|
| 1 | +import numpy as np |
| 2 | +import csv |
| 3 | +import os |
| 4 | +import matplotlib |
| 5 | +matplotlib.use('Agg') |
| 6 | +import matplotlib.pyplot as plt |
| 7 | + |
| 8 | +from pySDC.implementations.problem_classes.HeatEquation_1D_FD import heat1d |
| 9 | +from pySDC.implementations.problem_classes.AdvectionEquation_1D_FD import advection1d |
| 10 | +from pySDC.implementations.datatype_classes.mesh import mesh |
| 11 | +from pySDC.implementations.collocation_classes.gauss_radau_right import CollGaussRadau_Right |
| 12 | +from pySDC.implementations.sweeper_classes.generic_implicit import generic_implicit |
| 13 | +from pySDC.implementations.transfer_classes.TransferMesh import mesh_to_mesh |
| 14 | +from pySDC.implementations.controller_classes.allinclusive_multigrid_nonMPI import allinclusive_multigrid_nonMPI |
| 15 | + |
| 16 | +from pySDC.helpers.stats_helper import filter_stats, sort_stats |
| 17 | + |
| 18 | + |
| 19 | +def main(): |
| 20 | + """ |
| 21 | + Main driver running diffusion and advection tests |
| 22 | + """ |
| 23 | + QI = 'LU' |
| 24 | + # run_diffusion(QI=QI) |
| 25 | + # run_advection(QI=QI) |
| 26 | + |
| 27 | + QI = 'LU2' |
| 28 | + # run_diffusion(QI=QI) |
| 29 | + # run_advection(QI=QI) |
| 30 | + |
| 31 | + plot_results() |
| 32 | + |
| 33 | + |
| 34 | +def run_diffusion(QI): |
| 35 | + """ |
| 36 | + A simple test program to test PFASST convergence for the heat equation with random initial data |
| 37 | +
|
| 38 | + Args: |
| 39 | + QI: preconditioner |
| 40 | + """ |
| 41 | + |
| 42 | + # initialize level parameters |
| 43 | + level_params = dict() |
| 44 | + level_params['restol'] = 1E-08 |
| 45 | + level_params['nsweeps'] = [3, 1] |
| 46 | + |
| 47 | + # initialize sweeper parameters |
| 48 | + sweeper_params = dict() |
| 49 | + sweeper_params['collocation_class'] = CollGaussRadau_Right |
| 50 | + sweeper_params['num_nodes'] = [3] |
| 51 | + sweeper_params['QI'] = [QI, 'LU'] |
| 52 | + sweeper_params['spread'] = False |
| 53 | + |
| 54 | + # initialize problem parameters |
| 55 | + problem_params = dict() |
| 56 | + problem_params['nu'] = 0.1 # diffusion coefficient |
| 57 | + problem_params['freq'] = -1 # frequency for the test value |
| 58 | + problem_params['nvars'] = [127, 63] # number of degrees of freedom for each level |
| 59 | + |
| 60 | + # initialize step parameters |
| 61 | + step_params = dict() |
| 62 | + step_params['maxiter'] = 200 |
| 63 | + |
| 64 | + # initialize space transfer parameters |
| 65 | + space_transfer_params = dict() |
| 66 | + space_transfer_params['rorder'] = 2 |
| 67 | + space_transfer_params['iorder'] = 2 |
| 68 | + space_transfer_params['periodic'] = False |
| 69 | + |
| 70 | + # initialize controller parameters |
| 71 | + controller_params = dict() |
| 72 | + controller_params['logger_level'] = 30 |
| 73 | + controller_params['predict'] = False |
| 74 | + |
| 75 | + # fill description dictionary for easy step instantiation |
| 76 | + description = dict() |
| 77 | + description['problem_class'] = heat1d # pass problem class |
| 78 | + description['problem_params'] = problem_params # pass problem parameters |
| 79 | + description['dtype_u'] = mesh # pass data type for u |
| 80 | + description['dtype_f'] = mesh # pass data type for f |
