source: src/Fragmentation/Fragmentation.cpp@ d760bb

/*
 * Project: MoleCuilder
 * Description: creates and alters molecular systems
 * Copyright (C)  2010-2012 University of Bonn. All rights reserved.
 * 
 *
 *   This file is part of MoleCuilder.
 *
 *    MoleCuilder is free software: you can redistribute it and/or modify
 *    it under the terms of the GNU General Public License as published by
 *    the Free Software Foundation, either version 2 of the License, or
 *    (at your option) any later version.
 *
 *    MoleCuilder is distributed in the hope that it will be useful,
 *    but WITHOUT ANY WARRANTY; without even the implied warranty of
 *    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *    GNU General Public License for more details.
 *
 *    You should have received a copy of the GNU General Public License
 *    along with MoleCuilder.  If not, see <http://www.gnu.org/licenses/>.
 */

/*
 * Fragmentation.cpp
 *
 *  Created on: Oct 18, 2011
 *      Author: heber
 */

#ifdef HAVE_CONFIG_H
#include <config.h>
#endif

#include "CodePatterns/MemDebug.hpp"

#include "Fragmentation.hpp"

#include "CodePatterns/Assert.hpp"
#include "CodePatterns/Info.hpp"
#include "CodePatterns/Log.hpp"

#include "Atom/atom.hpp"
#include "Bond/bond.hpp"
#include "Descriptors/MoleculeDescriptor.hpp"
#include "Element/element.hpp"
#include "Element/periodentafel.hpp"
#include "Fragmentation/AdaptivityMap.hpp"
#include "Fragmentation/AtomMask.hpp"
#include "Fragmentation/fragmentation_helpers.hpp"
#include "Fragmentation/Graph.hpp"
#include "Fragmentation/helpers.hpp"
#include "Fragmentation/KeySet.hpp"
#include "Fragmentation/PowerSetGenerator.hpp"
#include "Fragmentation/UniqueFragments.hpp"
#include "Graph/BondGraph.hpp"
#include "Graph/AdjacencyList.hpp"
#include "Graph/ListOfLocalAtoms.hpp"
#include "molecule.hpp"
#include "World.hpp"


/** Constructor of class Fragmentation.
 *
 * \param _mol molecule for internal use (looking up atoms)
 * \param _FileChecker instance contains adjacency parsed from elsewhere
 * \param _saturation whether to treat hydrogen special and saturate dangling bonds or not
 */
Fragmentation::Fragmentation(molecule *_mol, AdjacencyList &_FileChecker, const enum HydrogenSaturation _saturation) :
  mol(_mol),
  saturation(_saturation),
  FileChecker(_FileChecker)
{}

/** Destructor of class Fragmentation.
 *
 */
Fragmentation::~Fragmentation()
{}


/** Performs a many-body bond order analysis for a given bond order.
 * -# parses adjacency, keysets and orderatsite files
 * -# RootStack is created for every subgraph (here, later we implement the "update 10 sites with highest energ
y contribution", and that's why this consciously not done in the following loop)
 * -# in a loop over all subgraphs
 *  -# calls FragmentBOSSANOVA with this RootStack and within the subgraph molecule structure
 *  -# creates molecule (fragment)s from the returned keysets (StoreFragmentFromKeySet)
 * -# combines the generated molecule lists from all subgraphs
 * -# saves to disk: fragment configs, adjacency, orderatsite, keyset files
 * Note that as we split "this" molecule up into a list of subgraphs, i.e. a MoleculeListClass, we have two sets
 * of vertex indices: Global always means the index in "this" molecule, whereas local refers to the molecule or
 * subgraph in the MoleculeListClass.
 * \param atomids atomic ids (local to Fragmentation::mol) to fragment, used in AtomMask
 * \param Order up to how many neighbouring bonds a fragment contains in BondOrderScheme::BottumUp scheme
 * \param prefix prefix string for every fragment file name (may include path)
 * \param &DFS contains bond structure analysis data
 * \return 1 - continue, 2 - stop (no fragmentation occured)
 */
int Fragmentation::FragmentMolecule(
    const std::vector<atomId_t> &atomids,
    int Order,
    const std::string &prefix,
    DepthFirstSearchAnalysis &DFS,
    const Graph &ParsedFragmentList)
{
  std::fstream File;
  bool CheckOrder = false;
  int TotalNumberOfKeySets = 0;

