Reorganize multiwinner directory structure; extract out PR measures.

CMakeLists.txt

@@ -189,25 +189,27 @@ add_library(qe_singlewinner_methods
 target_link_libraries(qe_singlewinner_methods ${GLPK_LIBRARIES})
 
 add_library(qe_multiwinner_methods
-	src/multiwinner/auction.cc
-	src/multiwinner/exhaustive/scored_method.cc
 	src/multiwinner/helper/errors.cc
-	src/multiwinner/compat_qbuck.cc
-	src/multiwinner/dhwl.cc
-	src/multiwinner/dhwl_mat.cc
-	src/multiwinner/meek_stv.cc
-	src/multiwinner/methods.cc
-	src/multiwinner/prbucklin.cc
-	src/multiwinner/psc.cc
-	src/multiwinner/quotas.cc
-	src/multiwinner/qbuck.cc
-	src/multiwinner/qpq.cc
-	src/multiwinner/qrange_stv.cc
-	src/multiwinner/randballots.cc
-	src/multiwinner/range_stv.cc
-	src/multiwinner/rusty/auxiliary/dsc.cc
-	src/multiwinner/shuntsstv.cc
-	src/multiwinner/stv.cc
+	src/multiwinner/methods/auction.cc
+	src/multiwinner/methods/exhaustive/scored_method.cc
+	src/multiwinner/methods/compat_qbuck.cc
+	src/multiwinner/methods/dhwl.cc
+	src/multiwinner/methods/dhwl_mat.cc
+	src/multiwinner/methods/meek_stv.cc
+	src/multiwinner/methods/methods.cc
+	src/multiwinner/methods/prbucklin.cc
+	src/multiwinner/methods/psc.cc
+	src/multiwinner/methods/quotas.cc
+	src/multiwinner/methods/qbuck.cc
+	src/multiwinner/methods/qpq.cc
+	src/multiwinner/methods/qrange_stv.cc
+	src/multiwinner/methods/randballots.cc
+	src/multiwinner/methods/range_stv.cc
+	src/multiwinner/methods/rusty/auxiliary/dsc.cc
+	src/multiwinner/methods/shuntsstv.cc
+	src/multiwinner/methods/stv.cc
+	src/multiwinner/pr_measures/clustering.cc
+	src/multiwinner/pr_measures/normal_fit.cc
 	src/stats/multiwinner/convex_hull.cc
 	src/stats/multiwinner/mwstats.cc)
 

src/main/multiwinner.cc

@@ -1,226 +1,15 @@
 // More multiwinner hacks/clustering tests.
 
-#include "multiwinner/methods.h"
-#include "multiwinner/qpq.h"
-
+#include "multiwinner/methods/all.h"
 #include "generator/spatial/all.h"
+
 #include "random/random.h"
 #include "tools/tools.h"
 
 #include "multiwinner/helper/errors.h"
 