| 81 | + description['sweeper_class'] = generic_implicit # pass sweeper (see part B) |
| 82 | + description['sweeper_params'] = sweeper_params # pass sweeper parameters |
| 83 | + description['step_params'] = step_params # pass step parameters |
| 84 | + description['space_transfer_class'] = mesh_to_mesh # pass spatial transfer class |
| 85 | + description['space_transfer_params'] = space_transfer_params # pass paramters for spatial transfer |
| 86 | + |
| 87 | + # set time parameters |
| 88 | + t0 = 0.0 |
| 89 | + Tend = 1.0 |
| 90 | + |
| 91 | + # set up number of parallel time-steps to run PFASST with |
| 92 | + |
| 93 | + fname = 'data/results_conv_diffusion_Linf_QI' + str(QI) + '.txt' |
| 94 | + file = open(fname, 'w') |
| 95 | + writer = csv.writer(file) |
| 96 | + writer.writerow(('num_proc', 'niter')) |
| 97 | + file.close() |
| 98 | + |
| 99 | + for i in range(0, 13): |
| 100 | + |
| 101 | + num_proc = 2 ** i |
| 102 | + level_params['dt'] = (Tend - t0) / num_proc |
| 103 | + description['level_params'] = level_params # pass level parameters |
| 104 | + |
| 105 | + out = 'Working on num_proc = %5i' % num_proc |
| 106 | + print(out) |
| 107 | + cfl = problem_params['nu'] * level_params['dt'] / (1.0 / (problem_params['nvars'][0] + 1)) ** 2 |
| 108 | + out = ' CFL number: %4.2e' % cfl |
| 109 | + print(out) |
| 110 | + |
| 111 | + # instantiate controller |
| 112 | + controller = allinclusive_multigrid_nonMPI(num_procs=num_proc, controller_params=controller_params, |
| 113 | + description=description) |
| 114 | + |
| 115 | + # get initial values on finest level |
| 116 | + P = controller.MS[0].levels[0].prob |
| 117 | + uinit = P.u_exact(t0) |
| 118 | + |
| 119 | + # call main function to get things done... |
| 120 | + uend, stats = controller.run(u0=uinit, t0=t0, Tend=Tend) |
| 121 | + |
| 122 | + # filter statistics by type (number of iterations) |
| 123 | + filtered_stats = filter_stats(stats, type='niter') |
| 124 | + |
| 125 | + # convert filtered statistics to list of iterations count, sorted by process |
| 126 | + iter_counts = sort_stats(filtered_stats, sortby='time') |
| 127 | + |
| 128 | + niters = np.array([item[1] for item in iter_counts]) |
| 129 | + |
| 130 | + out = ' Mean number of iterations: %4.2f' % np.mean(niters) |
| 131 | + print(out) |
| 132 | + |
| 133 | + file = open(fname, 'a') |
| 134 | + writer = csv.writer(file) |
| 135 | + writer.writerow((num_proc, np.mean(niters))) |
| 136 | + file.close() |
| 137 | + |
| 138 | + assert os.path.isfile(fname), 'ERROR: pickle did not create file' |
| 139 | + |
| 140 | + |
| 141 | +def run_advection(QI): |
| 142 | + """ |
| 143 | + A simple test program to test PFASST convergence for the periodic advection equation |
| 144 | +
|
| 145 | + Args: |
| 146 | + QI: preconditioner |
| 147 | + """ |
| 148 | + |
| 149 | + # initialize level parameters |
| 150 | + level_params = dict() |
| 151 | + level_params['restol'] = 1E-08 |
| 152 | + level_params['nsweeps'] = [3, 1] |
| 153 | + |
| 154 | + # initialize sweeper parameters |
| 155 | + sweeper_params = dict() |
| 156 | + sweeper_params['collocation_class'] = CollGaussRadau_Right |
| 157 | + sweeper_params['num_nodes'] = [3] |
| 158 | + sweeper_params['QI'] = [QI, 'LU'] # For the IMEX sweeper, the LU-trick can be activated for the implicit part |
