Changeset - 9905828198ec
[Not reviewed]
1 4 4
Tom Bannink - 8 years ago 2017-05-15 16:54:05
tombannink@gmail.com
Split cpp files and add more triangle exponent data
6 files changed:
0 comments (0 inline, 0 general)
cpp/Makefile
Show inline comments
 
@@ -3,8 +3,19 @@ INCLUDES += -I.
 

	
 
CXXFLAGS += -std=c++14 -O3 -Wall -Wextra -Wfatal-errors -Wno-deprecated-declarations $(INCLUDES)
 

	
 

	
 
all: switchchain switchchain_exponent switchchain_initialtris
 

	
 

	
 
switchchain:
 

	
 

	
 
switchchain_exponent:
 

	
 

	
 
switchchain_initialtris:
 

	
 

	
 
# target : dep1 dep2 dep3
 
# 	$@ = target
 
# 	$< = dep1
cpp/switchchain.cpp
Show inline comments
 
@@ -181,7 +181,7 @@ int main() {
 
    // Expect:  #tri = const * n^{ something }
 
    // The goal is to find the 'something' by finding the number of triangles
 
    // for different values of n and tau
 
    float tauValues[] = {2.1f, 2.2f, 2.3f, 2.4f, 2.5f, 2.6f, 2.7f, 2.8f};
 
    float tauValues[] = {2.1f, 2.2f, 2.3f, 2.4f, 2.5f, 2.6f, 2.7f, 2.8f, 2.9f};
 

	
 
    Graph g;
 
    Graph g1;
cpp/switchchain_exponent.cpp
Show inline comments
 
@@ -125,7 +125,7 @@ int main() {
 
                std::cout << "Running n = " << numVertices << ", tau = " << tau
 
                          << ". \t" << std::flush;
 

	
 
                int mixingTime = (32.0f - 26.0f*(tau - 2.0f)) * numVertices; //40000;
 
                int mixingTime = 8*(32.0f - 26.0f*(tau - 2.0f)) * numVertices; //40000;
 
                constexpr int measurements = 50;
 
                constexpr int measureSkip =
 
                    200; // Take a sample every ... steps
cpp/switchchain_initialtris.cpp
Show inline comments
 
new file 100644
 
#include "exports.hpp"
 
#include "graph.hpp"
 
#include "powerlaw.hpp"
 
#include <algorithm>
 
#include <fstream>
 
#include <iostream>
 
#include <numeric>
 
#include <random>
 
#include <vector>
 

	
 
// Its assumed that u,v are distinct.
 
// Check if all four vertices are distinct
 
bool edgeConflicts(const Edge& e1, const Edge& e2) {
 
    return (e1.u == e2.u || e1.u == e2.v || e1.v == e2.u || e1.v == e2.v);
 
}
 

	
 
class SwitchChain {
 
  public:
 
    SwitchChain()
 
        : mt(std::random_device{}()), permutationDistribution(0.5)
 
    // permutationDistribution(0, 2)
 
    {
 
        // random_device uses hardware entropy if available
 
        // std::random_device rd;
 
        // mt.seed(rd());
 
    }
 
    ~SwitchChain() {}
 

	
 
    bool initialize(const Graph& gstart) {
 
        if (gstart.edgeCount() == 0)
 
            return false;
 
        g = gstart;
 
        edgeDistribution.param(
 
            std::uniform_int_distribution<>::param_type(0, g.edgeCount() - 1));
 
        return true;
 
    }
 

	
 
    bool doMove() {
 
        int e1index, e2index;
 
        int timeout = 0;
 
        // Keep regenerating while conflicting edges
 
        do {
 
            e1index = edgeDistribution(mt);
 
            e2index = edgeDistribution(mt);
 
            if (++timeout % 100 == 0) {
 
                std::cerr << "Warning: sampled " << timeout
 
                          << " random edges but they keep conflicting.\n";
 
            }
 
        } while (edgeConflicts(g.getEdge(e1index), g.getEdge(e2index)));
 

	
 
        // Consider one of the three possible permutations
 
        // 1) e1.u - e1.v and e2.u - e2.v (original)
 
        // 2) e1.u - e2.u and e1.v - e2.v
 
        // 3) e1.u - e2.v and e1.v - e2.u
 
        bool switchType = permutationDistribution(mt);
 
        return g.exchangeEdges(e1index, e2index, switchType);
 
    }
 

	
 
    Graph g;
 
    std::mt19937 mt;
 
    std::uniform_int_distribution<> edgeDistribution;
 
    //std::uniform_int_distribution<> permutationDistribution;
 
    std::bernoulli_distribution permutationDistribution;
 
};
 

	
 
//
 
// Assumes degree sequence does NOT contain any zeroes!
 
