/** Create an iterator for the given iterator. If the verbosity is enough, progress bar information
* can be displayed.
* \param[in] iter : object to be encapsulated by a potential progress information
* \param[in] nbIterations : number of iterations to be done.
* \param[in] message : message used if progress information has to be displayed
* \return the created iterator.
*/
template<typename Item> dp::Iterator<Item>* createIterator (dp::Iterator<Item>* iter, size_t nbIterations=0, const char* message=0)
{
if (nbIterations > 0 && message != 0)
{
// We create some listener to be notified every 1000 iterations and attach it to the iterator.
dp::impl::SubjectIterator<Item>* iterSubject = new dp::impl::SubjectIterator<Item> (iter, nbIterations/100);
iterSubject->addObserver (createIteratorListener (nbIterations, message));
/** We assign the used iterator to be the subject iterator. */
iter = iterSubject;
}
/** We return the result. */
return iter;
}
void MPHFAlgorithm<span,Abundance_t,NodeState_t>::execute ()
{
/** We check whether we can use such a type. */
if (_buildOrLoad == true)
{
/** We need a progress object. */
tools::dp::IteratorListener* delegate = createIteratorListener(0,""); LOCAL (delegate);
setProgress (new ProgressCustom(delegate));
//if MPHF_BOOPHF and verbose 0, give a null progress to the builder, make it understand the internal progress bar of boophf needs to be removed
if((typeid(*delegate) == typeid(tools::dp::IteratorListener)))
setProgress (0);
// get number of threads from dispatcher
unsigned int nbThreads = this->getDispatcher()->getExecutionUnitsNumber();
/** We build the hash. */
{ TIME_INFO (getTimeInfo(), "build");
_abundanceMap->build (*_solidKmers, nbThreads, _progress);
}
/** We save the hash object in the dedicated storage group. */
{ TIME_INFO (getTimeInfo(), "save");
_dataSize = _abundanceMap->save (_group, _name);
}
/** We populate the hash table. */
populate ();
/** init a clean node state map */
initNodeStates ();
}
}
void BranchingAlgorithm<span, Node, Edge, Graph>::execute ()
{
/** We get an iterator over all graph nodes. */
GraphIterator<Node> itNodes = _graph->Graph::iterator();
/** We create a custom listener that makes the finish() method, normally called at end of iteration, do nothing this time.
* => we define our own 'finishPostponed' method that is called when all the information is ok. */
CustomListener<Count>* listener = new CustomListener<Count> (
createIteratorListener (itNodes.size(), progressFormat1),
_branchingCollection
);
/** We encapsulate this iterator with a potentially decorated iterated (for progress information). */
tools::dp::Iterator<Node>* iter = createIterator<Node> (
itNodes.get(),
itNodes.size(),
progressFormat1,
listener
);
LOCAL (iter);
/** We get a synchronized object on the data handled by functors. */
ThreadObject <FunctorData<Count,Type> > functorData;
FunctorNodes<Count,Type, Node, Edge, Graph> functorNodes (this->_graph, functorData);
/** We iterate the nodes. */
tools::dp::IDispatcher::Status status = getDispatcher()->iterate (iter, functorNodes);
/** Now, because we iterated with N threads, we have N vector of branching nodes. (N=nbcores used by the dispatcher)
* We need to merge them.
* TODO doc: why do they need to be sorted actually?!
* The next step are:
* 1) sort each vector
* 2) sort/merge the N vectors in the final collection
*/
/** Step 1 : sorting the N branching nodes vectors (with the dispatcher). */
vector<tools::dp::ICommand*> sortCmds;
for (size_t i=0; i<functorData.size(); i++) { sortCmds.push_back (new SortCmd<Count> (functorData[i].branchingNodes)); }
getDispatcher()->dispatchCommands (sortCmds);
/** Step 2 : sort/merge the N vectors.