| 159 | + sweeper_params['spread'] = False |
| 160 | + |
| 161 | + # initialize problem parameters |
| 162 | + problem_params = dict() |
| 163 | + problem_params['freq'] = 64 # frequency for the test value |
| 164 | + problem_params['nvars'] = [128, 64] # number of degrees of freedom for each level |
| 165 | + problem_params['order'] = 2 |
| 166 | + problem_params['type'] = 'center' |
| 167 | + problem_params['c'] = 0.1 |
| 168 | + |
| 169 | + # initialize step parameters |
| 170 | + step_params = dict() |
| 171 | + step_params['maxiter'] = 200 |
| 172 | + |
| 173 | + # initialize space transfer parameters |
| 174 | + space_transfer_params = dict() |
| 175 | + space_transfer_params['rorder'] = 2 |
| 176 | + space_transfer_params['iorder'] = 2 |
| 177 | + space_transfer_params['periodic'] = True |
| 178 | + |
| 179 | + # initialize controller parameters |
| 180 | + controller_params = dict() |
| 181 | + controller_params['logger_level'] = 30 |
| 182 | + controller_params['predict'] = False |
| 183 | + |
| 184 | + # fill description dictionary for easy step instantiation |
| 185 | + description = dict() |
| 186 | + description['problem_class'] = advection1d # pass problem class |
| 187 | + description['problem_params'] = problem_params # pass problem parameters |
| 188 | + description['dtype_u'] = mesh # pass data type for u |
| 189 | + description['dtype_f'] = mesh # pass data type for f |
| 190 | + description['sweeper_class'] = generic_implicit # pass sweeper (see part B) |
| 191 | + description['sweeper_params'] = sweeper_params # pass sweeper parameters |
| 192 | + description['step_params'] = step_params # pass step parameters |
| 193 | + description['space_transfer_class'] = mesh_to_mesh # pass spatial transfer class |
| 194 | + description['space_transfer_params'] = space_transfer_params # pass paramters for spatial transfer |
| 195 | + |
| 196 | + # set time parameters |
| 197 | + t0 = 0.0 |
| 198 | + Tend = 1.0 |
| 199 | + |
| 200 | + # set up number of parallel time-steps to run PFASST with |
| 201 | + |
| 202 | + fname = 'data/results_conv_advection_Linf_QI' + str(QI) + '.txt' |
| 203 | + file = open(fname, 'w') |
| 204 | + writer = csv.writer(file) |
| 205 | + writer.writerow(('num_proc', 'niter')) |
| 206 | + file.close() |
| 207 | + |
| 208 | + for i in range(0, 7): |
| 209 | + |
| 210 | + num_proc = 2 ** i |
| 211 | + level_params['dt'] = (Tend - t0) / num_proc |
| 212 | + description['level_params'] = level_params # pass level parameters |
| 213 | + |
| 214 | + out = 'Working on num_proc = %5i' % num_proc |
| 215 | + print(out) |
| 216 | + cfl = problem_params['c'] * level_params['dt'] / (1.0 / problem_params['nvars'][0]) |
| 217 | + out = ' CFL number: %4.2e' % cfl |
| 218 | + print(out) |
| 219 | + |
| 220 | + # instantiate controller |
| 221 | + controller = allinclusive_multigrid_nonMPI(num_procs=num_proc, controller_params=controller_params, |
| 222 | + description=description) |
| 223 | + |
| 224 | + # get initial values on finest level |
| 225 | + P = controller.MS[0].levels[0].prob |
| 226 | + uinit = P.u_exact(t0) |
| 227 | + |
| 228 | + # call main function to get things done... |
| 229 | + uend, stats = controller.run(u0=uinit, t0=t0, Tend=Tend) |