  LOG(0, std::endl);
  switch (saturation) {
    case DoSaturate:
      LOG(0, "I will treat hydrogen special and saturate dangling bonds with it.");
      break;
    case DontSaturate:
      LOG(0, "Hydrogen is treated just like the rest of the lot.");
      break;
    default:
      ASSERT(0, "Fragmentation::FragmentMolecule() - there is a saturation setting which I have no idea about.");
      break;
  }

  // ++++++++++++++++++++++++++++ INITIAL STUFF: Bond structure analysis, file parsing, ... ++++++++++++++++++++++++++++++++++++++++++
  bool FragmentationToDo = true;

  // ===== 1. Check whether bond structure is same as stored in files ====

  // === compare it with adjacency file ===
  {
    const std::vector<atomId_t> globalids = getGlobalIdsFromLocalIds(*mol, atomids);
    AdjacencyList WorldAdjacency(globalids);
    FragmentationToDo = FragmentationToDo && (FileChecker > WorldAdjacency);
  }

  // ===== 2. create AtomMask that takes Saturation condition into account
  AtomMask_t AtomMask(atomids);
  for (molecule::const_iterator iter = mol->begin(); iter != mol->end(); ++iter) {
    // remove in hydrogen and we do saturate
    if ((saturation == DoSaturate) && ((*iter)->getType()->getAtomicNumber() == 1)) // skip hydrogen
      AtomMask.setFalse((*iter)->getNr());
  }

  // ===== 4. check globally whether there's something to do actually (first adaptivity check)
  FragmentationToDo = FragmentationToDo && ParseOrderAtSiteFromFile(atomids, prefix);

  // =================================== Begin of FRAGMENTATION ===============================
  // ===== 6a. assign each keyset to its respective subgraph =====
  ListOfLocalAtoms_t ListOfLocalAtoms;
  Graph FragmentList;
  AssignKeySetsToFragment(ParsedFragmentList, ListOfLocalAtoms, FragmentList, true);

  // ===== 6b. prepare and go into the adaptive (Order<0), single-step (Order==0) or incremental (Order>0) cycle
  KeyStack RootStack;
  FragmentationToDo = false;  // if CheckOrderAtSite just ones recommends fragmentation, we will save fragments afterwards
  bool LoopDoneAlready = false;
  while ((CheckOrder = CheckOrderAtSite(AtomMask, ParsedFragmentList, Order, prefix, LoopDoneAlready))) {
    FragmentationToDo = FragmentationToDo || CheckOrder;
    LoopDoneAlready = true;   // last plus one entry is used as marker that we have been through this loop once already in CheckOrderAtSite()
    // ===== 6b. fill RootStack for each subgraph (second adaptivity check) =====
    FillRootStackForSubgraphs(RootStack, AtomMask);

    // call BOSSANOVA method
    FragmentBOSSANOVA(mol, FragmentList, RootStack);
  }
  LOG(2, "CheckOrder is " << CheckOrder << ".");

  // ==================================== End of FRAGMENTATION ============================================

  // ===== 8a. translate list into global numbers (i.e. ones that are valid in "this" molecule, not in MolecularWalker->Leaf)
  TranslateIndicesToGlobalIDs(FragmentList, TotalNumberOfKeySets, TotalGraph);

  LOG(1, "STATUS: We have created " << TotalGraph.size() << " fragments.");


  // store adaptive orders into file
  StoreOrderAtSiteFile(prefix);

  LOG(0, "End of bond fragmentation.");
  return ((int)(!FragmentationToDo)+1);    // 1 - continue, 2 - stop (no fragmentation occured)
};


