be291b6a6ab8a663beac72fefa213bc7216a1617,geomstats/learning/frechet_mean.py,,_ball_gradient_descent,#Any#Any#Any#Any#Any#Any#,177

Before Change

        while convergence > tau and max_iter > iteration:

            iteration += 1
            grad_tangent = 2 * log(torch.from_numpy(gs.repeat(barycenter, len(points), axis = 0)),
                                   torch.from_numpy(points)) * torch.from_numpy(weights)
            grad_tangent = grad_tangent.data.numpy()
            cc_barycenter = exp(torch.from_numpy(barycenter),
                                torch.from_numpy(lr * grad_tangent.sum(0, keepdims=True)))
            cc_barycenter = cc_barycenter.data.numpy()
            convergence = metric.dist(cc_barycenter, barycenter).max().item()

After Change

            //grad_tangent = 2 * log(torch.from_numpy(rep_barycenter), torch.from_numpy(points))

            barycenter_gs  = gs.squeeze(barycenter)
            points_gs = gs.squeeze(points)

            grad_tangent = gs.zeros((len(points),len(barycenter_gs),len(points[0][0])))


            for j in range(len(points)):
                for i in range(len(barycenter_gs)):

                    grad_tangent[j][i] = 2*metric.log(points_gs[j][0], barycenter_gs[i] )

            grad_tangent = grad_tangent * weights

            //cc_barycenter_gs = metric.exp(barycenter_gs,
                                       //lr * grad_tangent.sum(0, keepdims=True))

Instances