| 230 | + |
| 231 | + # filter statistics by type (number of iterations) |
| 232 | + filtered_stats = filter_stats(stats, type='niter') |
| 233 | + |
| 234 | + # convert filtered statistics to list of iterations count, sorted by process |
| 235 | + iter_counts = sort_stats(filtered_stats, sortby='time') |
| 236 | + |
| 237 | + niters = np.array([item[1] for item in iter_counts]) |
| 238 | + |
| 239 | + out = ' Mean number of iterations: %4.2f' % np.mean(niters) |
| 240 | + print(out) |
| 241 | + |
| 242 | + file = open(fname, 'a') |
| 243 | + writer = csv.writer(file) |
| 244 | + writer.writerow((num_proc, np.mean(niters))) |
| 245 | + file.close() |
| 246 | + |
| 247 | + assert os.path.isfile(fname), 'ERROR: pickle did not create file' |
| 248 | + |
| 249 | + |
| 250 | +def plot_results(cwd=''): |
| 251 | + """ |
| 252 | + Plotting routine for iteration counts |
| 253 | +
|
| 254 | + Args: |
| 255 | + cwd: current working directory |
| 256 | + """ |
| 257 | + |
| 258 | + setups = [('diffusion', 'LU', 'LU2'), ('advection', 'LU', 'LU2')] |
| 259 | + |
| 260 | + for type, QI1, QI2 in setups: |
| 261 | + |
| 262 | + fname = cwd + 'data/results_conv_' + type + '_Linf_QI' + QI1 + '.txt' |
| 263 | + file = open(fname, 'r') |
| 264 | + reader = csv.DictReader(file, delimiter=',') |
| 265 | + xvalues_1 = [] |
| 266 | + niter_1 = [] |
| 267 | + for row in reader: |
| 268 | + xvalues_1.append(int(row['num_proc'])) |
| 269 | + niter_1.append(float(row['niter'])) |
| 270 | + file.close() |
| 271 | + |
| 272 | + fname = cwd + 'data/results_conv_' + type + '_Linf_QI' + QI2 + '.txt' |
| 273 | + file = open(fname, 'r') |
| 274 | + reader = csv.DictReader(file, delimiter=',') |
| 275 | + xvalues_2 = [] |
| 276 | + niter_2 = [] |
| 277 | + for row in reader: |
| 278 | + xvalues_2.append(int(row['num_proc'])) |
| 279 | + niter_2.append(float(row['niter'])) |
| 280 | + file.close() |
| 281 | + |
| 282 | + # set up plotting parameters |
| 283 | + params = {'legend.fontsize': 20, |
| 284 | + 'figure.figsize': (12, 8), |
| 285 | + 'axes.labelsize': 20, |
| 286 | + 'axes.titlesize': 20, |
| 287 | + 'xtick.labelsize': 16, |
| 288 | + 'ytick.labelsize': 16, |
| 289 | + 'lines.linewidth': 3 |
| 290 | + } |
| 291 | + plt.rcParams.update(params) |
| 292 | + |
| 293 | + # set up figure |
| 294 | + plt.figure() |
| 295 | + plt.xlabel('number of time-steps (L)') |
| 296 | + plt.ylabel('#iterations') |
| 297 | + plt.xlim(min(xvalues_1 + xvalues_2) / 2.0, max(xvalues_1 + xvalues_2) * 2.0) |
| 298 | + plt.ylim(min(niter_1 + niter_2) - 1, max(niter_1 + niter_2) + 1) |
| 299 | + plt.grid() |
| 300 | + |
| 301 | + # plot |
| 302 | + plt.semilogx(xvalues_1, niter_1, 'r-', marker='s', markersize=10, label=QI1) |
| 303 | + plt.semilogx(xvalues_2, niter_2, 'b-', marker='o', markersize=10, label=QI2) |
| 304 | + |
| 305 | + plt.legend(loc=2, ncol=1, numpoints=1) |
| 306 | + |
| 307 | + # save plot, beautify |
| 308 | + fname = 'data/conv_test_niter_Linf_' + type + '.png' |
| 309 | + plt.savefig(fname, rasterized=True, bbox_inches='tight') |
| 310 | + |
| 311 | + assert os.path.isfile(fname), 'ERROR: plotting did not create file' |
| 312 | + |
| 313 | + |
| 314 | +if __name__ == "__main__": |
| 315 | + main() |
0 commit comments