def get_sig(m1, m2, n):
if m1 is None or m2 is None:
return 1.0
p1, p2 = len(m1.domain.attributes), len(m2.domain.attributes)
RSS1, RSS2 = m1.statistics["model summary"]["ExplVar"], m2.statistics["model summary"]["ExplVar"]
if RSS1<=RSS2 or p2<=p1 or n<=p2 or RSS2 <= 0:
return 1.0
F = ((RSS1 - RSS2)/(p2-p1))/(RSS2/(n-p2))
return statc.fprob(int(p2-p1),int(n-p2),F)
def get_sig(m1, m2, n):
if m1==None or m2==None:
return 1.0
p1, p2 = len(m1.domain.attributes), len(m2.domain.attributes)
RSS1, RSS2 = m1.statistics["model summary"]["ExplVar"], m2.statistics["model summary"]["ExplVar"]
if RSS1<=RSS2 or p2<=p1 or n<=p2 or RSS2 <= 0:
return 1.0
F = ((RSS1 - RSS2)/(p2-p1))/(RSS2/(n-p2))
return statc.fprob(int(p2-p1),int(n-p2),F)
def tubedRegression(cache, dimensions, progressCallback=None, **args):
if not cache.findNearest:
cache.findNearest = orange.FindNearestConstructor_BruteForce(
cache.data,
distanceConstructor=orange.ExamplesDistanceConstructor_Euclidean(),
includeSame=True)
if not cache.attrStat:
cache.attrStat = orange.DomainBasicAttrStat(cache.data)
normalizers = cache.findNearest.distance.normalizers
if progressCallback:
nExamples = len(cache.data)
nPoints = 100.0 / nExamples / len(dimensions)
effNeighbours = len(cache.contAttributes) > 1 and cache.nNeighbours or len(
cache.deltas)
for di, d in enumerate(dimensions):
contIdx = cache.contIndices[d]
minV, maxV = cache.attrStat[contIdx].min, cache.attrStat[contIdx].max
if minV == maxV:
continue
oldNormalizer = normalizers[cache.contIndices[d]]
normalizers[cache.contIndices[d]] = 0
for exi, ref_example in enumerate(cache.data):
if ref_example[contIdx].isSpecial():
cache.deltas[exi][d] = "?"
continue
ref_x = float(ref_example[contIdx])
Sx = Sy = Sxx = Syy = Sxy = n = 0.0
nn = cache.findNearest(ref_example, 0, True)
nn = [ex for ex in nn
if not ex[contIdx].isSpecial()][:effNeighbours]
mx = [abs(ex[contIdx] - ref_x) for ex in nn]
if not mx:
cache.deltas[exi][d] = "?"
continue
if max(mx) < 1e-10:
kw = math.log(.001)
else:
kw = math.log(.001) / max(mx)**2
for ex in nn[:effNeighbours]:
ex_x = float(ex[contIdx])
ex_y = float(ex.getclass())
w = math.exp(kw * (ex_x - ref_x)**2)
Sx += w * ex_x
Sy += w * ex_y
Sxx += w * ex_x**2
Syy += w * ex_y**2
Sxy += w * ex_x * ex_y
n += w
div = n * Sxx - Sx**2
if div: # and i<40:
b = (Sxy * n - Sx * Sy) / div
# div = Sx*Sy/n - Sxy
# if abs(div) < 1e-10:
# cache.errors[exi][d] = 1
# else:
# B = ((Syy - Sy**2/n) - (Sxx - Sx**2/n)) / 2 / div
#
# b_p = -B + math.sqrt(B**2+1)
# a = Sy/n - b_p * Sx/n
# error1 = 1/(1+b_p**2) * (Syy + a**2 + b_p**2*Sxx - 2*a*Sy + 2*a*b_p*Sx - 2*b_p*Sxy)
#
# b_2 = -B - math.sqrt(B**2+1)
# a = Sy/n - b_p * Sx/n
# error2 = 1/(1+b_p**2) * (Syy + a**2 + b_p**2*Sxx - 2*a*Sy + 2*a*b_p*Sx - 2*b_p*Sxy)
#
# if error1 < error2 and error1 >= 0:
# cache.errors[exi][d] = error1
# elif error2 >= 0:
# cache.errors[exi][d] = error2
# else:
# cache.errors[exi][d] = 42
# print error1, error2
a = (Sy - b * Sx) / n
err = (n * a**2 + b**2 * Sxx + Syy + 2 * a * b * Sx -
2 * a * Sy - 2 * b * Sxy)
tot = Syy - Sy**2 / n
mod = tot - err
merr = err / (n - 2)
if merr < 1e-10:
F = 0
Fprob = 1
else:
F = mod / merr
Fprob = statc.fprob(F, 1, int(n - 2))
cache.errors[exi][d] = Fprob
# print "%.4f" % Fprob,
#print ("%.3f\t" + "%.0f\t"*6 + "%f\t%f") % (w, ref_x, ex_x, n, a, b, merr, F, Fprob)
cache.deltas[exi][d] = b
else:
cache.deltas[exi][d] = "?"
