def __init__(self, testName): self.testName = testName f = open(testName, 'r') exec(f.read(), globals()) self.gnodes = [] # 'model' is set when the text file is exec'ed for var in model: observed = True dType = 'Numeric' if type(var) == type((0, )): name, parents = var[:2] if len(var) >= 3: observed = var[2] if len(var) >= 4: dType = var[3] else: name = var parents = [] gnode = rv.rv(name, parents, observed, dType, None, None) self.gnodes.append(gnode) # For dat file, use the input file name with the .csv extension tokens = testName.split('.') testFileRoot = str.join('.', tokens[:-1]) datFileName = testFileRoot + '.csv' d = getData.DataReader(datFileName) self.data = d.read() self.g = cGraph.cGraph(self.gnodes, self.data) self.clusters = {} self.exos = []
def ref(x, choice): if choice == 1: return math.tanh(x) + math.sin(conditional) else: return math.sin(x) + math.tanh(conditional) def kcdf(x1, x2=0, kparms=[]): # CDF Kernel sigma = kparms[0] return (1 + erf(abs(x2 - x1) / sigma * sqrt(2))) / 2 path = '../models/Cprobdata2.csv' d = getData.DataReader(path) data = d.read() P = ProbSpace(data) #conditional X value conditional = 3 #choice 1=X(tanh) , 0=Y(sinh) (choose 1 to condition on X, 0 for Y) choice = 1 rtime1 = time.time() if choice == 1: FilterData, parentProb, finalQuery = P.filter([('X', conditional)]) else: FilterData, parentProb, finalQuery = P.filter([('Y', conditional)])
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' 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
return result def testFCDF(x): result = (logisticCDF(x, -2, .5) + logisticCDF(x, 2, .5) + logisticCDF(x, -.5, .3)) / 3 return result if __name__ == '__main__': args = sys.argv if (len(args) > 1): test = args[1] else: test = 'models/rkhsTest.csv' d = getData.DataReader(test) data = d.read() X = data['X'][:10000] tps = [] # Uniform test points testMin = -5 testMax = 5 tp = testMin numTP = 200 # Number of test points for graphing interval = (testMax - testMin) / numTP X2 = [] # Resampled X stdx = np.std(X) # Std Dev of X # Generate a uniform range of test points, from testMin to testMax. for i in range(numTP + 1): tps.append(tp) tp += interval sigma = 1 / log(len(X),
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))