def plot_rbf_qr():
# in_mesh = mesh.GaussChebyshev_2D(12, 1, 4, 1)
in_mesh = mesh.GaussChebyshev_1D(12, 1, 4, 0) - 4
in_vals = func(in_mesh)
plot_mesh = np.linspace(np.min(in_mesh), np.max(in_mesh), 2000) # Use a fine mesh for plotting
shape_param = 1e-3
print("Proposed shape parameter =", shape_param)
rbf_qr = RBF_QR_1D(shape_param, in_mesh, in_vals)
fig1 = plt.figure("RBF-QR Interpolation")
ax1 = fig1.add_subplot(111)
ax1.set_title("Interpolation on N= " + str(len(in_mesh)) + " nodes with m=" + str(shape_param))
ax1.plot(in_mesh, in_vals, "d")
ax1.plot(plot_mesh, func(plot_mesh), "-", label="Original Function")
ax1.plot(plot_mesh, rbf_qr(plot_mesh), "--", label="Interpolation RBF_QR_1D, m=" + str(shape_param))
ax1.set_ylabel("y")
ax1.set_xlabel("x")
ax1.legend(loc=2)
ax1.grid()
fig2 = plt.figure("Error")
ax2 = fig2.add_subplot(111)
ax2.set_title("Error of RBF_QR on N=" + str(len(in_mesh)) + " nodes with m=" + str(shape_param))
ax2.plot(plot_mesh, np.abs(func(plot_mesh) - rbf_qr(plot_mesh)), "-", label="Error RBF_QR_1D")
ax2.plot(plot_mesh, np.full(plot_mesh.shape, np.finfo(np.float64).eps), "-", label="Machine precision (eps)")
ax2.set_yscale("log")
ax2.set_ylabel("Error")
ax2.set_xlabel("x")
ax2.legend(loc=2)
ax2.grid()
plt.show()
def combined():
shape_parameter_space = np.logspace(0, -10, num=50)
chebgauss_order_space = range(1, 40, 1)
X, Y = np.meshgrid(shape_parameter_space, chebgauss_order_space)
rmse = np.empty(X.shape)
for i, j in tqdm(list(np.ndindex(X.shape))):
shape = X[i, j]
order = Y[i, j]
in_mesh = mesh.GaussChebyshev_1D(order, 1, 4, 0)
in_vals = func(in_mesh)
test_mesh = np.linspace(np.min(in_mesh), np.max(in_mesh), 100)
rbf_qr = RBF_QR_1D(shape, in_mesh, in_vals)
rmse[i, j] = rbf_qr.RMSE(func, test_mesh)
fig = plt.figure()
ax = fig.gca(projection='3d')
ax.plot_surface(np.log10(X), 4 * Y, np.log10(rmse))
ax.set_xlabel("Shape parameter (log)")
ax.set_ylabel("Meshsize")
ax.set_zlabel("RMSE (log)")
plt.show()
def evalShapeQR():
in_mesh = mesh.GaussChebyshev_1D(4, 1, 4, 0)
in_vals = func(in_mesh)
test_mesh = np.linspace(np.min(in_mesh), np.max(in_mesh), 2000)
shape_parameter_space = np.logspace(0.8, 0, num=100)
rmse = []
for shape_param in shape_parameter_space:
print("m = ", shape_param)
rbf_qr = RBF_QR_1D(shape_param, in_mesh, in_vals)
rmse.append(rbf_qr.RMSE(func, test_mesh))
rmse = np.array(rmse)
plt.plot(shape_parameter_space, rmse, "-", label="RMSE")
plt.title("RMSE of RBF-QR for different epsilon, both axes log, mapping " + str(len(in_mesh))
+ " nodes to " + str(len(test_mesh)) + " nodes")
plt.ylabel("RMSE")
plt.xlabel("epsilon")
plt.yscale("log")
# plt.xscale("log") #mplot3d bug
plt.legend(loc=2)
plt.grid()
plt.show()
def plot_mesh_sizes():
GC = True
# mesh_sizes = np.arange(2, 64) if GC else np.linspace(4, 1024, 200)
mesh_sizes = np.arange(2, 32) if GC else np.linspace(4, 496, 200)
test_mesh = np.linspace(1, 4, 4000)
test_vals = func(test_mesh)
m = 15
separated = []
integrated = []
no_pol = []
for size in mesh_sizes:
in_mesh = mesh.GaussChebyshev_1D(size, 0.25, 4,
1) if GC else np.linspace(1, 4, size)
in_vals = func(in_mesh)
shape = rescaleBasisfunction(Gaussian, m, in_mesh)
print("Computing with in mesh size =", int(size),
", resulting shape parameter =", shape)
# bf = functools.partial(Gaussian, shape=shape)
bf = create_BFs(Gaussian, m, in_mesh)
separated.append(SeparatedConsistent(bf, in_mesh, in_vals))
integrated.append(IntegratedConsistent(bf, in_mesh, in_vals))
no_pol.append(NoneConsistent(bf, in_mesh, in_vals))
if GC: mesh_sizes = mesh_sizes * 4 * 4 # for GC
fig, ax1 = plt.subplots()
print("Computing RMSE")
ax1.semilogy(mesh_sizes, [i.RMSE(func, test_mesh) for i in separated],
"--",
label="RMSE separated Polynomial")
ax1.semilogy(mesh_sizes, [i.RMSE(func, test_mesh) for i in integrated],
"--",
label="RMSE integrated polynomial")
ax1.semilogy(mesh_sizes, [i.RMSE(func, test_mesh) for i in no_pol],
"--",
label="RMSE no polynomial")
ax1.set_ylabel("RMSE")
# ax1.legend(loc = 3)
ax2 = ax1.twinx()
print("Computing conditon")
ax2.semilogy(mesh_sizes, [i.condC for i in separated],
label="Condition separated / no polynomial")
ax2.semilogy(mesh_sizes, [i.condC for i in integrated],
label="Condition integrated polynomial")
ax2.set_ylabel("Condition Number")
# ax2.legend(loc = 2, framealpha = 1)
multi_legend(ax1, ax2, loc=7)
ax1.set_xlabel("Input Mesh Size")
# ax1.set_xlabel("Gauss-Chebyshev Order")
# plt.title("m = " + str(m))
rm = [i.RMSE(func, test_mesh) for i in separated]
plt.grid()
# plt.show()
plt.savefig("rc-gc-size.pdf")
def gc_m_order():
""" Keep number of elements constant, increase order, thus also points.