//
 
// method2 = true  -> take highest degree and finish its pairing completely
 
// method2 = false -> take new highest degree after every pairing
 
bool greedyConfigurationModel(DegreeSequence& ds, Graph& g, auto& rng, bool method2) {
 
    // Similar to Havel-Hakimi but instead of pairing up with the highest ones
 
    // that remain, simply pair up with random ones
 
    unsigned int n = ds.size();
 

	
 
    // degree, vertex index
 
    std::vector<std::pair<unsigned int, unsigned int>> degrees(n);
 
    for (unsigned int i = 0; i < n; ++i) {
 
        degrees[i].first = ds[i];
 
        degrees[i].second = i;
 
    }
 

	
 
    std::vector<decltype(degrees.begin())> available;
 
    available.reserve(n);
 

	
 
    // Clear the graph
 
    g.reset(n);
 

	
 
    while (!degrees.empty()) {
 
        std::shuffle(degrees.begin(), degrees.end(), rng);
 
        // Get the highest degree:
 
        // If there are multiple highest ones, we pick a random one,
 
        // ensured by the shuffle.
 
        // The shuffle is needed anyway for the remaining part
 
        unsigned int dmax = 0;
 
        auto uIter = degrees.begin();
 
        for (auto iter = degrees.begin(); iter != degrees.end(); ++iter) {
 
            if (iter->first >= dmax) {
 
                dmax = iter->first;
 
                uIter = iter;
 
            }
 
        }
 

	
 
        if (dmax > degrees.size() - 1)
 
            return false;
 

	
 
        if (dmax == 0) {
 
            std::cerr << "ERROR 1 in GCM.\n";
 
        }
 

	
 
        unsigned int u = uIter->second;
 

	
 
        if (method2) {
 
            // Take the highest degree out of the vector
 
            degrees.erase(uIter);
 

	
 
            // Now assign randomly to the remaining vertices
 
            // Since its shuffled, we can pick the first 'dmax' ones
 
            auto vIter = degrees.begin();
 
            while (dmax--) {
 
                if (vIter->first == 0)
 
                    std::cerr << "ERROR in GCM2.\n";
 
                if (!g.addEdge({u, vIter->second}))
 
                    std::cerr << "ERROR. Could not add edge in GCM2.\n";
 
                vIter->first--;
 
                if (vIter->first == 0)
 
                    vIter = degrees.erase(vIter);
 
                else
 
                    vIter++;
 
            }
 
        } else {
 
            // Pair with a random vertex that is not u itself and to which
 
            // u has not paired yet!
 
            available.clear();
 
            for (auto iter = degrees.begin(); iter != degrees.end(); ++iter) {
 
                if (iter->second != u && !g.hasEdge({u, iter->second}))
 
                    available.push_back(iter);
 
            }
 
            if (available.empty())
 
                return false;
 
            std::uniform_int_distribution<> distr(0, available.size() - 1);
 
            auto vIter = available[distr(rng)];
 
            // pair u to v
 
            if (vIter->first == 0)
 
                std::cerr << "ERROR 2 in GCM1.\n";
 
            if (!g.addEdge({u, vIter->second}))
 
                std::cerr << "ERROR. Could not add edge in GCM1.\n";
 
            // Purge anything with degree zero
 
            // Be careful with invalidating the other iterator!
 
            // Degree of u is always greater or equal to the degree of v
 
            if (dmax == 1) {
 
                // Remove both
 
                // Erasure invalidates all iterators AFTER the erased one
 
                if (vIter > uIter) {
 
                    degrees.erase(vIter);
 
                    degrees.erase(uIter);
 
                } else {
 
                    degrees.erase(uIter);
 
                    degrees.erase(vIter);
 
                }
 
            } else {
 
                // Remove only v if it reaches zero
 
                uIter->first--;
 
                vIter->first--;
 
                if (vIter->first == 0)
 
                    degrees.erase(vIter);
 
            }
 
            //degrees.erase(std::remove_if(degrees.begin(), degrees.end(),
 
            //                             [](auto x) { return x.first == 0; }));
 
        }
 
    }
 
    return true;
 
}
 

	
 
int main() {
 
    // Generate a random degree sequence
 
    std::mt19937 rng(std::random_device{}());
 

	
 
    // Goal:
 
    // Degrees follow a power-law distribution with some parameter tau
 
    // Expect:  #tri = const * n^{ something }
 
    // The goal is to find the 'something' by finding the number of triangles
 
    // for different values of n and tau
 
    float tauValues[] = {2.1f, 2.2f, 2.3f, 2.4f, 2.5f, 2.6f, 2.7f, 2.8f, 2.9f};
 

	
 
    Graph g;
 

	
 
    std::ofstream outfile("graphdata_initialtris.m");
 
    outfile << '{';
 
    bool outputComma = false;
 

	
 
    for (int numVertices = 200; numVertices <= 2000; numVertices += 400) {
 
        for (float tau : tauValues) {
 

	
 
            DegreeSequence ds(numVertices);
 
            powerlaw_distribution degDist(tau, 1, numVertices);
 
            //std::poisson_distribution<> degDist(12);
 

	
 
            // For a single n,tau take samples over several instances of
 
            // the degree distribution.
 
