import argparse import numpy as np import os from python2D.optimizations.eit_variable.textures_eit import * from python2D.optimizations.eit_variable.optimize_eit import * from PDE2D.utils import * from PDE2D.BoundaryShape import * from PDE2D.Solver import * from PDE2D import PATH, GreenSampling, Split root_directory = os.path.join(PATH, "output2D", "optimizations", "variable-eit") def main(): parser = argparse.ArgumentParser(description='''Optimization Sphere''') parser.add_argument("--restensor", default = 24, type = int) parser.add_argument('--spe', default = 17, type=int) parser.add_argument('--dirichletspe', default = 23, type=int) parser.add_argument('--primalspe', default = 20, type=int) parser.add_argument('--objspe', default = 20, type=int) parser.add_argument('--seedobj', default = 16, type=int) parser.add_argument('--seed', default = 42, type=int) parser.add_argument('--confiter', default = 6, type=int) parser.add_argument('--iternum', default = 16, type=int) parser.add_argument("--lr", default = "0.1", type = float) parser.add_argument("--epsilon", default = "1e-3", type = float) parser.add_argument("--plot", action="store_true") parser.add_argument("--normalizedgrad", action="store_true") parser.add_argument("--split", default = "normal", type = str) parser.add_argument("--noaccel", action="store_true") parser.add_argument("--splitdepth", default=254, type=int) parser.add_argument("--computevariance", action = "store_true") parser.add_argument("--numdirichlet", default = 1, type = int) parser.add_argument("--dirichletoffset", default = 0, type = float) parser.add_argument("--dirichletradius", default = 0.001, type = float) parser.add_argument("--regL1", default = "0.0001", type = float) parser.add_argument("--regTV", default = "0.0001", type = float) parser.add_argument("--injectionset", default = "skip1-skip3-skip5-skip7", type = str) parser.add_argument("--visset", default = "none", type = str) parser.add_argument("--scaletexture", default = 1.0, type=float) parser.add_argument("--condthreshold", default = 1.2, type = float) parser.add_argument("--gradthreshold", default = 10, type = float) parser.add_argument("--mergedistance", default = 0.30, type = float) parser.add_argument("--conf", default = "1", type = str) parser.add_argument("--verbose", action = "store_true") parser.add_argument("--centeredsingle", action = "store_true") parser.add_argument("--kill", action = "store_true") parser.add_argument("--killstep", type = int, default = 150) parser.add_argument("--killrate", type = float, default = 0.99) parser.add_argument("--measuretime", action = "store_true") args = parser.parse_args() radius = 0.22 bbox = [[-1.1 * radius,-1.1 * radius],[1.1 * radius, 1.1 * radius]] compute_variance = args.computevariance max_dirichlet = args.numdirichlet centered_dirichlet = args.centeredsingle dirichlet_offset = args.dirichletoffset dirichlet_radius = args.dirichletradius use_accel = not args.noaccel split_depth = args.splitdepth e_shell = args.epsilon * radius delete_injection = True normalized_grad = args.normalizedgrad plot = args.plot seed_obj = args.seedobj seed = args.seed is_delta = True spe_obj = 2 ** args.objspe spe = 2 ** args.spe spe_primal = 2 ** args.primalspe spe_dirichlet = 2 ** args.dirichletspe num_electrodes = 16 conf_per_iter = args.confiter num_iter = args.iternum learning_rate = args.lr λ_L1 = args.regL1 λ_TV = args.regTV bg_conductance = 1 cond_threshold = args.condthreshold * bg_conductance grad_threshold = args.gradthreshold merge_distance = args.mergedistance res_tensor = args.restensor resolution_tensor = [res_tensor, res_tensor] if args.split == "none": split = Split.Naive elif args.split == "agressive": split = Split.Agressive elif args.split == "normal": split = Split.Normal else: raise Exception("No such split is defined.") conf_name = f"conf{args.conf}" normalized_name = "normalized" if normalized_grad else "unnormalized" if max_dirichlet == 0: dirichlet_name = "cn" else: dirichlet_name = "singlecentered" if centered_dirichlet else f"maxdirichlet{max_dirichlet}-merge{args.mergedistance}-spe{args.dirichletspe}" res_name = f"res{res_tensor}" spe_name = f"spe{args.primalspe}_{args.spe}" seed_name = f"seed{seed}" #e_name = f"epsilon{args.epsilon}" scale_name = f"scale{args.scaletexture}" reg_name = f"L1_{λ_L1}-TV_{λ_TV}" out_boundary = CircleWithElectrodes(radius = radius, num_electrodes=num_electrodes, is_delta = is_delta, injection_set = args.injectionset, centered = True) # 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=num_electrodes) num_conf = out_boundary.num_confs image_obj = objectives[conf_name] image_obj *= args.scaletexture image_obj += bg_conductance image = np.zeros(resolution_tensor) image[int(res_tensor * 3 / 8) : int(res_tensor * 5 / 8), int(res_tensor * 3 / 8) : int(res_tensor * 5 / 8)] = 0.1 image += bg_conductance kill_step = args.killstep if args.kill else