/** Performs BOSSANOVA decomposition at selected sites, increasing the cutoff by one at these sites.
 * -# constructs a complete keyset of the molecule
 * -# In a loop over all possible roots from the given rootstack
 *  -# increases order of root site
 *  -# calls PowerSetGenerator with this order, the complete keyset and the rootkeynr
 *  -# for all consecutive lower levels PowerSetGenerator is called with the suborder, the higher order keyset
as the restricted one and each site in the set as the root)
 *  -# these are merged into a fragment list of keysets
 * -# All fragment lists (for all orders, i.e. from all destination fields) are merged into one list for return
 * Important only is that we create all fragments, it is not important if we create them more than once
 * as these copies are filtered out via use of the hash table (KeySet).
 * \param *out output stream for debugging
 * \param Fragment&*List list of already present keystacks (adaptive scheme) or empty list
 * \param &RootStack stack with all root candidates (unequal to each atom in complete molecule if adaptive scheme is applied)
 * \return pointer to Graph list
 */
void Fragmentation::FragmentBOSSANOVA(molecule *mol, Graph &FragmentList, KeyStack &RootStack)
{
  Info FunctionInfo(__func__);
  std::vector<Graph*> *FragmentLowerOrdersList = NULL;
  int NumLevels = 0;
  int NumMolecules = 0;
  int TotalNumMolecules = 0;
  int *NumMoleculesOfOrder = NULL;
  int Order = 0;
  int UpgradeCount = RootStack.size();
  KeyStack FragmentRootStack;
  int RootKeyNr = 0;
  int RootNr = 0;

  // FragmentLowerOrdersList is a 2D-array of pointer to MoleculeListClass objects, one dimension represents the ANOVA expansion of a single order (i.e. 5)
  // with all needed lower orders that are subtracted, the other dimension is the BondOrder (i.e. from 1 to 5)
  NumMoleculesOfOrder = new int[UpgradeCount];
  FragmentLowerOrdersList = new std::vector<Graph*>[UpgradeCount];

  for(int i=0;i<UpgradeCount;i++)
    NumMoleculesOfOrder[i] = 0;

  // Construct the complete KeySet which we need for topmost level only (but for all Roots)
  KeySet CompleteMolecule;
  for (molecule::const_iterator iter = mol->begin(); iter != mol->end(); ++iter) {
    CompleteMolecule.insert((*iter)->GetTrueFather()->getNr());
  }

  // this can easily be seen: if Order is 5, then the number of levels for each lower order is the total sum of the number of levels above, as
  // each has to be split up. E.g. for the second level we have one from 5th, one from 4th, two from 3th (which in turn is one from 5th, one from 4th),
  // hence we have overall four 2th order levels for splitting. This also allows for putting all into a single array (FragmentLowerOrdersList[])
  // with the order along the cells as this: 5433222211111111 for BondOrder 5 needing 16=pow(2,5-1) cells (only we use bit-shifting which is faster)
  RootNr = 0;   // counts through the roots in RootStack
  while ((RootNr < UpgradeCount) && (!RootStack.empty())) {
    RootKeyNr = RootStack.front();
    RootStack.pop_front();
    atom *Walker = mol->FindAtom(RootKeyNr);
    // check cyclic lengths
    //if ((MinimumRingSize[Walker->GetTrueFather()->getNr()] != -1) && (Walker->GetTrueFather()->AdaptiveOrder+1 > MinimumRingSize[Walker->GetTrueFather()->getNr()])) {
    //  LOG(0, "Bond order " << Walker->GetTrueFather()->AdaptiveOrder << " of Root " << *Walker << " greater than or equal to Minimum Ring size of " << MinimumRingSize << " found is not allowed.");
    //} else
    {
      // set adaptive order to desired max order
      Walker->GetTrueFather()->AdaptiveOrder = Walker->GetTrueFather()->MaxOrder;
      Order = Walker->AdaptiveOrder = Walker->GetTrueFather()->AdaptiveOrder;

      // allocate memory for all lower level orders
      NumLevels = Order;
      FragmentLowerOrdersList[RootNr].resize(NumLevels, NULL);
      for( size_t i=0;i<NumLevels;++i)
        FragmentLowerOrdersList[RootNr][i] = new Graph;

      // initialise Order-dependent entries of UniqueFragments structure
      UniqueFragments FragmentSearch(1., FragmentLowerOrdersList[RootNr], Walker);
      PowerSetGenerator PSG(&FragmentSearch, Walker->AdaptiveOrder);

      // create top order where nothing is reduced
      LOG(0, "==============================================================================================================");
      LOG(0, "Creating KeySets up till Bond Order " << Order << " for " << *Walker << ", " << (RootStack.size()-RootNr) << " Roots remaining."); // , NumLevels is " << NumLevels << "

      // Create list of Graphs of current Bond Order (i.e. F_{ij})
      NumMoleculesOfOrder[RootNr] = PSG(CompleteMolecule, saturation);