-// TODO: Make error type parametric, or just inherit the function/
-// template it or something.
-
-class proportionality_measure {
-	public:
-		virtual void prepare(const positions_election & p_e) = 0;
-		virtual double get_error(const std::list<size_t> & outcome) = 0;
-};
-
-// Clustering-based proportionality measure
-
-class cluster_proportionality : public proportionality_measure {
-	private:
-		size_t num_clusters;
-		std::vector<std::vector<double> > clusters;
-		std::vector<size_t> cluster_for_candidate;
-
-		// Fraction of voters covered by each cluster.
-		std::vector<double> cluster_voter_proportions;
-
-		// ... and fractions of winning candidates.
-		std::vector<double> cluster_winner_proportions;
-
-		size_t get_optimum_cluster(const std::vector<double> & point,
-			const std::vector<std::vector<double> > & centers) const;
-
-	public:
-		void set_num_clusters(size_t num_clusters_in) {
-			num_clusters = num_clusters_in;
-			clusters.resize(num_clusters);
-			cluster_voter_proportions.resize(num_clusters);
-			cluster_winner_proportions.resize(num_clusters);
-
-		}
-		cluster_proportionality(size_t num_clusters_in) {
-			set_num_clusters(num_clusters_in);
-		}
-
-		void prepare(const positions_election & p_e);
-		double get_error(const std::list<size_t> & outcome);
-};
-
-size_t cluster_proportionality::get_optimum_cluster(
-	const std::vector<double> & point,
-	const std::vector<std::vector<double> > & centers) const {
-
-	size_t record_cluster = 0;
-	double record_distance = -1;
-	bool set_record = false;
-
-	for (size_t cluster_idx = 0; cluster_idx < centers.size();
-		++cluster_idx) {
-
-		double current_distance = lp_distance(2, point,
-				centers[cluster_idx]);
-
-		if (!set_record || current_distance < record_distance) {
-			set_record = true;
-			record_cluster = cluster_idx;
-			record_distance = current_distance;
-		}
-	}
-
-	return record_cluster;
-}
-
-void cluster_proportionality::prepare(const positions_election & p_e) {
-
-	size_t num_voters = p_e.voters_pos.size(),
-		   num_candidates = p_e.candidates_pos.size();
-
-	// Pick the k first voter coordinates as the cluster centers.
-	if (num_voters < num_clusters) {
-		throw std::invalid_argument("Clustering proportionality:"
-			" too few voters - need to be more than the number of clusters.");
-	}
-
-	clusters = std::vector<std::vector<double> >(
-			p_e.voters_pos.begin(), p_e.voters_pos.begin()+num_clusters);
-
-	std::fill(cluster_voter_proportions.begin(),
-		cluster_voter_proportions.end(), 0);
-
-	for (size_t voter = 0; voter < num_voters; ++voter) {
-		size_t opt = get_optimum_cluster(
-				p_e.voters_pos[voter], clusters);
-		cluster_voter_proportions[opt] += 1.0/num_voters;
-	}
-
-	cluster_for_candidate = std::vector<size_t>(num_candidates, 0);
-
-	for (size_t cand = 0; cand < num_candidates; ++cand) {
-		cluster_for_candidate[cand] = get_optimum_cluster(
-				p_e.candidates_pos[cand], clusters);
-	}
-}
-
-double cluster_proportionality::get_error(
-	const std::list<size_t> & outcome) {
-
-	size_t num_seats = outcome.size();
-
-	std::fill(cluster_winner_proportions.begin(),
-		cluster_winner_proportions.end(), 0);
-
-	for (size_t winner: outcome) {
-		cluster_winner_proportions[cluster_for_candidate[winner]]++;
-	}
-
-	for (double & cluster_prop: cluster_winner_proportions) {
-		cluster_prop /= (double)num_seats;
-	}
-
-	return sli(cluster_winner_proportions,
-			cluster_voter_proportions);
-}
-
-// Check the error between a normal distribution fitted to the winners
-// and the normal reference distribution.
-
-class normal_proportionality : public proportionality_measure {
-	private:
-		gaussian_generator reference;
-		std::vector<std::vector<double> > candidate_positions;
-		rng rnd;
-
-	public:
-		// TODO, use a different seed
-		normal_proportionality(gaussian_generator ref_in) : rnd(1) {
-			reference = ref_in;
-		}
-
-		void prepare(const positions_election & p_e);
-		double get_error(const std::list<size_t> & outcome);
-};
-
-void normal_proportionality::prepare(const positions_election & p_e) {
-	candidate_positions = p_e.candidates_pos;
-}
-
-double normal_proportionality::get_error(
-	const std::list<size_t> & outcome) {
-
-	size_t num_seats = outcome.size();
-
-	// Standard deviation is undefined if we have only one winner.
-	if (num_seats < 2) {
-		throw std::invalid_argument("normal_proportionality: "
-			"Need at least two winners");
-	}
-
-	size_t dim, dims = reference.get_num_dimensions();
-
-	std::vector<double> mu_estimate(dims, 0), sigma_estimate(dims, 0);
-
-	// Get means first.
-	for (size_t winner: outcome) {
-		for (dim = 0; dim < dims; ++dim) {
-			mu_estimate[dim] += candidate_positions[winner][dim];
-		}
-		for (dim = 0; dim < dims; ++dim) {
-			mu_estimate[dim] /= (double)num_seats;
-		}
-	}
-
-	// Get standard deviations.
-	for (size_t winner: outcome) {
-		for (dim = 0; dim < dims; ++dim) {
-			sigma_estimate[dim] += square(
-					candidate_positions[winner][dim] - mu_estimate[dim]);
-		}
-		for (dim = 0; dim < dims; ++dim) {
-			// Setting this to 1 makes it large-biased, because a bunch of
-			// candidates clustered around the center gives a more accurate mean
-			// than if they're dispersed.
-			sigma_estimate[dim] = sqrt(sigma_estimate[dim] / (double)num_seats);
-		}
-	}
-
-	/*std::cout << "mu: ";
-	std::copy(mu_estimate.begin(), mu_estimate.end(), std::ostream_iterator<double>(std::cout, " "));
-	std::cout << "\n";
-	std::cout << "sigma: ";
-	std::copy(sigma_estimate.begin(), sigma_estimate.end(),
-		std::ostream_iterator<double>(std::cout, " "));
-	std::cout << std::endl;*/
-
-	gaussian_generator estimated;
-	estimated.set_params(dims, false);
-	estimated.set_center(mu_estimate);
-	estimated.set_dispersion(sigma_estimate);
-
-	// Pick random points from the reference distribution and
-	// get the pdf from both. Use Sainte-Laguë index-like measure.
-
-	// TODO: KL divergence? Closed form integral instead of this
-	// slow Monte Carlo stuff?
-
-	size_t iters = 500;
-	double error_sum = 0;
-
-	for (size_t i = 0; i < iters; ++i) {
-		std::vector<double> q = reference.rnd_vector(dims, rnd);
-
-		double observed = estimated.pdf(q),
-			   expected = reference.pdf(q);
-
-		error_sum += square(observed-expected)/expected;
-	}
-
-	return error_sum / (double)iters;
-}
+#include "multiwinner/pr_measures/clustering.h"
+#include "multiwinner/pr_measures/normal_fit.h"
 
 // normsum, etc.
 
