import argparse import mitsuba as mi import drjit as dr mi.set_variant("cuda_ad_rgb") import numpy as np import sys from PDE2D.Solver import WostConstant, DataHolder from python2D.optimizations.eit_discrete.configurations import data from python2D.optimizations.eit_discrete.optimize import * from PDE2D.BoundaryShape import * import os sys.path.append("../../../") from PDE2D import PATH root_directory = f"{PATH}/output2D/optimizations/discrete-eit" def main(): parser = argparse.ArgumentParser(description='''Optimization Sphere''') parser.add_argument('--spe', default = 18, type=int) parser.add_argument('--objspe', default = 20, type=int) parser.add_argument('--objseed', default = 32, type=int) parser.add_argument('--seed', default = 543, type=int) parser.add_argument("--distder", default = "5e-2", type = float) parser.add_argument('--resnormalder', default = 128, type=int) parser.add_argument('--sppnormalder', default = 18, type=int) parser.add_argument('--confiter', default = 4, type=int) parser.add_argument('--iternum', default = 256, type=int) parser.add_argument("--lr", default = "0.025", type = float) parser.add_argument("--epsilon", default = "1e-5", type = float) parser.add_argument("--plot", action="store_true") parser.add_argument("--injectionset", default = "skip2", type = str) parser.add_argument("--visset", default = "none", type = str) parser.add_argument("--conf", default = 1, type = int) parser.add_argument("--measuretime", action = "store_true") parser.add_argument("--sdf", action = "store_true") parser.add_argument("--sdfres", type = int, default = 64) args = parser.parse_args() is_sdf = args.sdf e_shell = args.epsilon delete_injection = True #normalized_grad = args.normalizedgrad plot = args.plot seed_obj = args.objseed seed = args.seed dist_normalder = args.distder spp_normalder = 2 ** args.sppnormalder res_normalder = args.resnormalder spe_obj = 2 ** args.objspe spe = 2 ** args.spe num_iter = args.iternum learning_rate = args.lr conf = args.conf if is_sdf: sdf_begin = data["begin-sdf"].repeat(args.sdfres / 16, axis = 0).repeat(args.sdfres / 16, axis = 1) sdf_obj = data[f"obj-sdf{conf}"] in_boundary = SDFGrid(tensor_np= sdf_begin, box_length=2.1, box_center=[0,0], epsilon = args.epsilon, redistance = True, name = "inboundary") in_boundary_obj = SDFGrid(tensor_np= sdf_obj, box_length=2.1, box_center=[0,0], epsilon = args.epsilon, redistance = True, name = "inboundaryobj") opt_params = ["inboundary.dirichlet.tensor"] def postprocess(opt, radius, box_length, box_center, bbox, resolution): shape1 = CircleShape(radius = radius) shape1 = shape1.generate_sdf_grid(resolution =resolution, box_length = box_length, box_center = box_center, redistance = False) tensor = opt["inboundary.dirichlet.tensor"].numpy().squeeze() shape2 = SDFGrid(tensor_np=tensor, box_length= box_length , box_center = box_center, redistance = False) new_tensor = get_intersection_tensor(shape1, shape2, tensor.shape, bbox).squeeze() opt["inboundary.dirichlet.tensor"] = mi.TensorXf(new_tensor[...,np.newaxis]) post_process = lambda opt : postprocess(opt, 0.97, in_boundary.box_length, in_boundary.box_center, in_boundary.bbox, in_boundary.resolution) else: origin = data[f"origin-conf{conf}"] radius = data[f"radius-conf{conf}"] origin_obj = data[f"obj-origin-conf{conf}"] radius_obj = data[f"obj-radius-conf{conf}"] in_boundary_obj = CircleShape(origin_obj, radius_obj, name = "inboundaryobj") in_boundary = CircleShape(origin, radius, name = "inboundary") opt_params = ["inboundary.dirichlet.origin", "inboundary.dirichlet.radius"] def postprocess(opt, min_radius, max_region): opt["inboundary.dirichlet.radius"] = dr.select(opt["inboundary.dirichlet.radius"] < min_radius, min_radius, opt["inboundary.dirichlet.radius"]) origin_dir = dr.normalize(opt["inboundary.dirichlet.origin"]) diff = dr.norm(opt["inboundary.dirichlet.origin"]) + opt["inboundary.dirichlet.radius"] - max_region opt["inboundary.dirichlet.origin"] = dr.select(diff >= 0, (max_region - opt["inboundary.dirichlet.radius"]) * origin_dir, opt["inboundary.dirichlet.origin"]) post_process = lambda opt : postprocess(opt, 0.03 * out_radius, (1 - 0.0001) * out_radius) out_radius = 1 bbox_plot = [[-1.05 * out_radius, -1.05 * out_radius],[1.05 * out_radius, 1.05 * out_radius]] out_boundary = CircleWithElectrodes(radius = out_radius, injection_set = args.injectionset, is_delta = True) shape_obj = BoundaryWithDirichlets(out_boundary, [in_boundary_obj], dirichlet_values = [[0]], epsilon=e_shell) shape = BoundaryWithDirichlets(out_boundary, [in_boundary], dirichlet_values = [[0]], epsilon=e_shell) data_holder_obj = DataHolder(shape_obj) data_holder = DataHolder(shape) wos_obj = WostConstant(data_holder_obj, seed = args.objseed) wos = WostConstant(data_holder, opt_params = opt_params) num_conf = out_boundary.num_confs # The primal configurations that are going to be evaluated at each iteration for visualization. vis_set = [] if args.visset != "none": vis_set = out_boundary.get_injection_confs(args.injectionset, args.visset, num_electrodes=16) folder_name0 = "sdf" if is_sdf else "circle" folder_name1 = f"config{conf}-{args.injectionset}" folder_name2 = f"distder{dist_normalder}-confperiter{args.confiter}" folder_name3 = f"spe{args.spe}-normalspp{args.sppnormalder}" path = os.path.join(root_directory, folder_name0, folder_name1, folder_name2, folder_name3) path_obj = os.path.join(root_directory, "objectives", folder_name0, folder_name1) def create_path(path): isExist = os.path.exists(path) if not isExist: os.makedirs(path) create_path(path) create_path(path_obj) obj_results = [] for s in range(seed_obj): file = f"{s}.npy" file_el = f"elnums.npy" filepath = os.path.join(path_obj, file) filepath_el = os.path.join(path_obj, file_el) if not os.path.isfile(filepath): print(f"Generating objective results for seed {s}.") tensor, electrode_nums = compute_primals(wos_obj, spe_obj, s, delete_injection, confs_iter=16, num_electrodes=16, conf_numbers = [dr.opaque(UInt32, i, ) for i in range(num_conf)]) np.save(filepath, tensor) np.save(filepath_el, electrode_nums) obj_iter = np.load(filepath, allow_pickle = True) obj_results.append(obj_iter) obj_results = np.mean(np.array(obj_results), axis = 0) electrode_nums = np.load(filepath_el, allow_pickle=True) print("Objective Results are loaded.") wos.input.shape.out_boundary.voltages = np.array(obj_results) optimize_eit(path, wos, wos_obj, spe, spe, seed, args.confiter, res_normalder, spp_normalder, dist_normalder, num_iter, learning_rate, post_process, plot, bbox_plot = bbox_plot, delete_injection = delete_injection, measure_time = args.measuretime, vis_confs = vis_set, is_sdf = is_sdf) if __name__ == "__main__": main()