def test_lwr_quadratic():
"""Test LWLR on 2nd order polynomial models of dimensions from 1 to 1.
(should pass most of the time)"""
result = True
for i in range(20):
a = random.random()
b = random.random()
c = random.random()
f = lambda x: [a*x[0]*x[0] + b*x[0] + c]
model = LWLRForwardModel(1, 1, 0.1)
robot = robots.fun.RobotFunction(f, 1, 1)
for i in range(100):
x = np.random.rand(1)
y = f(x)
model.add_xy(x, y)
tb = models.testbed.Testbed(robot, fmodel = model)
tb.uniform_motor(20)
errors = tb.run_forward()
result = result and sum(errors) < len(errors)*1e-3
# for i in xrange(10):
# x = np.random.rand(1)
# y = f(x)
# yp = model.predict_y(x)
# # import tools
# # print tools.dist(y, yp)
# result = result and np.allclose(y, yp, rtol = 1e-1, atol = 1e-1)
return result
def test_lwr_linear():
"""Test LWLR on random linear models of dimensions from 1 to 20.
It should return exact results, give of take floating point imprecisions."""
result = True
for i in range(20):
n = random.randint(1, 20)
m = random.randint(1, 5)
f = random_linear(n, m)
model = LWLRForwardModel(n, m, 1.0)
for i in range(2*n):
x = np.random.rand(n)
y = f(x)
model.add_xy(x, y)
for i in range(10):
x = np.random.rand(n).ravel()
y = f(x)
yp = model.predict_y(x)
#print 'y ', y
#print 'yp', yp
if not np.allclose(y, yp, rtol = 1e-10, atol = 1e-10):
print(n, m)
result = result and np.allclose(y, yp, rtol = 1e-10, atol = 1e-10)
return result
def test_lwr_linear_testbed():
"""Same test as previously, this time using testbed api."""
result = True
for i in range(20):
n = random.randint(1, 20)
m = random.randint(1, 20)
f = random_linear(n, m)
robot = robots.fun.RobotFunction(f, n, m)
model = LWLRForwardModel(n, m, 1.0)
for i in range(2*n):
x = np.random.rand(n)
y = f(x)
model.add_xy(x, y)
tb = models.testbed.Testbed(robot, fmodel = model)
tb.uniform_motor(10)
errors = tb.run_forward()
result = result and sum(errors) < len(errors)*1e-10
return result
def test_lwr1D_linear():
"""Simplest test possible (well, not quite, but close)."""
result = True
f = lambda x : 2.0*x
model = LWLRForwardModel(1, 1, k = 3, sigma = 0.1)
for i in range(10):
x = np.random.rand(1)
y = f(x)
#print x, y
model.add_xy(x, y)
for i in range(10):
x = np.random.rand(1).ravel()
y = f(x)
yp = model.predict_y(x)
#print 'y ', y
#print 'yp', yp
result = result and np.allclose(y, yp, rtol = 1e-5, atol = 1e-5)
return result
import numpy as np
from models.crossval import CrossValidation
from models.forward.lwr import LWLRForwardModel
model = LWLRForwardModel(1, 1, sigma = 1.0)
for i in np.arange(-100, 100, 0.1):
model.add_xy([i], [i*i])
for sigma in [0.0, 0.0001, 0.01, 0.02, 0.05, 0.1, 0.25, 0.5, 1.0, 1.5, 2.0, 5.0, 10.0, 1000.0]:
model.sigma = sigma
cv = CrossValidation.from_model(model)
print "%5.2f\t%7.4f" % (sigma, cv.k_folds(10))