Different plots for different m's.
to have a comparision to equi-distant meshes """
ms = [2, 4, 6, 8]
gc_orders = np.arange(2, 24)
test_mesh = np.linspace(1, 4, 5000)
test_vals = func(test_mesh)
f, sp = plt.subplots(2, 2, sharex='col', sharey='row')
sp_lin = [sp[0][0], sp[0][1], sp[1][0], sp[1][1]]
for i, m in enumerate(ms):
print("Working on m =", m)
separated = []
integrated = []
no_pol = []
for gc_order in gc_orders:
in_mesh = mesh.GaussChebyshev_1D(gc_order, 1, 4, 1)
in_vals = func(in_mesh)
# shape = rescaleBasisfunction(Gaussian, m, in_mesh)
shape = -1
bf = create_BFs(Gaussian, m, in_mesh)
# bf = functools.partial(Gaussian, shape=shape)
separated.append(SeparatedConsistent(bf, in_mesh, in_vals))
integrated.append(IntegratedConsistent(bf, in_mesh, in_vals))
no_pol.append(NoneConsistent(bf, in_mesh, in_vals))
ax1 = sp_lin[i]
ax1.semilogy(gc_orders, [i.RMSE(func, test_mesh) for i in separated],
"--",
label="RMSE separated Polynomial")
ax1.semilogy(gc_orders, [i.RMSE(func, test_mesh) for i in integrated],
"--",
label="RMSE integrated polynomial")
ax1.semilogy(gc_orders, [i.RMSE(func, test_mesh) for i in no_pol],
"--",
label="RMSE no polynomial")
ax1.set_ylabel("RMSE")
# ax1.set_ylim(10e-6, 10e0)
ax1.legend(loc=3)
ax2 = ax1.twinx()
ax2.semilogy(gc_orders, [i.condC for i in separated],
label="Condition separated / no polynomial")
ax2.semilogy(gc_orders, [i.condC for i in integrated],
label="Condition integrated polynomial")
ax2.set_ylabel("Condition Number")
ax2.legend()
ax1.set_xlabel("Gauss-Chebyshev Order")
ax1.set_title(
"m = {} (shape parameter = {:.3f}), element size = 1, domain size = 4"
.format(m, shape))
ax1.grid()
plt.show()
""" Demo for conservative error metrics. """
import csv
import numpy as np, matplotlib.pyplot as plt, pandas as pd
import mesh, basisfunctions, rbf, testfunctions
testfunction = testfunctions.Highfreq()
in_mesh = mesh.GaussChebyshev_1D(order=12,
element_size=0.25,
domain_size=1,
domain_start=0)
# in_mesh = np.linspace(0, 1, 48)
out_mesh = np.linspace(np.min(in_mesh), np.max(in_mesh), 20)
# in_mesh = np.geomspace(1, 100, 90)
# in_mesh = in_mesh / 100
# out_mesh = np.linspace(0, 1, 80)
in_vals = testfunction(in_mesh)
m = 6
bf = basisfunctions.Gaussian(
basisfunctions.Gaussian.shape_param_from_m(m, in_mesh))
interp = rbf.SeparatedConservative(bf, in_mesh, in_vals)
one_interp = rbf.NoneConservative(bf,
in_mesh,
np.ones_like(in_vals),
rescale=False)
tf = lambda x: testfunctions.Highfreq()(x * 0.5 + 0.5)
# tf = func
BF = basisfunctions.Gaussian
test_mesh = np.linspace(-0.6, 0.6, 4000)
# test_mesh = np.linspace(0.2, 0.8, 4000)
test_vals = tf(test_mesh)
orders = np.arange(4, 128, 1)
# element_size = 0.0625
res = []
for order in orders:
in_mesh = mesh.GaussChebyshev_1D(order=order,
element_size=0.25,
domain_size=2,
domain_start=-1)
# in_mesh = np.polynomial.chebyshev.chebgauss(order)[0]
in_vals = tf(in_mesh)
shape_param = BF.shape_param_from_m(6, in_mesh)
bf = BF(shape_param)
interp = rbf.SeparatedConsistent(bf, in_mesh, in_vals)
rbfqr = rbf_qr.RBF_QR_1D(1e-5, in_mesh, in_vals)
out = interp(test_mesh)
outq = rbfqr(test_mesh)
res.append({
