apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-06 18:42:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

[statespace] changed tests to check model integrity with GP model

Browse source
This commit is contained in:
Max Zwiessele 2016-04-05 10:18:49 +01:00
parent 13cf717231
commit 7984e17805
1 changed files with 160 additions and 152 deletions
Download patch file Download diff file Expand all files Collapse all files

78
GPy/testing/gpy_kernels_state_space_tests.py
View file

@ -20,7 +20,7 @@ class StateSpaceKernelsTests(np.testing.TestCase):
def run_for_model(self, X, Y, ss_kernel, kalman_filter_type = 'regular', def run_for_model(self, X, Y, ss_kernel, kalman_filter_type = 'regular',
use_cython=False, check_gradients=True, use_cython=False, check_gradients=True,
optimize=True, optimize_max_iters=1000,predict_X=None, optimize=True, optimize_max_iters=250, predict_X=None,
compare_with_GP=True, gp_kernel=None, compare_with_GP=True, gp_kernel=None,
mean_compare_decimal=10, var_compare_decimal=7): mean_compare_decimal=10, var_compare_decimal=7):
@ -31,34 +31,35 @@ class StateSpaceKernelsTests(np.testing.TestCase):
if check_gradients: if check_gradients:
self.assertTrue(m1.checkgrad()) self.assertTrue(m1.checkgrad())
if optimize: if 1:#optimize:
m1.optimize(optimizer='lbfgsb',max_iters=optimize_max_iters) m1.optimize(optimizer='lbfgsb',max_iters=2)
if compare_with_GP and (predict_X is None): if compare_with_GP and (predict_X is None):
predict_X = X predict_X = X
if (predict_X is not None): self.assertTrue(compare_with_GP)
x_pred_reg_1 = m1.predict(predict_X)
x_quant_reg_1 = m1.predict_quantiles(predict_X)
if compare_with_GP: if compare_with_GP:
np.random.seed(254856)
m2 = GPy.models.GPRegression(X,Y, gp_kernel) m2 = GPy.models.GPRegression(X,Y, gp_kernel)
m2.optimize(optimizer='lbfgsb', max_iters=optimize_max_iters) #m2.randomize()
#print(m2) m2.optimize(max_iters=optimize_max_iters)
m1[:] = m2[:]
if (predict_X is not None):
x_pred_reg_1 = m1.predict(predict_X)
x_quant_reg_1 = m1.predict_quantiles(predict_X)
x_pred_reg_2 = m2.predict(predict_X) x_pred_reg_2 = m2.predict(predict_X)
x_quant_reg_2 = m2.predict_quantiles(predict_X) x_quant_reg_2 = m2.predict_quantiles(predict_X)
# Test values np.testing.assert_array_almost_equal(x_pred_reg_1[0], x_pred_reg_2[0], 3)
#print np.max(np.abs(x_pred_reg_1[0]-x_pred_reg_2[0])) np.testing.assert_array_almost_equal(x_pred_reg_1[1], x_pred_reg_2[1], 3)
np.testing.assert_almost_equal(np.max(np.abs(x_pred_reg_1[0]- \ np.testing.assert_array_almost_equal(x_quant_reg_1[0], x_quant_reg_2[0], 3)
x_pred_reg_2[0])), 0, decimal=mean_compare_decimal) np.testing.assert_array_almost_equal(x_quant_reg_1[1], x_quant_reg_2[1], 3)
np.testing.assert_almost_equal(m1.log_likelihood(), m2.log_likelihood(), 3)
np.testing.assert_array_almost_equal(m1.gradient, m2.gradient, 2)
# Test variances
#print np.max(np.abs(x_pred_reg_1[1]-x_pred_reg_2[1]))
np.testing.assert_almost_equal(np.max(np.abs(x_pred_reg_1[1]- \
x_pred_reg_2[1])), 0, decimal=var_compare_decimal)
def test_Matern32_kernel(self,): def test_Matern32_kernel(self,):
np.random.seed(234) # seed the random number generator np.random.seed(234) # seed the random number generator
@ -73,7 +74,7 @@ class StateSpaceKernelsTests(np.testing.TestCase):
predict_X=X, predict_X=X,
compare_with_GP=True, compare_with_GP=True,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=10, var_compare_decimal=7) mean_compare_decimal=5, var_compare_decimal=5)