      // output resulting number
      LOG(1, "Number of resulting KeySets is: " << NumMoleculesOfOrder[RootNr] << ".");
      if (NumMoleculesOfOrder[RootNr] != 0) {
        NumMolecules = 0;
      }
      // now, we have completely filled each cell of FragmentLowerOrdersList[] for the current Walker->AdaptiveOrder
      //NumMoleculesOfOrder[Walker->AdaptiveOrder-1] = NumMolecules;
      TotalNumMolecules += NumMoleculesOfOrder[RootNr];
//      LOG(1, "Number of resulting molecules for Order " << (int)Walker->GetTrueFather()->AdaptiveOrder << " is: " << NumMoleculesOfOrder[RootNr] << ".");
      RootStack.push_back(RootKeyNr); // put back on stack
      RootNr++;
    }
  }
  LOG(0, "==============================================================================================================");
  LOG(1, "Total number of resulting molecules is: " << TotalNumMolecules << ".");
  LOG(0, "==============================================================================================================");

  // now, FragmentLowerOrdersList is complete, it looks - for BondOrder 5 - as this (number is the ANOVA Order of the terms therein)
  // 5433222211111111
  // 43221111
  // 3211
  // 21
  // 1

  // Subsequently, we combine all into a single list (FragmentList)
  CombineAllOrderListIntoOne(FragmentList, FragmentLowerOrdersList, RootStack, mol);
  FreeAllOrdersList(FragmentLowerOrdersList, RootStack, mol);
  delete[](NumMoleculesOfOrder);
};

/** Estimates by educated guessing (using upper limit) the expected number of fragments.
 * The upper limit is
 * \f[
 *  n = N \cdot C^k
 * \f]
 * where \f$C=2^c\f$ and c is the maximum bond degree over N number of atoms.
 * \param *out output stream for debugging
 * \param order bond order k
 * \return number n of fragments
 */
int Fragmentation::GuesstimateFragmentCount(int order)
{
  size_t c = 0;
  int FragmentCount;
  // get maximum bond degree
  for (molecule::const_iterator iter = mol->begin(); iter != mol->end(); ++iter) {
    const BondList& ListOfBonds = (*iter)->getListOfBonds();
    c = (ListOfBonds.size() > c) ? ListOfBonds.size() : c;
  }
  FragmentCount = (saturation == DoSaturate ? mol->getNoNonHydrogen() : mol->getAtomCount()) *(1 << (c*order));
  LOG(1, "Upper limit for this subgraph is " << FragmentCount << " for "
      << mol->getNoNonHydrogen() << " non-H atoms with maximum bond degree of " << c << ".");
  return FragmentCount;
};


/** Checks whether the OrderAtSite is still below \a Order at some site.
 * \param AtomMask defines true/false per global Atom::Nr to mask in/out each nuclear site, used to activate given number of site to increment order adaptively
 * \param *GlobalKeySetList list of keysets with global ids (valid in "this" molecule) needed for adaptive increase
 * \param Order desired Order if positive, desired exponent in threshold criteria if negative (0 is single-step)
 * \param path path to ENERGYPERFRAGMENT file (may be NULL if Order is non-negative)
 * \param LoopDoneAlready indicate whether we have done a fragmentation loop already
 * \return true - needs further fragmentation, false - does not need fragmentation
 */
bool Fragmentation::CheckOrderAtSite(AtomMask_t &AtomMask, const Graph &GlobalKeySetList, int Order, const std::string &path, bool LoopDoneAlready)
{
  bool status = false;

  if (Order < 0) { // adaptive increase of BondOrder per site
    if (LoopDoneAlready)  // break after one step
      return false;

    // transmorph graph keyset list into indexed KeySetList
    AdaptivityMap * IndexKeySetList = GlobalKeySetList.GraphToAdaptivityMap();

    // parse the EnergyPerFragment file
    IndexKeySetList->ScanAdaptiveFileIntoMap(path); // (Root No., (Value, Order)) !
    // then map back onto (Value, (Root Nr., Order)) (i.e. sorted by value to pick the highest ones)
    IndexKeySetList->ReMapAdaptiveCriteriaListToValue(mol);