            // 500 samples seems to give reasonable results
 
            for (int degreeSample = 0; degreeSample < 200; ++degreeSample) {
 
                // Generate a graph
 
                // might require multiple tries
 
                for (int i = 1; ; ++i) {
 
                    std::generate(ds.begin(), ds.end(),
 
                                  [&degDist, &rng] { return degDist(rng); });
 
                    // First make the sum even
 
                    unsigned int sum = std::accumulate(ds.begin(), ds.end(), 0);
 
                    if (sum % 2) {
 
                        continue;
 
                        // Can we do this: ??
 
                        ds.back()++;
 
                    }
 

	
 
                    if (g.createFromDegreeSequence(ds))
 
                        break;
 

	
 
                    // When 10 tries have not worked, output a warning
 
                    if (i % 10 == 0) {
 
                        std::cerr << "Warning: could not create graph from "
 
                                     "degree sequence. Trying again...\n";
 
                    }
 
                }
 

	
 
                //
 
                // Test the GCM1 and GCM2 success rate
 
                //
 
                long long gcmTris1 = 0;
 
                long long gcmTris2 = 0;
 
                int successrate1 = 0;
 
                int successrate2 = 0;
 
                for (int i = 0; i < 100; ++i) {
 
                    Graph gtemp;
 
                    // Take new highest degree every time
 
                    if (greedyConfigurationModel(ds, gtemp, rng, false)) {
 
                        ++successrate1;
 
                        gcmTris1 += gtemp.countTriangles();
 
                    }
 
                    // Finish all pairings of highest degree first
 
                    if (greedyConfigurationModel(ds, gtemp, rng, true)) {
 
                        ++successrate2;
 
                        gcmTris2 += gtemp.countTriangles();
 
                    }
 
                }
 

	
 
                SwitchChain chain;
 
                if (!chain.initialize(g)) {
 
                    std::cerr << "Could not initialize Markov chain.\n";
 
                    return 1;
 
                }
 

	
 
                std::cout << "Running n = " << numVertices << ", tau = " << tau
 
                          << ". \t" << std::flush;
 

	
 
                int mixingTime = (32.0f - 20.0f * (tau - 2.0f)) * numVertices;
 
                constexpr int measurements = 20;
 
                constexpr int measureSkip = 200;
 

	
 
                int movesDone = 0;
 

	
 
                long long trianglesTotal = 0;
 

	
 
                for (int i = 0; i < mixingTime; ++i) {
 
                    if (chain.doMove())
 
                        ++movesDone;
 
                }
 
                for (int i = 0; i < measurements; ++i) {
 
                    for (int j = 0; j < measureSkip; ++j)
 
                        if (chain.doMove())
 
                            ++movesDone;
 
                    trianglesTotal = chain.g.countTriangles();
 
                }
 

	
 
                std::cout << movesDone << '/' << mixingTime + measurements * measureSkip
 
                          << " moves succeeded ("
 
                          << 100.0f * float(movesDone) /
 
                                 float(mixingTime + measurements * measureSkip)
 
                          << "%).";
 
                //std::cout << std::endl;
 

	
 
                if (outputComma)
 
                    outfile << ',' << '\n';
 
                outputComma = true;
 

	
 
                float avgTriangles =
 
                    float(trianglesTotal) / float(measurements);
 
                outfile << '{';
 
                outfile << '{' << numVertices << ',' << tau << '}';
 
                outfile << ',' << avgTriangles;
 
                outfile << ',' << '{' << gcmTris1 << ',' << successrate1 << '}';
 
                outfile << ',' << '{' << gcmTris2 << ',' << successrate2 << '}';
 
                outfile << '}' << std::flush;
 

	
 
                std::cout << std::endl;
 
            }
 
        }
 
    }
 
    outfile << '}';
 
    return 0;
 
}
data/README
Show inline comments
 
Contents of each file
 

	
 
graphdata_exponent.m
 
graphdata_exponent_mixN.m
 
    output: {{n,tau},avgTriangles}
 
    n from 200 to 2000 with step 200
 
    degreeSamples = 500 + 1000
 
    degreeSamples = 500 + 1000 for mix1 and 1000 for mix4,mix8
 
    initial ErdosGallai
 
    mixingTime = (32.0f - 26.0f*(tau - 2.0f)) * n
 
    mixingTime = N* (32.0f - 26.0f*(tau - 2.0f)) * n
 
    measurements = 50
 
    measureSkip = 200
 

	
 
graphdata_gcm_partial.m
 
    ??
 
    output: {{n,tau},triangleseq,ds, {...??...} }
 

	
 
graphdata_partial.m
 
    output: {{n,tau},triangleseq,ds,greedyTriangles1,greedyTriangles2,greedySeq1,greedySeq2}
data/graphdata_exponent_mix1.m
Show inline comments
 
file renamed from data/graphdata_exponent.m to data/graphdata_exponent_mix1.m

Changeset was too big and was cut off... Show full diff anyway

0 comments (0 inline, 0 general)