def test_Matern52_kernel(self,): def test_Matern52_kernel(self,):
np.random.seed(234) # seed the random number generator np.random.seed(234) # seed the random number generator
@ -87,7 +88,7 @@ class StateSpaceKernelsTests(np.testing.TestCase):
self.run_for_model(X, Y, ss_kernel, check_gradients=True, self.run_for_model(X, Y, ss_kernel, check_gradients=True,
optimize = True, predict_X=X, optimize = True, predict_X=X,
compare_with_GP=True, gp_kernel=gp_kernel, compare_with_GP=True, gp_kernel=gp_kernel,
mean_compare_decimal=8, var_compare_decimal=7) mean_compare_decimal=5, var_compare_decimal=5)
def test_RBF_kernel(self,): def test_RBF_kernel(self,):
np.random.seed(234) # seed the random number generator np.random.seed(234) # seed the random number generator
@ -95,13 +96,14 @@ class StateSpaceKernelsTests(np.testing.TestCase):
plot = False, points_num=50, x_interval = (0, 20), random=True) plot = False, points_num=50, x_interval = (0, 20), random=True)
X.shape = (X.shape[0],1); Y.shape = (Y.shape[0],1) X.shape = (X.shape[0],1); Y.shape = (Y.shape[0],1)
ss_kernel = GPy.kern.sde_RBF(1,active_dims=[0,]) ss_kernel = GPy.kern.sde_RBF(1, Y.var(), X.ptp()/2., active_dims=[0,])
gp_kernel = GPy.kern.RBF(1,active_dims=[0,]) gp_kernel = GPy.kern.RBF(1, Y.var(), X.ptp()/2., active_dims=[0,])
self.run_for_model(X, Y, ss_kernel, check_gradients=True, self.run_for_model(X, Y, ss_kernel, check_gradients=True,
predict_X=X, predict_X=X,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=1, var_compare_decimal=1) optimize_max_iters=1000,
mean_compare_decimal=2, var_compare_decimal=2)
def test_periodic_kernel(self,): def test_periodic_kernel(self,):
np.random.seed(322) # seed the random number generator np.random.seed(322) # seed the random number generator
@ -170,22 +172,25 @@ class StateSpaceKernelsTests(np.testing.TestCase):
self.run_for_model(X, Y, ss_kernel, check_gradients=True, self.run_for_model(X, Y, ss_kernel, check_gradients=True,
predict_X=X, predict_X=X,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=10, var_compare_decimal=7) mean_compare_decimal=5, var_compare_decimal=5)
def test_exponential_kernel(self,): def test_exponential_kernel(self,):
np.random.seed(234) # seed the random number generator np.random.seed(12345) # seed the random number generator
(X,Y) = generate_linear_data(x_points=None, tangent=1.0, add_term=20.0, noise_var=2.0, (X,Y) = generate_linear_data(x_points=None, tangent=1.0, add_term=20.0, noise_var=2.0,
plot = False, points_num=50, x_interval = (0, 20), random=True) plot = False, points_num=10, x_interval = (0, 20), random=True)
X.shape = (X.shape[0],1); Y.shape = (Y.shape[0],1) X.shape = (X.shape[0],1); Y.shape = (Y.shape[0],1)
ss_kernel = GPy.kern.sde_Exponential(1, active_dims=[0,]) ss_kernel = GPy.kern.sde_Exponential(1, Y.var(), X.ptp()/2., active_dims=[0,])
gp_kernel = GPy.kern.Exponential(1, active_dims=[0,]) gp_kernel = GPy.kern.Exponential(1, Y.var(), X.ptp()/2., active_dims=[0,])
Y -= Y.mean()
self.run_for_model(X, Y, ss_kernel, check_gradients=True, self.run_for_model(X, Y, ss_kernel, check_gradients=True,
predict_X=X, predict_X=X,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=5, var_compare_decimal=6) optimize_max_iters=1000,
mean_compare_decimal=2, var_compare_decimal=2)
def test_kernel_addition(self,): def test_kernel_addition(self,):
