A=5 | C=7) vs measured (Bayes(P(A | C)), P(A | C), diff): ',
invpC_A, pC_A, err, ' Exp: ~ 0')
# P(0 <= IVB < 1 | 1 <= IVA < 2) = P(1 <= IVA < 2 | 0 <= IVB < 1) * P(0 <= IVB < 1) / P(1 <= IVA < 2)
pA_B = ps.P(('IVA', 1, 2), ('IVB', 0, 1))
pB = ps.P(('IVB', 0, 1))
pA = ps.P(('IVA', 1, 2))
pB_A = ps.P(('IVB', 0, 1), ('IVA', 1, 2))
invpB_A = pA_B * pB / pA
err = abs(invpB_A - pB_A)
print(
'Inverse P(0 <= IVB < 1 | 1 <= IVA < 2) vs measured (Bayes(P(IVB | IVA)), P(IVB | IVA), diff): ',
invpB_A, pB_A, err, ' Exp: ~ 0')
print()
print('Testing Prediction and Classification:')
testDat = {'A': [2, 3, 6], 'B': [5, 2, 6]}
predDat = ps.Predict('C', testDat)
for p in range(len(predDat)):
val = predDat[p]
a = testDat['A'][p]
b = testDat['B'][p]
print('Prediction(C) for A = ', a, ', B = ', b, ', = pred(C) = ', val,
' Exp:', a + b)
predDat = ps.Classify('C', testDat)
for p in range(len(predDat)):
val = predDat[p]
a = testDat['A'][p]
b = testDat['B'][p]
print('Classification(C) for A = ', a, ', B = ', b, ', = pred(C) = ',
val, ' Exp:', a + b)
testDat = {'N': [.5, 1, 1.5, 2, 2.5, 3], 'B': [1, 2, 3, 4, 5, 6]}
predDists = ps.PredictDist('N2', testDat)
for p in range(len(predDists)):
d = predDists[p]
n = testDat['N'][p]
b = testDat['B'][p]
print('Prediction(N2) for N = ', n, ', B = ', b,
', = pred(N2 (mean, std)) = ', d.E(), d.stDev(), ' Exp:', n + 1,
', 1')
print()
end = time.time()
duration = end - start
print('Test Time = ', round(duration))
def run(filename):
r = getData.DataReader(filename)
dat = r.read()
start = time.time()
# split data between 'training' and test
vars = list(dat.keys())
datLen = len(dat[vars[0]])
trainLen = datLen - 100
tr = {}
te = {}
for var in dat.keys():
datL = list(dat[var])
tr[var] = datL[:trainLen]
te[var] = datL[trainLen:]
#print('te = ',te.keys(), te)
print()
print('Testing probability module\'s prediction capabilities.')
ps = ProbSpace(tr, density=1, power=1)
print()
print('Testing non-linear regression with continuous variables.')
d = ps.distr('Y')
print('stats(Y) = ', d.mean(), d.stDev(), d.skew(), d.kurtosis())
# Note: Predict will automatically remove Y from the test data
Ymean = d.mean()
expected = te['Y']
results = ps.Predict('Y', te)
#print('results = ', results)
SSE = 0.0 # Sum of squared error
SST = 0.0 # Sum of squared deviation
for i in range(len(results)):
val = results[i]
exp = expected[i]
X = []
for x in ['X1', 'X2', 'X3']:
X.append(te[x][i])
#print('X = ', X, ', pred = ', val, ', expected = ', exp, ', err = ', val - exp)
SSE += (val - exp)**2
SST += (exp - Ymean)**2
print('R2 = ', 1 - SSE / SST)
print()
print('Testing Classification with discontinuous discrete data')
d = ps.distr('DY')
print('stats(DY) = ', d.minVal(), d.maxVal(), d.mean(), d.stDev(),
d.skew(), d.kurtosis())
expected = te['DY']
results = ps.Classify('DY', te)
#print('results = ', results)
cumErr = 0
for i in range(len(results)):
val = results[i]
exp = expected[i]
X = []
for x in ['DX1', 'DX2', 'DX3', 'DX4']:
X.append(te[x][i])
#print('X = ', X, ', pred = ', val, ', expected = ', exp, ', err = ', val != exp)
if val != exp:
cumErr += 1
print('Accuracy = ', 1 - (cumErr / len(results)))
end = time.time()
duration = end - start
print('Test Time = ', round(duration))