def cdf(self, x):
out = 9/32.*(x+1)**2*(x<1/3.)
out += (3*x+1)/4.*(x>=1/3.)
return out
def bnd(self):
return -1,1
MyDist = pc.construct(
pdf=pdf, cdf=cdf, bnd=bnd)
#end
dist = MyDist()
print dist.mom(1, order=100)
# 0.277778240534
#end
print dist.mom(1, order=5, composite=1/3.)
# 0.277777777778
#end
mom = pc.momgen(
order=100,
domain=dist,
composite=1/3.)
dist.addattr(mom=mom)
print dist.mom([1,2,3])
# [ 0.27777778 0.2962963 0.15925926]
#end
E,V = np.empty((2,len(Z)))
N = 4*pc.terms(order, 2)
q = Q.sample(N, "H")
for i in xrange(len(Z)):
print "i", i
z = Z[i]
solver = Solver(z)
trans = lambda q: \
[z/q[0]*(z<q[3]) + \
(q[3]/q[0]+(z-q[3])/q[1])*(z>=q[3])*(z<q[4]) + \
(q[3]/q[0]+(q[4]-q[3])/q[1]+(z-q[4])/q[2])*(z>=q[4]),
q[3]/q[0]+(q[4]-q[3])/q[1]+(1-q[4])/q[2]]
dist = pc.Dist(_length=2)
dist._mom = pc.momgen(trans, 15, Q, rule="C",
composit=[.15,.9,.15, z, z])
orth = pc.orth_chol(order, dist, normed=0)
y = np.array(map(solver, q.T))
approx = pc.fitter_lr(orth, trans(q), y,
rule="T", order=1, alpha=1e-8)
E[i] = pc.E(approx, dist)
V[i] = pc.Var(approx, dist)
E = np.abs(E-E0)
V = np.abs(V-V0)
os.system('echo "%d\n%s\n%s" >> 3a.log' % (N, repr(E), repr(V)))
os.system('echo "%s, %s" >> 3a.log' % (np.mean(E), np.mean(V)))
N = 4 * pc.terms(order, 2)
q = Q.sample(N, "H")
for i in xrange(len(Z)):
print "i", i
z = Z[i]
solver = Solver(z)
trans = lambda q: \
[z/q[0]*(z<q[3]) + \
(q[3]/q[0]+(z-q[3])/q[1])*(z>=q[3])*(z<q[4]) + \
(q[3]/q[0]+(q[4]-q[3])/q[1]+(z-q[4])/q[2])*(z>=q[4]),
q[3]/q[0]+(q[4]-q[3])/q[1]+(1-q[4])/q[2]]
dist = pc.Dist(_length=2)
dist._mom = pc.momgen(trans,
15,
Q,
rule="C",
composit=[.15, .9, .15, z, z])
orth = pc.orth_chol(order, dist, normed=0)
y = np.array(map(solver, q.T))
approx = pc.fitter_lr(orth, trans(q), y, rule="T", order=1, alpha=1e-8)
E[i] = pc.E(approx, dist)
V[i] = pc.Var(approx, dist)
E = np.abs(E - E0)
V = np.abs(V - V0)
os.system('echo "%d\n%s\n%s" >> 3a.log' % (N, repr(E), repr(V)))
os.system('echo "%s, %s" >> 3a.log' % (np.mean(E), np.mean(V)))
