12345qiupeng
/
gtsam
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

remove extraneous file

release/4.3a0
Varun Agrawal 2021-07-15 00:25:40 -04:00
parent 2dd22c64fd
commit 6db646d800
1 changed files with 0 additions and 81 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

81
gtsam/linear/Coreset.h
View File

@ -1,81 +0,0 @@
#include <gtsam/linear/LossFunctions.h>
#include <Eigen/unsupported/Eigen/CX11/Tensor>
namespace gtsam {
Vector FastCaratheodory(Matrix& P, Vector& weights, size_t coreset_size) {
size_t n = P.rows(), d = P.cols();
size_t m = 2 * d + 2;
if (n < d + 1) {
return weights;
}
Vector weights = weights / weights.sum();
size_t chunk_size = ceil(n / m);
size_t current_m = ceil(n / chunk_size);
size_t add_z = chunk_size - size_t(n % chunk_size);
Matrix u(weights.size(), 1);
u.col(0) = weights;
if (add_z != chunk_size) {
Matrix zeros = Matrix::Zero(add_z, d);
Matrix P_new = Matrix(P.rows() + zeros.rows(), P.cols() + zeros.cols());
P_new << P, zeros;
zeros = Matrix::Zero(add_z, u.cols());
Matrix u_new(u.rows() + zeros.rows(), u.cols() + zeros.cols());
u_new << u, zeros;
}
Vector idxarray = Vector::LinSpaced(n, 0, n - 1);
Eigen::Tensor p_groups;
// p_groups = P.reshape(current_m, chunk_size, P.shape[1])
// u_groups = u.reshape(current_m, chunk_size)
// idx_group = idxarray.reshape(current_m, chunk_size)
// u_nonzero = np.count_nonzero(u)
// if not coreset_size:
// coreset_size = d+1
// while u_nonzero > coreset_size:
// groups_means = np.einsum('ijk,ij->ik',p_groups, u_groups)
// group_weigts = np.ones(groups_means.shape[0], dtype =
// dtype)*1/current_m
// Cara_u_idx = Caratheodory(groups_means , group_weigts,dtype = dtype )
// IDX = np.nonzero(Cara_u_idx)
// new_P = p_groups[IDX].reshape(-1,d)
// subset_u = (current_m * u_groups[IDX] * Cara_u_idx[IDX][:,
// np.newaxis]).reshape(-1, 1) new_idx_array =
// idx_group[IDX].reshape(-1,1)
// ##############################################################################3
// u_nonzero = np.count_nonzero(subset_u)
// chunk_size = math.ceil(new_P.shape[0]/ m)
// current_m = math.ceil(new_P.shape[0]/ chunk_size)
// add_z = chunk_size - int(new_P.shape[0] % chunk_size)
// if add_z != chunk_size:
// new_P = np.concatenate((new_P, np.zeros((add_z, new_P.shape[1]),
// dtype = dtype))) subset_u = np.concatenate((subset_u,
// np.zeros((add_z, subset_u.shape[1]),dtype = dtype)))
// new_idx_array = np.concatenate((new_idx_array, np.zeros((add_z,
// new_idx_array.shape[1]),dtype = dtype)))
// p_groups = new_P.reshape(current_m, chunk_size, new_P.shape[1])
// u_groups = subset_u.reshape(current_m, chunk_size)
// idx_group = new_idx_array.reshape(current_m , chunk_size)
// ###########################################################
// new_u = np.zeros(n)
// subset_u = subset_u[(new_idx_array < n)]
// new_idx_array = new_idx_array[(new_idx_array <
// n)].reshape(-1).astype(int) new_u[new_idx_array] = subset_u return u_sum
// * new_u
}
} // namespace gtsam