    // pick the ones still below threshold and mark as to be adaptively updated
    if (IndexKeySetList->IsAdaptiveCriteriaListEmpty()) {
      ELOG(2, "Unable to parse file, incrementing all.");
      status = true;
    } else {
      // mark as false all sites that are below threshold already
      status = IndexKeySetList->MarkUpdateCandidates(AtomMask, Order, mol);
    }

    delete[](IndexKeySetList);
  } else { // global increase of Bond Order
    for(molecule::iterator iter = mol->begin(); iter != mol->end(); ++iter) {
      if (AtomMask.isTrue((*iter)->getNr())) { // skip masked out
        (*iter)->MaxOrder = (Order != 0 ? Order : (*iter)->MaxOrder+1);
        // remove all that have reached desired order
        if ((*iter)->AdaptiveOrder >= (*iter)->MaxOrder) // && ((*iter)->AdaptiveOrder < MinimumRingSize[(*iter)->getNr()]))
          AtomMask.setFalse((*iter)->getNr());
        else
          status = true;
      }
    }
    if ((!Order) && (!LoopDoneAlready))  // single stepping, just check
      status = true;

    if (!status) {
      if (Order == 0)
        LOG(1, "Single stepping done.");
      else
        LOG(1, "Order at every site is already equal or above desired order " << Order << ".");
    }
  }

  PrintAtomMask(AtomMask, mol->getAtomCount()); // for debugging

  return status;
};

/** Stores pairs (Atom::Nr, Atom::AdaptiveOrder) into file.
 * Atoms not present in the file get "-1".
 * \param &path path to file ORDERATSITEFILE
 * \return true - file writable, false - not writable
 */
bool Fragmentation::StoreOrderAtSiteFile(const std::string &path)
{
  string line;
  ofstream file;

  line = path + ORDERATSITEFILE;
  file.open(line.c_str());
  LOG(1, "Writing OrderAtSite " << ORDERATSITEFILE << " ... ");
  if (file.good()) {
    for_each(mol->begin(),mol->end(),bind2nd(mem_fun(&atom::OutputOrder), &file));
    file.close();
    LOG(1, "done.");
    return true;
  } else {
    LOG(1, "failed to open file " << line << ".");
    return false;
  }
};


/** Parses pairs(Atom::Nr, Atom::AdaptiveOrder) from file and stores in molecule's Atom's.
 * Atoms not present in the file get "0".
 * \param atomids atoms to fragment, used in AtomMask
 * \param &path path to file ORDERATSITEFILEe
 * \return true - file found and scanned, false - file not found
 * \sa ParseKeySetFile() and CheckAdjacencyFileAgainstMolecule() as this is meant to be used in conjunction with the two
 */
bool Fragmentation::ParseOrderAtSiteFromFile(const std::vector<atomId_t> &atomids, const std::string &path)
{
  Info FunctionInfo(__func__);
  typedef unsigned char order_t;
  typedef std::map<atomId_t, order_t> OrderArray_t;
  OrderArray_t OrderArray;
  AtomMask_t MaxArray(atomids);
  bool status;
  int AtomNr, value;
  string line;
  ifstream file;

  line = path + ORDERATSITEFILE;
  file.open(line.c_str());
  if (file.good()) {
    while (!file.eof()) { // parse from file
      AtomNr = -1;
      file >> AtomNr;
      if (AtomNr != -1) {   // test whether we really parsed something (this is necessary, otherwise last atom is set twice and to 0 on second time)
        file >> value;
        OrderArray[AtomNr] = value;
        file >> value;
        MaxArray.setValue(AtomNr, (bool)value);
        //LOG(2, "AtomNr " << AtomNr << " with order " << (int)OrderArray[AtomNr] << " and max order set to " << (int)MaxArray[AtomNr] << ".");
      }
    }
    file.close();

    // set atom values
    for(molecule::iterator iter=mol->begin();iter!=mol->end();++iter){
      (*iter)->AdaptiveOrder = OrderArray[(*iter)->getNr()];
      (*iter)->MaxOrder = OrderArray[(*iter)->getNr()]; //MaxArray.isTrue((*iter)->getNr());
    }
    //SetAtomValueToIndexedArray( OrderArray, &atom::getNr(), &atom::AdaptiveOrder );
    //SetAtomValueToIndexedArray( MaxArray, &atom::getNr(), &atom::MaxOrder );

