apunkt/GPy
1
0
Fork 0
mirror of https://github.com/SheffieldML/GPy.git synced 2026-05-09 03:52:39 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

Added a hack fix as suggested by max, zeroing any negative values (should really be numerically negative values on diagonal)

Browse source
This commit is contained in:
Alan Saul 2014-03-14 15:23:32 +00:00
parent a56928e11a
commit 4757265b24
3 changed files with 30 additions and 20 deletions
Download patch file Download diff file Expand all files Collapse all files

1
GPy/kern/_src/independent_outputs.py
View file

@ -40,6 +40,7 @@ class IndependentOutputs(CombinationKernel):
The index of the functions is given by the last column in the input X The index of the functions is given by the last column in the input X
the rest of the columns of X are passed to the underlying kernel for computation (in blocks). the rest of the columns of X are passed to the underlying kernel for computation (in blocks).
Kern is wrapped with a slicer metaclass
""" """
def __init__(self, kern, index_dim=-1, name='independ'): def __init__(self, kern, index_dim=-1, name='independ'):
assert isinstance(index_dim, int), "IndependentOutputs kernel is only defined with one input dimension being the indeces" assert isinstance(index_dim, int), "IndependentOutputs kernel is only defined with one input dimension being the indeces"

4
GPy/kern/_src/stationary.py
View file

@ -95,7 +95,9 @@ class Stationary(Kern):
#X2, = self._slice_X(X2) #X2, = self._slice_X(X2)
X1sq = np.sum(np.square(X),1) X1sq = np.sum(np.square(X),1)
X2sq = np.sum(np.square(X2),1) X2sq = np.sum(np.square(X2),1)
return np.sqrt(-2.*np.dot(X, X2.T) + (X1sq[:,None] + X2sq[None,:])) r2 = -2.*np.dot(X, X2.T) + X1sq[:,None] + X2sq[None,:]
r2[r2<0] = 0. # A bit hacky
return np.sqrt(r2)
@Cache_this(limit=5, ignore_args=()) @Cache_this(limit=5, ignore_args=())
def _scaled_dist(self, X, X2=None): def _scaled_dist(self, X, X2=None):

17
GPy/testing/kernel_tests.py
View file

@ -120,6 +120,8 @@ def check_kernel_gradient_functions(kern, X=None, X2=None, output_ind=None, verb
if verbose: if verbose:
print("Checking covariance function is positive definite.") print("Checking covariance function is positive definite.")
#if isinstance(kern, GPy.kern.IndependentOutputs):
#import ipdb; ipdb.set_trace() # XXX BREAKPOINT
result = Kern_check_model(kern, X=X).is_positive_semi_definite() result = Kern_check_model(kern, X=X).is_positive_semi_definite()
if result and verbose: if result and verbose:
print("Check passed.") print("Check passed.")
@ -306,17 +308,22 @@ class KernelTestsNonContinuous(unittest.TestCase):
D = self.D D = self.D
self.X = np.random.randn(N,D) self.X = np.random.randn(N,D)
self.X2 = np.random.randn(N1,D) self.X2 = np.random.randn(N1,D)
self.X_block = np.zeros((N+N1, D+D+1)) #self.X_block = np.zeros((N+N1, D+D+1))
#self.X_block[0:N, 0:D] = self.X
#self.X_block[N:N+N1, D:D+D] = self.X2
#self.X_block[0:N, -1] = 0
#self.X_block[N:N+N1, -1] = 1
self.X_block = np.zeros((N+N1, D+1))
self.X_block[0:N, 0:D] = self.X self.X_block[0:N, 0:D] = self.X
self.X_block[N:N+N1, D:D+D] = self.X2 self.X_block[N:N+N1, 0:D] = self.X2
self.X_block[0:N, -1] = 1 self.X_block[0:N, -1] = 0
self.X_block[N:N+1, -1] = 2 self.X_block[N:N+N1, -1] = 1
self.X_block = self.X_block[self.X_block.argsort(0)[:, -1], :] self.X_block = self.X_block[self.X_block.argsort(0)[:, -1], :]
def test_IndependentOutputs(self): def test_IndependentOutputs(self):
k = GPy.kern.RBF(self.D) k = GPy.kern.RBF(self.D)
kern = GPy.kern.IndependentOutputs(k, -1) kern = GPy.kern.IndependentOutputs(k, -1)
self.assertTrue(check_kernel_gradient_functions(kern, X=self.X_block, X2=self.X_block, verbose=verbose)) self.assertTrue(check_kernel_gradient_functions(kern, X=self.X_block, verbose=verbose))
if __name__ == "__main__": if __name__ == "__main__":
print "Running unit tests, please be (very) patient..." print "Running unit tests, please be (very) patient..."