dr.inf kill_rate = args.killrate grad_points = out_boundary.create_boundary_points(distance = 0, res = 512, spp = 1, discrete_points=True)[0] α_obj = TextureCoefficient("diffusion", bbox, image_obj, grad_zero_points=grad_points, out_val = bg_conductance) grad_points = out_boundary.create_boundary_points(distance = 0, res = 512, spp = 1, discrete_points=True)[0] α = TextureCoefficient("diffusion", bbox, image, grad_zero_points=grad_points, out_val = bg_conductance) opt_variable_name = "diffusion.texture.tensor" shape_obj = BoundaryWithDirichlets(out_boundary=out_boundary, dirichlet_boundaries=[], epsilon=e_shell, dirichlet_values = []) data_holder_obj = DataHolder(shape_obj, α = α_obj) if centered_dirichlet: obj_dir_point = np.array([[0,0]]) else: obj_dir_point = data_holder_obj.compute_high_conductance_points(max_num_points=1, cond_threshold=cond_threshold, grad_threshold=grad_threshold, merge_distance=merge_distance) data_holder_obj.shape.update_in_boundaries_circle(origins = obj_dir_point, radius = dirichlet_radius * radius, dirichlet_values = [dirichlet_offset]) wos_obj = WostVariable(data_holder_obj, green_sampling=GreenSampling.Polynomial, use_accelaration = use_accel) opt_params = [opt_variable_name] if max_dirichlet == 0: shape = BoundaryWithDirichlets(out_boundary=out_boundary, dirichlet_boundaries=[], epsilon = e_shell, dirichlet_values = []) else: shape = BoundaryWithDirichlets(out_boundary=out_boundary, dirichlet_boundaries=[CircleShape(origin = Point2f(0, 0), radius = dirichlet_radius * radius)], epsilon = e_shell, dirichlet_values = [dirichlet_offset]) data_holder = DataHolder(shape, α = α) wos = WostVariable(data_holder, green_sampling=GreenSampling.Polynomial, use_accelaration = use_accel, opt_params = opt_params) shape_dummy = BoundaryWithDirichlets(out_boundary=out_boundary, dirichlet_boundaries=[CircleShape(radius = dirichlet_radius * radius)], epsilon = e_shell, dirichlet_values = (np.ones([1, num_conf]) * dirichlet_offset).tolist()) data_holder_dummy = DataHolder(shape_dummy, α = α) wos_dummy = WostVariable(data_holder_dummy, green_sampling=GreenSampling.Polynomial, use_accelaration = use_accel, opt_params = opt_params) def postprocess(opt, min_val, max_val): opt[opt_variable_name] = dr.clip(opt[opt_variable_name], min_val, max_val) post_process = lambda opt : postprocess(opt, bg_conductance, 1.1 * np.max(image_obj)) folder0 = f"{conf_name}-{scale_name}" folder1 = f"{args.injectionset}" folder2 = dirichlet_name folder3 = f"{res_name}-{spe_name}-{seed_name}-{normalized_name}" folder4 = reg_name if args.kill: folder4 += f"-kill{kill_step}_{kill_rate}" path_obj = os.path.join(root_directory, "objectives", folder0, folder1) path = os.path.join(root_directory, folder0, folder1, folder2, folder3, folder4) print(path) create_path(path_obj) create_path(path) image_obj_ = plot_coeff(wos_obj.input.α, wos.input.shape, bbox, path, "objective", resolution = [256, 256]) max_range = [bg_conductance, 1.1 * np.max(image_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, std, electrode_nums = compute_primals(wos_obj, split, spe_obj, 0, s, delete_injection, split_depth, compute_variance, confs_iter = num_electrodes, num_electrodes=num_electrodes, conf_numbers = [dr.opaque(UInt32, i, ) for i in range(num_conf)]) np.save(filepath, tensor) np.save(filepath_el, electrode_nums) if compute_variance: filepath_std = os.path.join(path_obj, f"{s}_std.npy") np.save(filepath_std, std) 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) obj_results_std = np.zeros_like(obj_results) wos.input.shape.out_boundary.voltages_std = obj_results_std #obj_results, obj_results_std, electrode_nums = compute_primals(wos_obj, split, spe_obj, seed_obj, delete_injection, split_depth, compute_variance) #wos.input.shape.out_boundary.voltages = np.array(obj_results) #wos.input.shape.out_boundary.voltages_std = np.array(obj_results_std) if plot: iter_plot(wos_obj, bbox, path, "objective", obj_results, obj_results_std, electrode_nums, compute_std = False) optimize_eit(path = path, wos = wos, wos_obj = wos_obj, wos_dummy = wos_dummy, split = split, spe = spe, primal_spe = spe_primal, dirichlet_spe = spe_dirichlet, seed = seed, conf_per_iter = conf_per_iter, max_split_depth = split_depth, num_iter = num_iter, learning_rate = learning_rate, λ_L1 = λ_L1, λ_TV = λ_TV, post_process = post_process, cond_threshold = cond_threshold, grad_threshold = grad_threshold, max_dirichlet = max_dirichlet, dirichlet_radius = dirichlet_radius * radius, dirichlet_offset = dirichlet_offset, merge_distance = merge_distance, normalize_grad = normalized_grad, plot = plot, bbox_plot = bbox, delete_injection = delete_injection, compute_std = compute_variance, verbose = args.verbose, max_range = max_range, centered_dirichlet = centered_dirichlet, kill_step = kill_step, kill_rate = kill_rate, measure_time=args.measuretime, vis_confs=vis_set) if __name__ == "__main__": main()