    LOG(1, "\t ... done.");
    status = true;
  } else {
    LOG(1, "\t ... failed to open file " << line << ".");
    status = false;
  }

  return status;
};

/** Fills the root stack for sites to be used as root in fragmentation depending on order or adaptivity criteria
 * Again, as in \sa FillBondStructureFromReference steps recursively through each Leaf in this chain list of molecule's.
 * \param &RootStack stack to be filled
 * \param AtomMask defines true/false per global Atom::Nr to mask in/out each nuclear site
 * \return true - stack is non-empty, fragmentation necessary, false - stack is empty, no more sites to update
 */
void Fragmentation::FillRootStackForSubgraphs(KeyStack &RootStack, const AtomMask_t &AtomMask)
{
  for(molecule::const_iterator iter = mol->begin(); iter != mol->end(); ++iter) {
    const atom * const Father = (*iter)->GetTrueFather();
    if (AtomMask.isTrue(Father->getNr())) // apply mask
      if ((saturation == DontSaturate) || ((*iter)->getType()->getAtomicNumber() != 1)) // skip hydrogen
        RootStack.push_front((*iter)->getNr());
  }
}

/** The indices per keyset are compared to the respective father's Atom::Nr in each subgraph and thus put into \a **&FragmentList.
 * \param *KeySetList list with all keysets
 * \param ListOfLocalAtoms Lookup table for each subgraph and index of each atom in global molecule, may be NULL on start, then it is filled
 * \param **&FragmentList list to be allocated and returned
 * \param FreeList true - ***ListOfLocalAtoms is free'd before return, false - it is not
 * \retuen true - success, false - failure
 */
bool Fragmentation::AssignKeySetsToFragment(const Graph &KeySetList, ListOfLocalAtoms_t &ListOfLocalAtoms, Graph &FragmentList, bool FreeList)
{
  Info FunctionInfo(__func__);
  bool status = true;
  size_t KeySetCounter = 0;

  // fill ListOfLocalAtoms if NULL was given
  if (!mol->FillListOfLocalAtoms(ListOfLocalAtoms, mol->getAtomCount())) {
    LOG(1, "Filling of ListOfLocalAtoms failed.");
    return false;
  }

  if (KeySetList.size() != 0) { // if there are some scanned keysets at all
    // assign scanned keysets
    KeySet TempSet;
    for (Graph::const_iterator runner = KeySetList.begin(); runner != KeySetList.end(); runner++) { // key sets contain global numbers!
      if (ListOfLocalAtoms[mol->FindAtom(*((*runner).first.begin()))->getNr()] != NULL) {// as we may assume that that bond structure is unchanged, we only test the first key in each set
        // translate keyset to local numbers
        for (KeySet::iterator sprinter = (*runner).first.begin(); sprinter != (*runner).first.end(); sprinter++)
          TempSet.insert(ListOfLocalAtoms[mol->FindAtom(*sprinter)->getNr()]->getNr());
        // insert into FragmentList
        FragmentList.insert(GraphPair(TempSet, pair<int, double> (KeySetCounter++, (*runner).second.second)));
      }
      TempSet.clear();
    }
  } else
    LOG(1, "KeySetList is NULL or empty.");

  if (FreeList) {
    // free the index lookup list
    ListOfLocalAtoms.clear();
  }
  return status;
}

/** Translate list into global numbers (i.e. ones that are valid in "this" molecule, not in MolecularWalker->Leaf)
 * \param &FragmentList Graph with local numbers per fragment
 * \param &TotalNumberOfKeySets global key set counter
 * \param &TotalGraph Graph to be filled with global numbers
 */
void Fragmentation::TranslateIndicesToGlobalIDs(Graph &FragmentList, int &TotalNumberOfKeySets, Graph &TotalGraph)
{
  Info FunctionInfo(__func__);
  for (Graph::iterator runner = FragmentList.begin(); runner != FragmentList.end(); runner++) {
    KeySet TempSet;
    for (KeySet::iterator sprinter = (*runner).first.begin(); sprinter != (*runner).first.end(); sprinter++)
      TempSet.insert((mol->FindAtom(*sprinter))->GetTrueFather()->getId());
    TotalGraph.insert(GraphPair(TempSet, pair<int, double> (TotalNumberOfKeySets++, (*runner).second.second)));
  }
}