First version of the irace API

eo/contrib/irace/fastga.cpp

@@ -0,0 +1,195 @@
+#include <iostream>
+#include <cstdlib>
+#include <string>
+
+#include <eo>
+#include <ga.h>
+#include <utils/checkpointing>
+#include <eoInt.h>
+#include <problems/eval/eoEvalIOH.h>
+
+#include <IOHprofiler_ecdf_logger.h>
+#include <f_w_model_one_max.hpp>
+
+// using Particle = eoRealParticle<eoMaximizingFitness>;
+using Ints = eoInt<eoMinimizingFitnessT<int>, size_t>;
+using Bits = eoBit<eoMinimizingFitnessT<int>, int>;
+
+// by enumerating candidate operators and their parameters.
+eoAlgoFoundryFastGA<Bits>& make_foundry(
+        eoFunctorStore& store,
+        eoInit<Bits>& init,
+        eoEvalFunc<Bits>& eval_onemax,
+        const size_t max_evals,
+        const size_t generations
+    )
+{
+    auto& foundry = store.pack< eoAlgoFoundryFastGA<Bits> >(init, eval_onemax, max_evals /*, max_restarts = max */);
+
+    /***** Continuators ****/
+    foundry.continuators.add< eoGenContinue<Bits> >(generations);
+    // for(size_t i=1; i<10; i++) {
+    //     foundry.continuators.add< eoGenContinue<Bits> >(i);
+    // }
+    // for(size_t i=10; i < 100; i+=2 ) {
+    //     foundry.continuators.add< eoSteadyFitContinue<Bits> >(10,i);
+    // }
+
+    for(double i=0.1; i<1.0; i+=0.1) {
+        foundry.crossover_rates.add<double>(i);
+        foundry.mutation_rates.add<double>(i);
+    }
+
+    for(size_t i=5; i<100; i+=10) {
+        foundry.pop_sizes.add<size_t>(i);
+    }
+
+    /***** Crossovers ****/
+    for(double i=0.1; i<0.9; i+=0.1) {
+        foundry.crossovers.add< eoUBitXover<Bits> >(i); // preference over 1
+    }
+    for(size_t i=1; i < 11; i+=1) {
+        foundry.crossovers.add< eoNPtsBitXover<Bits> >(i); // nb of points
+    }
+    foundry.crossovers.add< eo1PtBitXover<Bits> >();
+
+    /***** Mutations ****/
+    double p = 1.0; // Probability of flipping eath bit.
+    foundry.mutations.add< eoUniformBitMutation<Bits> >(p); // proba of flipping k bits, k drawn in uniform distrib
+    foundry.mutations.add< eoStandardBitMutation<Bits> >(p); // proba of flipping k bits, k drawn in binomial distrib
+    foundry.mutations.add< eoConditionalBitMutation<Bits> >(p); // proba of flipping k bits, k drawn in binomial distrib, minus zero
+    foundry.mutations.add< eoShiftedBitMutation<Bits> >(p); // proba of flipping k bits, k drawn in binomial distrib, changing zeros to one
+    foundry.mutations.add< eoNormalBitMutation<Bits> >(p); // proba of flipping k bits, k drawn in normal distrib
+    foundry.mutations.add< eoFastBitMutation<Bits> >(p); // proba of flipping k bits, k drawn in powerlaw distrib
+    for(size_t i=1; i < 11; i+=1) {
+        foundry.mutations.add< eoDetSingleBitFlip<Bits> >(i); // mutate k bits without duplicates
+    }
+
+    /***** Selectors *****/
+    foundry.selectors.add< eoRandomSelect<Bits> >();
+    foundry.selectors.add< eoSequentialSelect<Bits> >();
+    foundry.selectors.add< eoProportionalSelect<Bits> >();
+    for(size_t i=2; i < 10; i+=1) { // Tournament size.
+        foundry.selectors.add< eoDetTournamentSelect<Bits> >(i);
+    }
+    for(double i=0.51; i<0.91; i+=0.1) { // Tournament size as perc of pop.
