def Run(self, W, x, eps, seed): x = x.flatten() prng = np.random.RandomState(seed) striped_mapping = mapper.Striped(self.domain, self.stripe_dim).mapping() x_sub_list = meta.SplitByPartition(striped_mapping).transform(x) Ms = [] ys = [] scale_factors = [] group_idx = sorted(set(striped_mapping)) for i in group_idx: x_i = x_sub_list[group_idx.index(i)] P_i = support.projection_matrix(striped_mapping, i) M_bar = selection.HB(x_i.shape).select() y_i = measurement.Laplace(M_bar, eps).measure(x_i, prng) noise_scale_factor = laplace_scale_factor(M_bar, eps) M_i = M_bar * P_i Ms.append(M_i) ys.append(y_i) scale_factors.append(noise_scale_factor) x_hat = inference.LeastSquares().infer(Ms, ys, scale_factors) return x_hat
def Run(self, W, x, eps, seed): assert len(x.shape) == 2, "Adaptive Grid only works for 2D domain" shape_2d = x.shape x = x.flatten() prng = np.random.RandomState(seed) Ms = [] ys = [] M = selection.UniformGrid(shape_2d, self.data_scale, eps, ag_flag=True, c=self.c).select() y = measurement.Laplace(M, self.alpha*eps).measure(x, prng) x_hat = inference.LeastSquares().infer(M, y) Ms.append(M) ys.append(y) # Prepare parition object for later SplitByParition. # This Partition selection operator is missing from Figure 2, plan 12 in the paper. uniform_mapping = mapper.UGridPartition(shape_2d, self.data_scale, eps, ag_flag=True, c=self.c).mapping() x_sub_list = meta.SplitByPartition(uniform_mapping).transform(x) sub_domains = support.get_subdomain_grid(uniform_mapping, shape_2d) ll, hi =[], [] for i in sorted(set(uniform_mapping)): x_i = x_sub_list[i] P_i = support.projection_matrix(uniform_mapping, i) x_hat_i = P_i * x_hat sub_domain_shape = sub_domains[i] M_i = selection.AdaptiveGrid(sub_domain_shape, x_hat_i, (1-self.alpha)*eps, c2=self.c2).select() y_i = measurement.Laplace(M_i, (1-self.alpha)*eps).measure(x_i, prng) offset = np.unravel_index(P_i.matrix.nonzero()[1][0], shape_2d) ll.extend(M_i._lower + np.array(offset)) hi.extend(M_i._higher + np.array(offset)) ys.append(y_i) Ms.append(workload.RangeQueries(shape_2d, np.array(ll), np.array(hi))) x_hat = inference.LeastSquares().infer(Ms, ys) return x_hat
def Run(self, W, x, eps, seed): assert len(x.shape) == 2, "Adaptive Grid only works for 2D domain" shape_2d = x.shape x = x.flatten() prng = np.random.RandomState(seed) Ms = [] ys = [] M = selection.UniformGrid(shape_2d, self.data_scale, eps, ag_flag=True, c=self.c).select() if not isinstance(M, np.ndarray): M = M.toarray() y = measurement.Laplace(M, self.alpha*eps).measure(x, prng) x_hat = inference.LeastSquares().infer(M, y) Ms.append(M) ys.append(y) # Prepare parition object for later SplitByParition. # This Partition selection operator is missing from Figure 2, plan 12 in the paper. uniform_mapping = mapper.UGridPartition(shape_2d, self.data_scale, eps, ag_flag=True, c=self.c).mapping() x_sub_list = meta.SplitByPartition(uniform_mapping).transform(x) sub_domains = support.get_subdomain_grid(uniform_mapping, shape_2d) for i in sorted(set(uniform_mapping)): x_i = x_sub_list[i] P_i = support.projection_matrix(uniform_mapping, i) x_hat_i = P_i * x_hat sub_domain_shape = sub_domains[i] M_i = selection.AdaptiveGrid(sub_domain_shape, x_hat_i, (1-self.alpha)*eps, c2=self.c2).select() if not isinstance(M, np.ndarray): M_i = M_i.toarray() y_i = measurement.Laplace(M_i, (1-self.alpha)*eps).measure(x_i, prng) M_i_o = M_i * P_i Ms.append(M_i_o) ys.append(y_i) x_hat = inference.LeastSquares().infer(Ms, ys, [1.0]*len(ys)) return x_hat
