def main():
from bootstrapping_olympics.extra.reprep import ReprepPublisher
N = 20
K = 2
# M = np.random.rand(N, N, K)
M = np.zeros((K, N, N))
for i, j in itertools.product(range(N), range(N)):
if i <= j:
M[:, i, j] = 0
else:
M[:, i, j] = np.random.rand(K)
M[:, :, :] = 0
M[0, 5, 10] = 1
M[1, 15, 10] = -1
bds = BDSEstimator2()
y_mean = np.random.rand(N)
A = np.random.randn(N, N)
T = 10000
Au = np.random.randn(K, K)
error_M = []
error_M2 = []
for t in range(T):
u = np.dot(Au, np.random.randn(K))
# if (T - 1) % 100 == 0:
# u = 0 * u
y = y_mean + np.dot(A, np.random.randn(N))
y_dot = bds_dynamics(M, y, u)
dt = 0.1
bds.update(y=y, y_dot=y_dot, u=u, dt=dt)
if t > T / 2:
bds.fits1.reset()
bds.fits2.reset()
if t % 100 == 0:
Mest = bds.get_M()
e = np.abs(M - Mest).mean()
error_M.append(e)
Mest2, M2info = bds.get_M2() #@UnusedVariable
e2 = np.abs(M - Mest2).mean()
error_M2.append(e2)
sys.stderr.write('%8d/%d %-10.5g %-10.5g \n' % (t, T, e, e2))
pass
rp = ReprepPublisher('bds_estimator_test_learn')
Mest = bds.get_M()
Merr = np.abs(M - Mest)
bds.publish(rp)
for i in range(K):
rp.array_as_image('M%s' % i, M[i, :, :])
for i in range(K):
rp.array_as_image('est_M%s' % i, Mest[i, :, :])
for i in range(K):
rp.array_as_image('Merr%s' % i, Merr[i, :, :])
with rp.plot('error') as pylab:
pylab.semilogy(np.array(error_M), 'x')
pylab.semilogy(np.array(error_M2), 'ro')
pylab.axis((0, len(error_M), 0, max(error_M) * 1.1))
rp.r.to_html('bds_estimator_test_learn.html')