if progressCallback:
progressCallback((nExamples * di + exi) * nPoints)
normalizers[cache.contIndices[d]] = oldNormalizer
def tubedRegression(cache, dimensions, progressCallback=None, **args):
if not cache.findNearest:
cache.findNearest = orange.FindNearestConstructor_BruteForce(
cache.data, distanceConstructor=orange.ExamplesDistanceConstructor_Euclidean(), includeSame=True
)
if not cache.attrStat:
cache.attrStat = orange.DomainBasicAttrStat(cache.data)
normalizers = cache.findNearest.distance.normalizers
if progressCallback:
nExamples = len(cache.data)
nPoints = 100.0 / nExamples / len(dimensions)
effNeighbours = len(cache.contAttributes) > 1 and cache.nNeighbours or len(cache.deltas)
for di, d in enumerate(dimensions):
contIdx = cache.contIndices[d]
minV, maxV = cache.attrStat[contIdx].min, cache.attrStat[contIdx].max
if minV == maxV:
continue
oldNormalizer = normalizers[cache.contIndices[d]]
normalizers[cache.contIndices[d]] = 0
for exi, ref_example in enumerate(cache.data):
if ref_example[contIdx].isSpecial():
cache.deltas[exi][d] = "?"
continue
ref_x = float(ref_example[contIdx])
Sx = Sy = Sxx = Syy = Sxy = n = 0.0
nn = cache.findNearest(ref_example, 0, True)
nn = [ex for ex in nn if not ex[contIdx].isSpecial()][:effNeighbours]
mx = [abs(ex[contIdx] - ref_x) for ex in nn]
if not mx:
cache.deltas[exi][d] = "?"
continue
if max(mx) < 1e-10:
kw = math.log(0.001)
else:
kw = math.log(0.001) / max(mx) ** 2
for ex in nn[:effNeighbours]:
ex_x = float(ex[contIdx])
ex_y = float(ex.getclass())
w = math.exp(kw * (ex_x - ref_x) ** 2)
Sx += w * ex_x
Sy += w * ex_y
Sxx += w * ex_x ** 2
Syy += w * ex_y ** 2
Sxy += w * ex_x * ex_y
n += w
div = n * Sxx - Sx ** 2
if div: # and i<40:
b = (Sxy * n - Sx * Sy) / div
# div = Sx*Sy/n - Sxy
# if abs(div) < 1e-10:
# cache.errors[exi][d] = 1
# else:
# B = ((Syy - Sy**2/n) - (Sxx - Sx**2/n)) / 2 / div
#
# b_p = -B + math.sqrt(B**2+1)
# a = Sy/n - b_p * Sx/n
# error1 = 1/(1+b_p**2) * (Syy + a**2 + b_p**2*Sxx - 2*a*Sy + 2*a*b_p*Sx - 2*b_p*Sxy)
#
# b_2 = -B - math.sqrt(B**2+1)
# a = Sy/n - b_p * Sx/n
# error2 = 1/(1+b_p**2) * (Syy + a**2 + b_p**2*Sxx - 2*a*Sy + 2*a*b_p*Sx - 2*b_p*Sxy)
#
# if error1 < error2 and error1 >= 0:
# cache.errors[exi][d] = error1
# elif error2 >= 0:
# cache.errors[exi][d] = error2
# else:
# cache.errors[exi][d] = 42
# print error1, error2
a = (Sy - b * Sx) / n
err = n * a ** 2 + b ** 2 * Sxx + Syy + 2 * a * b * Sx - 2 * a * Sy - 2 * b * Sxy
tot = Syy - Sy ** 2 / n
mod = tot - err
merr = err / (n - 2)
if merr < 1e-10:
F = 0
Fprob = 1
else:
F = mod / merr
Fprob = statc.fprob(F, 1, int(n - 2))
cache.errors[exi][d] = Fprob
# print "%.4f" % Fprob,
# print ("%.3f\t" + "%.0f\t"*6 + "%f\t%f") % (w, ref_x, ex_x, n, a, b, merr, F, Fprob)
cache.deltas[exi][d] = b
else:
cache.deltas[exi][d] = "?"
if progressCallback:
progressCallback((nExamples * di + exi) * nPoints)
normalizers[cache.contIndices[d]] = oldNormalizer
