def segment_packed_cov(segments, width=None):
# cov over time within each segment
if width is None:
width = segments.tag.shape[2]
n_segments = segments.shape[0]
packing_matrix = pack_symmatrix_spmat(width)
cov = custom_cov(segments).reshape((n_segments, width ** 2))
rval = theano.sparse.structured_dot(cov, packing_matrix)
if cov.type.broadcastable[0]:
# sparse matrices have no broadcastable property,
# so the broadcastable[0] is always False, even when it shouldn't be
rval = theano.tensor.Rebroadcast((0, True))(rval)
return rval
def segment_packed_cov(segments, width=None):
#cov over time within each segment
if width is None:
width = segments.tag.shape[2]
n_segments = segments.shape[0]
packing_matrix = pack_symmatrix_spmat(width)
cov = custom_cov(segments).reshape((n_segments, width**2))
rval = theano.sparse.structured_dot(cov, packing_matrix)
if cov.type.broadcastable[0]:
# sparse matrices have no broadcastable property,
# so the broadcastable[0] is always False, even when it shouldn't be
rval = theano.tensor.Rebroadcast((0, True))(rval)
return rval
def segment_cov(segments):
# cov over time within each segment
n_segments = segments.shape[0]
width = segments.shape[2]
return custom_cov(segments).reshape((n_segments, width ** 2))
def segment_cov(segments):
#cov over time within each segment
n_segments = segments.shape[0]
width = segments.shape[2]
return custom_cov(segments).reshape((n_segments, width**2))