#np.random.seed(329) # seed the random number generator #np.random.seed(329) # seed the random number generator
@ -249,25 +254,27 @@ class StateSpaceKernelsTests(np.testing.TestCase):
return ss_kernel, gp_kernel return ss_kernel, gp_kernel
ss_kernel, gp_kernel = get_new_kernels() ss_kernel, gp_kernel = get_new_kernels()
#import ipdb;ipdb.set_trace()
self.run_for_model(X, Y, ss_kernel, kalman_filter_type = 'svd', self.run_for_model(X, Y, ss_kernel, kalman_filter_type = 'svd',
use_cython=True, optimize_max_iters=10, check_gradients=True, use_cython=True, optimize_max_iters=10, check_gradients=True,
predict_X=X, predict_X=X,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=-1, var_compare_decimal=-1) mean_compare_decimal=2, var_compare_decimal=2)
ss_kernel, gp_kernel = get_new_kernels() ss_kernel, gp_kernel = get_new_kernels()
self.run_for_model(X, Y, ss_kernel, kalman_filter_type = 'regular', self.run_for_model(X, Y, ss_kernel, kalman_filter_type = 'regular',
use_cython=False, optimize_max_iters=10, check_gradients=True, use_cython=False, optimize_max_iters=10, check_gradients=True,
predict_X=X, predict_X=X,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=-1, var_compare_decimal=-1) mean_compare_decimal=2, var_compare_decimal=2)
ss_kernel, gp_kernel = get_new_kernels() ss_kernel, gp_kernel = get_new_kernels()
self.run_for_model(X, Y, ss_kernel, kalman_filter_type = 'svd', self.run_for_model(X, Y, ss_kernel, kalman_filter_type = 'svd',
use_cython=False, optimize_max_iters=10, check_gradients=True, use_cython=False, optimize_max_iters=10, check_gradients=True,
predict_X=X, predict_X=X,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=-1, var_compare_decimal=0) mean_compare_decimal=2, var_compare_decimal=2)
def test_forecast(self,): def test_forecast(self,):
""" """
@ -317,21 +324,22 @@ class StateSpaceKernelsTests(np.testing.TestCase):
use_cython=False, optimize_max_iters=30, check_gradients=True, use_cython=False, optimize_max_iters=30, check_gradients=True,
predict_X=X_test, predict_X=X_test,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=0, var_compare_decimal=-1) mean_compare_decimal=2, var_compare_decimal=2)
ss_kernel, gp_kernel = get_new_kernels() ss_kernel, gp_kernel = get_new_kernels()
self.run_for_model(X_train, Y_train, ss_kernel, kalman_filter_type = 'svd', self.run_for_model(X_train, Y_train, ss_kernel, kalman_filter_type = 'svd',
use_cython=False, optimize_max_iters=30, check_gradients=False, use_cython=False, optimize_max_iters=30, check_gradients=False,
predict_X=X_test, predict_X=X_test,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=-1, var_compare_decimal=-1) mean_compare_decimal=2, var_compare_decimal=2)
ss_kernel, gp_kernel = get_new_kernels() ss_kernel, gp_kernel = get_new_kernels()
self.run_for_model(X_train, Y_train, ss_kernel, kalman_filter_type = 'svd', self.run_for_model(X_train, Y_train, ss_kernel, kalman_filter_type = 'svd',
use_cython=True, optimize_max_iters=30, check_gradients=False, use_cython=True, optimize_max_iters=30, check_gradients=False,
predict_X=X_test, predict_X=X_test,
gp_kernel=gp_kernel, gp_kernel=gp_kernel,
mean_compare_decimal=-1, var_compare_decimal=-1) mean_compare_decimal=2, var_compare_decimal=2)
if __name__ == "__main__": if __name__ == "__main__":
print("Running state-space inference tests...") print("Running state-space inference tests...")