def function_fitness_test():
func_errors = [(polynomial_sample, 1e-10), (sin_cos_sample, 2e-5)]
mls = MLS()
mls.set_diff_order(1)
interval = sample_interval()
v = np.linspace(*interval, num=4)
centers = [(x, y) for x in v for y in v]
coords = sample_coords()
rad = (interval[1] - interval[0]) / 50.0 * 4
pt_vals = np.ndarray((len(coords), 3))
radss = [(rad,)]
rads = np.ndarray((len(coords),), dtype=np.double)
rads.fill(rad)
rads += np.linspace(0.1, 0.2, len(rads))
radss.append(rads)
for rads in radss:
for func, err in func_errors:
acts = []
i = 0
for coord in coords:
val = func(coord)
pt_vals[i] = val
i += 1
for center in centers:
coords_in_rad = []
indes = []
i = 0
for coord in coords:
d = np.array(coord) - center
d = np.dot(d, d) ** 0.5
if d <= rad:
coords_in_rad.append(coord)
indes.append(i)
i += 1
if len(rads) > 1:
act_rads = rads[indes]
else:
act_rads = rads
res = mls.values(center, coords_in_rad , act_rads)
act_val = np.dot(res[:, 0], pt_vals[indes, 0])
act_val_x = np.dot(res[:, 1], pt_vals[indes, 0])
act_val_y = np.dot(res[:, 2], pt_vals[indes, 0])
acts = np.array((act_val, act_val_x, act_val_y), dtype=np.double)
exps = np.array(func(center), dtype=np.double)
max_err = np.max(np.abs(acts - exps))
ok_(max_err < err, str(func) + "acts:" + str(acts) + " exps:" + str(exps) + " center:" + str(center) + "max_err:" + str(max_err) + " err_lim:" + str(err))
def partition_of_unity_test():
mls = MLS()
mls.set_diff_order(1)
interval = sample_interval()
v = np.linspace(*interval, num=5)
centers = [(x, y) for x in v for y in v]
coords = sample_coords()
radss = [((interval[1] - interval[0]) / 50.0 * 4,)]
rads = np.ndarray((len(coords),), dtype=np.double)
rads.fill(radss[0][0])
rads += np.linspace(0.1, 0.2, len(rads))
radss.append(rads)
for rads in radss:
acts = []
for center in centers:
coords_in_rad = []
act_rads = []
i = 0
for coord in coords:
d = np.array(coord) - center
d = np.dot(d, d) ** 0.5
if len(rads) > 1:
rad = rads[i]
else:
rad = rads[0]
if d <= rad:
coords_in_rad.append(coord)
if len(rads) > 1:
act_rads.append(rad)
i += 1
if len(rads) == 1:
act_rads = rads
res = mls.values(center, coords_in_rad , act_rads)
acts.append(np.sum(res, axis=0))
acts = np.array(acts, dtype=np.double)
acts = abs(np.max(acts, axis=0))
ok_(acts[0] - 1 < 1e-9)
ok_(acts[1] < 1e-9)
ok_(acts[1] < 1e-9)