+        foundry.selectors.add< eoStochTournamentSelect<Bits> >(i);
+    }
+
+    /***** Replacements ****/
+    foundry.replacements.add< eoPlusReplacement<Bits> >();
+    foundry.replacements.add< eoCommaReplacement<Bits> >();
+    foundry.replacements.add< eoSSGAWorseReplacement<Bits> >();
+    for(double i=0.51; i<0.91; i+=0.1) {
+        foundry.replacements.add< eoSSGAStochTournamentReplacement<Bits> >(i);
+    }
+    for(size_t i=2; i < 10; i+=1) {
+        foundry.replacements.add< eoSSGADetTournamentReplacement<Bits> >(i);
+    }
+
+    return foundry;
+}
+
+Bits::Fitness fake_func(const Bits&) { return 0; }
+
+
+int main(int argc, char* argv[])
+{
+    /***** Global parameters. *****/
+    enum { ERROR_USAGE = 100 };
+
+    const size_t dimension = 1000;
+    const size_t max_evals = 2 * dimension;
+    const size_t buckets = 100;
+
+    eoFunctorStore store;
+
+    /***** Parameters managed by the caller. *****/
+    assert(argc == 11);
+    const int pb_instance = std::atoi(argv[1]);
+
+    std::string s(argv[2]);
+    eo::rng.reseed(std::stoull(s));
+
+    Ints encoded_algo(8);
+    encoded_algo[0] = std::atoi(argv[3]); // continuator
+    encoded_algo[1] = std::atoi(argv[4]); // crossover_rate
+    encoded_algo[2] = std::atoi(argv[5]); // crossover
+    encoded_algo[3] = std::atoi(argv[6]); // mutation_rate
+    encoded_algo[4] = std::atoi(argv[7]); // mutation
+    encoded_algo[5] = std::atoi(argv[8]); // selection
+    encoded_algo[6] = std::atoi(argv[9]); // pop_size
+    encoded_algo[7] = std::atoi(argv[10]); // replacement
+
+    const size_t pop_size = encoded_algo[6];
+    const size_t generations = max_evals / pop_size;
+
+    eo::log << eo::setlevel(eo::warnings);
+
+    /***** IOH logger *****/
+    IOHprofiler_RangeLinear<size_t> target_range(0, dimension, buckets);
+    IOHprofiler_RangeLinear<size_t> budget_range(0, max_evals, buckets);
+    IOHprofiler_ecdf_logger<int, size_t, size_t> logger(target_range, budget_range);
+
+    logger.set_complete_flag(true);
+    logger.set_interval(0);
+    logger.activate_logger();
+
+    /***** IOH problem *****/
+    double w_model_suite_dummy_para = 0;
+    int w_model_suite_epitasis_para = 0;
+    int w_model_suite_neutrality_para = 0;
+    int w_model_suite_ruggedness_para = 0;
+
+    W_Model_OneMax w_model_om;
+    std::string problem_name = "OneMax";
+    problem_name = problem_name
+                    + "_D" + std::to_string((int)(w_model_suite_dummy_para * dimension))
+                    + "_E" + std::to_string(w_model_suite_epitasis_para)
+                    + "_N" + std::to_string(w_model_suite_neutrality_para)
+                    + "_R" + std::to_string(w_model_suite_ruggedness_para);
+
+    /// This must be called to configure the w-model to be tested.
+    w_model_om.set_w_setting(w_model_suite_dummy_para,w_model_suite_epitasis_para,
+                                    w_model_suite_neutrality_para,w_model_suite_ruggedness_para);
+
+    /// Set problem_name based on the configuration.
+    w_model_om.IOHprofiler_set_problem_name(problem_name);
+
+    /// Set problem_id as 1
+    w_model_om.IOHprofiler_set_problem_id(1);
+    w_model_om.IOHprofiler_set_instance_id(pb_instance);
+
+    /// Set dimension.
+    w_model_om.IOHprofiler_set_number_of_variables(dimension);
+
+    /***** Bindings *****/
+    logger.track_problem(w_model_om);
+
+    eoEvalIOHproblem<Bits> onemax_eval(w_model_om, logger);
+    eoPopLoopEval<Bits> pop_onemax(onemax_eval);
+
+    /***** Instanciate and run the algo *****/
+
+    eoUniformGenerator<int> ugen(0, 1);
+    eoInitFixedLength<Bits> onemax_init(/*bitstring size=*/dimension, ugen);
+    auto& foundry = make_foundry(store, onemax_init, onemax_eval, max_evals, generations);
+
+    size_t n = foundry.continuators.size() * foundry.crossovers.size() * foundry.mutations.size() * foundry.selectors.size() * foundry.replacements.size();
+    std::clog << n << " possible algorithms instances." << std::endl;
+
+    // Evaluation of a forged encoded_algo on the sub-problem
+    eoEvalFoundryFastGA<Ints, Bits> eval_foundry(
+            foundry, onemax_init, pop_onemax, /*penalization=*/ 0);
+
+    // Actually instanciate and run the algorithm.
+    eval_foundry(encoded_algo);
+
+    /***** IOH perf stats *****/
+    IOHprofiler_ecdf_sum ecdf_sum;
+    // iRace expects minimization
+    size_t perf = -1 * ecdf_sum(logger.data());
+
+    // Output
+    std::cout << perf << std::endl;
+}