def Run(self, W, x, eps, seed): x = x.flatten() prng = np.random.RandomState(seed) striped_vectors = mapper.Striped(self.domain, self.stripe_dim).partitions() hd_vector = support.combine_all(striped_vectors) striped_mapping = hd_vector.flatten() x_sub_list = meta.SplitByPartition(striped_mapping).transform(x) Ms = [] ys = [] scale_factors = [] group_idx = sorted(set(striped_mapping)) # Given a group id on the full vector, recover the group id for each partition # put back in loop to save memory self.subgroups = {} for i in group_idx: selected_idx = np.where(hd_vector == i) ans = [p[i[0]] for p, i in zip(striped_vectors, selected_idx)] self.subgroups[i] = ans for i in group_idx: x_i = x_sub_list[group_idx.index(i)] # overwriting standard projection for efficiency W_i = self.project_workload(W, striped_vectors, hd_vector, i) dawa = pmapper.Dawa(eps, self.ratio, self.approx) mapping = dawa.mapping(x_i, prng) reducer = transformation.ReduceByPartition(mapping) x_bar = reducer.transform(x_i) W_bar = W_i * support.expansion_matrix(mapping) M_bar = selection.GreedyH(x_bar.shape, W_bar).select() if not isinstance(M_bar, np.ndarray): M_bar = M_bar.toarray() y_i = measurement.Laplace( M_bar, eps * (1 - self.ratio)).measure(x_bar, prng) noise_scale_factor = laplace_scale_factor( M_bar, eps * (1 - self.ratio)) # convert the measurement back to the original domain for inference P_i = support.projection_matrix(striped_mapping, i) M_i = (M_bar * support.reduction_matrix(mapping)) * P_i Ms.append(M_i) ys.append(y_i) scale_factors.append(noise_scale_factor) x_hat = inference.LeastSquares().infer(Ms, ys, scale_factors) return x_hat
def split_by_partition(self, mapping): private_node_ids = self.kernel_service.partition( self.private_node_id, mapping) data_nodes = self.data_manager.partition( meta.SplitByPartition(mapping), self.prng, self.data_node) return [ self._clone(private_node_id, data_node) for private_node_id, data_node in zip(private_node_ids, data_nodes) ]
def Run(self, W, x, eps, seed): x = x.flatten() prng = np.random.RandomState(seed) striped_mapping = mapper.Striped(self.domain, self.stripe_dim).mapping() x_sub_list = meta.SplitByPartition(striped_mapping).transform(x) Ms = [] ys = [] scale_factors = [] group_idx = sorted(set(striped_mapping)) W = get_matrix(W) for i in group_idx: x_i = x_sub_list[group_idx.index(i)] P_i = support.projection_matrix(striped_mapping, i) W_i = W * P_i.T dawa = pmapper.Dawa(eps, self.ratio, self.approx) mapping = dawa.mapping(x_i, prng) reducer = transformation.ReduceByPartition(mapping) x_bar = reducer.transform(x_i) W_bar = W_i * support.expansion_matrix(mapping) M_bar = selection.GreedyH(x_bar.shape, W_bar).select() if not isinstance(M_bar, np.ndarray): M_bar = M_bar.toarray() y_i = measurement.Laplace( M_bar, eps * (1 - self.ratio)).measure(x_bar, prng) noise_scale_factor = laplace_scale_factor( M_bar, eps * (1 - self.ratio)) M_i = (M_bar * support.reduction_matrix(mapping)) * P_i Ms.append(M_i) ys.append(y_i) scale_factors.append(noise_scale_factor) x_hat = inference.LeastSquares().infer(Ms, ys, scale_factors) return x_hat
def partition(self, node_id, mapping): node = self.private_manager.graph().id_node_map[node_id] part = meta.SplitByPartition(mapping) private_nodes = self.private_manager.partition(part, after=node) return [private_node.id for private_node in private_nodes]