RBmatlab presentation 10.10.2011

Outline

1. (New) object-oriented RBmatlab order of commands
2. (Repetition of) high-dimensional problem implementation
3. New object-oriented Greedy algorithms:
   • Cached snapshots generator
   • Plugin based abstract greedy algorithm
   • Interface based extension algorithms
   • Tree type storage of generated data
4. Further improvements
   • (Fake) MPI Implementation for use with the zicluster
   • Classes for post-processing of generated data
   • Checkpointing
   • New Doxygen documentation with mtocpp
   • Tools for Job-Management

Examples

Order of commands

descr      = thermalblock_model;
dmodel     = gen_detailed_model(descr);
model_data = gen_model_data(dmodel);
sim_data   = dmodel.detailed_simulation(model_data);
plot_sim_data(dmodel, model_data, sim_data);

rmodel     = gen_reduced_model(dmodel);
detailed_data = gen_detailed_data(rmodel, model_data);
reduced_data  = gen_reduced_data(rmodel, detailed_data);
rb_sim_data   = rb_simulation(rmodel, reduced_data);
rb_sim_data   = rb_reconstruction(rmodel, detailed_data, rb_sim_data);
plot_sim_data(rmodel, model_data, rb_sim_data);

Snapshot generators

params.model_size = 'small';
descr = newton_oo_model(params);
descr.verbose = 10;
dmodel = gen_detailed_model(descr);
sg     = SnapshotsGenerator.Trajectories(dmodel, 'rb');
model_data = gen_model_data(dmodel);
U = sg.generate(dmodel, model_data);
U = sg.generate(dmodel, model_data);

descr.xnumintervals = 80;
dmodel = gen_detailed_model(descr);
sg     = SnapshotsGenerator.Trajectories(dmodel, 'rb');
model_data = gen_model_data(dmodel);
U = sg.generate(dmodel, model_data);


op_gen  = SnapshotsGenerator.SpaceOpEvals(dmodel, 'implicit', sg, descr.L_I_local_ptr);
LU = op_gen.generate(dmodel, model_data);

Greedy algorithms

Abstract greedy algorithm:

X-greedy

POD-greedy algorithm:

POD-greedy

EI-greedy algorithm:

PODEI-greedy algorithm:

EI greedy plugin

descr = minimal_ei_model
bg_descr.rb_problem_type = 'Test';
bg_descr.detailed_data_constructor = @Test.DetailedData;
bg_descr.reduced_data_constructor = @Test.ReducedData;

dmodel = LinEvol.DetailedModel(descr);
M_train               = ParameterSampling.Random(5);
rb_generator          = SnapshotsGenerator.Random(dmodel);
ei_gen1               = SnapshotsGenerator.SpaceOpEvals(dmodel, 'implicit', rb_generator, descr.L_I_local_ptr);
ei_plugin             = Greedy.Plugin.EI(ei_gen1);
ei_plugin.stop_Mmax   = 200;

ei_greedy             = Greedy.Algorithm(ei_plugin, M_train);
bg_descr.bg_algorithm = ei_greedy;
rmodel                = Test.ReducedModel(dmodel, bg_descr);

model_data            = gen_model_data(dmodel);
detailed_data         = gen_detailed_data(rmodel, model_data);

ei_gen2      = SnapshotsGenerator.SpaceOpEvals(dmodel, 'explicit', rb_generator, descr.L_I_local_ptr);

ei_gen_sum    = ei_gen1 + ei_gen2;
ei_plugin_sum = Greedy.Plugin.EI(ei_gen_sum);
ei_plugin_sum.stop_Mmax = 200;

ei_greedy2   = Greedy.Algorithm(ei_extension2, M_train);

bg_descr.bg_algorithm = ei_greedy2;
rmodel2        = Test.ReducedModel(dmodel, bg_descr);
detailed_data2 = gen_detailed_data(rmodel2, model_data);