* We use a priority queue for the merge process. */
typedef typename vector<Count>::iterator BranchingIterator;
typedef pair<BranchingIterator,BranchingIterator> BranchingIteratorPair;
/** We use a cache to improve IO performances. */
CollectionCache<Count> branchingCache (*_branchingCollection, 16*1024, 0);
Type checksum; checksum.setVal( 0);
/** We initialize our priority queue. */
priority_queue <BranchingIteratorPair, vector<BranchingIteratorPair>, Compare<BranchingIteratorPair> > pq;
for (size_t i=0; i<functorData.size(); i++)
{
if (functorData[i].branchingNodes.empty() == false)
{
pq.push (make_pair (functorData[i].branchingNodes.begin(), functorData[i].branchingNodes.end()));
}
}
/** We process the merge/sort. */
while (!pq.empty())
{
/** We get the top iterators pair from the priority queue and remove from it. */
BranchingIteratorPair it = pq.top();
pq.pop();
/** We check that the current iterator is not finished. */
if (it.first != it.second)
{
/** We insert the Count object into the final bag. */
branchingCache.insert (*it.first);
/** Stats */
checksum += it.first->value;
/** We update the priority queue. */
++(it.first); if (it.first != it.second) { pq.push (it); }
}
}
/** We have to flush the cache to be sure every items is put into the cached collection. */
branchingCache.flush ();
/** We call our 'custom' finish method. */
listener->finishPostponed();
/** We save the kind in the storage. */
_storage(getName()).addProperty ("kind", toString(_kind));
/* print the number of branching nodes (could be important for debugging, if a user experiences a crash and copypastes stdout) */
//cout << "Graph has " << _branchingCollection->getNbItems() << " branching nodes." << endl;
/** We gather some statistics. */
getInfo()->add (1, "stats");
getInfo()->add (2, "nb_branching", "%ld", _branchingCollection->getNbItems());
getInfo()->add (2, "percentage", "%.1f", (itNodes.size() > 0 ? 100.0*(float)_branchingCollection->getNbItems()/(float)itNodes.size() : 0));
stringstream ss; ss << checksum;
getInfo()->add (2, "checksum_branching", "%s", ss.str().c_str());
if (getInput()->get(STR_TOPOLOGY_STATS) && getInput()->getInt(STR_TOPOLOGY_STATS) > 0)
{
/** We get some topological information. */
for (size_t i=0; i<functorData.size(); i++)
{
for (map<InOut_t, size_t>::iterator it = functorData[i].topology.begin(); it != functorData[i].topology.end(); ++it)
{
functorData->topology[it->first] += it->second;
}
}
/** We sort the statistics. */
vector < pair<InOut_t,size_t> > topologyStats;
for (map<InOut_t, size_t>::iterator it = functorData->topology.begin(); it != functorData->topology.end(); ++it) { topologyStats.push_back (*it); }
sort (topologyStats.begin(), topologyStats.end(), CompareFct);
getInfo()->add (1, "topology");
for (size_t i=0; i<topologyStats.size(); i++)
{
getInfo()->add (3, "neighborhood", "[in=%ld out=%ld] nb : %8ld percent. : %5.2f",
topologyStats[i].first.first, topologyStats[i].first.second, topologyStats[i].second,
_branchingCollection->getNbItems() > 0 ?
100.0*(float)topologyStats[i].second / (float)_branchingCollection->getNbItems() : 0
);
}
}
getInfo()->add (1, "time");
getInfo()->add (2, "build", "%.3f", status.time / 1000.0);
}
void MPHFAlgorithm<span,Abundance_t,NodeState_t>::populate ()
{
size_t nb_iterated = 0;
size_t n = _abundanceMap->size();
_nb_abundances_above_precision = 0;
/** We need a progress object. */
tools::dp::IteratorListener* delegate = createIteratorListener(_solidCounts->getNbItems(),messages[3]); LOCAL (delegate);
setProgress (new ProgressCustom(delegate));
SubjectIterator<Count>* itKmers = new SubjectIterator<Count> (_solidCounts->iterator(), _solidCounts->getNbItems()/100);
itKmers->addObserver (_progress);
LOCAL (itKmers);
// TODO parallize that
std::vector<int> & _abundanceDiscretization = _abundanceMap->_abundanceDiscretization ;
int max_abundance_discrete = _abundanceDiscretization[_abundanceDiscretization.size()-2];
// set counts and at the same time, test the mphf
for (itKmers->first(); !itKmers->isDone(); itKmers->next())
{
//cout << "kmer: " << itKmers->item().value.toString(21) << std::endl;
/** We get the hash code of the current item. */
typename AbundanceMap::Hash::Code h = _abundanceMap->getCode (itKmers->item().value);
/** Little check. */
if (h >= n) { throw Exception ("MPHF check: value out of bounds"); }
/** We get the abundance of the current kmer. */
int abundance = itKmers->item().abundance;
if (abundance > max_abundance_discrete)
{
_nb_abundances_above_precision++;
abundance = max_abundance_discrete;
}
//get first cell strictly greater than abundance
std::vector<int>::iterator up = std::upper_bound(_abundanceDiscretization.begin(), _abundanceDiscretization.end(), abundance);
up--; // get previous cell
int idx = up- _abundanceDiscretization.begin() ;
/** We set the abundance of the current kmer. */
_abundanceMap->at (h) = idx;
nb_iterated ++;
}
if (nb_iterated != n && n > 3)
{
throw Exception ("ERROR during abundance population: itKmers iterated over %d/%d kmers only", nb_iterated, n);
}
#if 1
// you know what? let's always test if the MPHF does not have collisions, it won't hurt.
check ();
#endif
/** We gather some statistics. */
getInfo()->add (1, "stats");
getInfo()->add (2, "nb_keys", "%ld", _abundanceMap->size());
getInfo()->add (2, "data_size", "%ld", _dataSize);
getInfo()->add (2, "bits_per_key", "%.3f", (float)(_dataSize*8)/(float)_abundanceMap->size());
getInfo()->add (2, "prec", "%d", MAX_ABUNDANCE);
getInfo()->add (2, "nb_abund_above_prec", "%d", _nb_abundances_above_precision);
getInfo()->add (1, getTimeInfo().getProperties("time"));
}