@@ -236,6 +25,8 @@ int main() {
 
 	size_t num_clusters = 2; // say
 
+	size_t maxiters = 5000;
+
 	std::cout << gauss.pdf(std::vector<double>(5, 0.1)) << "\n";
 
 	cluster_proportionality test(num_clusters);
@@ -244,7 +35,9 @@ int main() {
 	for (double delta = 0.1; delta <= 1; delta += 0.1) {
 
 		double error = 0;
-		for (int i = 0; i < 5000; ++i) {
+		for (int i = 0; i < maxiters; ++i) {
+
+			std::cerr << i << "/" << maxiters << "    \r" << std::flush;
 
 			positions_election p_e = gauss.generate_election_result(
 					num_voters, num_candidates, false, rnd);
@@ -260,6 +53,7 @@ int main() {
 			error += ntest.get_error(qpq_council);
 		}
 
+		std::cerr << "\n";
 		std::cout << delta << "\t" << error << std::endl;
 	}
 }

src/main/multiwinner_spatial.cc

@@ -19,36 +19,36 @@
 
 #include "multiwinner/helper/errors.cc"
 
-#include "multiwinner/methods.h"
-#include "multiwinner/exhaustive/birational.h"
-#include "multiwinner/exhaustive/harmonic.h"
-#include "multiwinner/exhaustive/isoelastic.h"
-#include "multiwinner/exhaustive/lpv.h"
-#include "multiwinner/exhaustive/psi.h"
-#include "multiwinner/exhaustive/quota_helper.h"
-
-#include "multiwinner/rusty/fc_kemeny.h"
-#include "multiwinner/rusty/mono_webst_640.h"
-#include "multiwinner/rusty/mono_webst_c37.h"
-#include "multiwinner/rusty/mono_webst_f03.h"
-#include "multiwinner/rusty/mw_kemeny2_34e.h"
-#include "multiwinner/rusty/mw_kemeny_db0.h"
-
-#include "multiwinner/auction.h"
-#include "multiwinner/compat_qbuck.h"
-#include "multiwinner/dhwl.h"
-#include "multiwinner/meek_stv.h"
-#include "multiwinner/prbucklin.h"
-#include "multiwinner/psc.h"
-#include "multiwinner/qbuck.h"
-#include "multiwinner/set_webster.h"
-#include "multiwinner/shuntsstv.h"
-#include "multiwinner/stv.h"
-
-#include "multiwinner/randballots.h"
-#include "multiwinner/qpq.h"
-#include "multiwinner/qrange_stv.h"
-#include "multiwinner/range_stv.h"
+#include "multiwinner/methods/methods.h"
+#include "multiwinner/methods/exhaustive/birational.h"
+#include "multiwinner/methods/exhaustive/harmonic.h"
+#include "multiwinner/methods/exhaustive/isoelastic.h"
+#include "multiwinner/methods/exhaustive/lpv.h"
+#include "multiwinner/methods/exhaustive/psi.h"
+#include "multiwinner/methods/exhaustive/quota_helper.h"
+
+#include "multiwinner/methods/rusty/fc_kemeny.h"
+#include "multiwinner/methods/rusty/mono_webst_640.h"
+#include "multiwinner/methods/rusty/mono_webst_c37.h"
+#include "multiwinner/methods/rusty/mono_webst_f03.h"
+#include "multiwinner/methods/rusty/mw_kemeny2_34e.h"
+#include "multiwinner/methods/rusty/mw_kemeny_db0.h"
+
+#include "multiwinner/methods/auction.h"
+#include "multiwinner/methods/compat_qbuck.h"
+#include "multiwinner/methods/dhwl.h"
+#include "multiwinner/methods/meek_stv.h"
+#include "multiwinner/methods/prbucklin.h"
+#include "multiwinner/methods/psc.h"
+#include "multiwinner/methods/qbuck.h"
+#include "multiwinner/methods/set_webster.h"
+#include "multiwinner/methods/shuntsstv.h"
+#include "multiwinner/methods/stv.h"
+
+#include "multiwinner/methods/randballots.h"
+#include "multiwinner/methods/qpq.h"
+#include "multiwinner/methods/qrange_stv.h"
+#include "multiwinner/methods/range_stv.h"
 
 #include "hack/msvc_random.h"
 

src/multiwinner/methods/all.h

@@ -0,0 +1,18 @@
+#include "auction.h"
+#include "compat_qbuck.h"
+#include "cpo_stv.h"
+#include "dhwl.h"
+#include "dhwl_mat.h"
+#include "meek_stv.h"
+#include "methods.h"
+#include "prbucklin.h"
+#include "psc.h"
+#include "qbuck.h"
+#include "qpq.h"
+#include "qrange_stv.h"
+#include "quotas.h"
+#include "randballots.h"
+#include "range_stv.h"
+#include "set_webster.h"
+#include "shuntsstv.h"
+#include "stv.h"