float KernelKmeansClusterer::distanceToCluster(int sampleId,
const boost::unordered_set<int>& clusterIdSet) {
if (clusterIdSet.empty()) {
return FLT_MAX;
}
// the sum of weight of the cluster
float weightSum = 0.0f;
// the sum of kernels from the sample to all samples in the cluster, multiplied by the weight
float kernelSum = 0.0f;
// the sum of kernels from each pair of samples in the cluster, multiplied by weights
float kernelPairSum = 0.0f;
for (boost::unordered_set<int>::const_iterator iter = clusterIdSet.begin();
iter != clusterIdSet.end(); ++iter) {
float weight = mWeightArray[*iter];
weightSum += weight;
float kernelResult = (*mpKernel)(sampleId, *iter);
kernelSum += weight * kernelResult;
for (boost::unordered_set<int>::const_iterator _iter =
clusterIdSet.begin(); _iter != clusterIdSet.end(); ++_iter) {
float _kernelResult = (*mpKernel)(*iter, *_iter);
kernelPairSum += weight * mWeightArray[*_iter] * _kernelResult;
}
}
float kernelSelf = (*mpKernel)(sampleId, sampleId);
if (weightSum == 0.0f) {
return FLT_MAX;
}
return kernelSelf - 2 * kernelSum / weightSum
+ kernelPairSum / (weightSum * weightSum);
}
inline Observer& Observer::operator=(const Observer& o) {
iterator i;
for (i=observables_.begin(); i!=observables_.end(); ++i)
(*i)->unregisterObserver(this);
observables_ = o.observables_;
for (i=observables_.begin(); i!=observables_.end(); ++i)
(*i)->registerObserver(this);
return *this;
}
inline void ObservableSettings::registerDeferredObservers(boost::unordered_set<Observer*>& observers) {
if (updatesDeferred())
{
QL_TRACE("adding " << observers.size() << " observers to the deferred list");
deferredObservers_.insert(observers.begin(), observers.end());
}
}
//calculates average of vectors
boost::unordered_map<string, double> Cluster::_centroid(boost::unordered_set<int>& cluster_list) {
unordered_string_map centroid;
int cluster_size = cluster_list.size();
for (boost::unordered_set<int>::iterator it = cluster_list.begin(); it != cluster_list.end(); ++it) {
//add each element of vector to centroid
for (unordered_string_map::iterator ivec = doc_term_index[ *it ].begin(); ivec != doc_term_index[ *it ].end(); ++ivec) {
//element (term) doesn't exist, we add it to the map
if ( centroid.count ( ivec->first ) == 0) {
//cout << "ivec second: " << ivec->second <<endl;
//cout << "cluster_size: " << cluster_size <<endl;
centroid.insert(unordered_string_map::value_type( ivec->first, double( (double)ivec->second / cluster_size ) ));
}
else {
centroid[ivec->first] += double( ((double)ivec->second / cluster_size) );
}
}
}
return centroid;
}
inline void ObservableSettings::enableUpdates() {
updatesEnabled_ = true;
updatesDeferred_ = false;
// if there are outstanding deferred updates, do the notification
if (deferredObservers_.size() > 0)
{
bool successful = true;
std::string errMsg;
QL_TRACE("deferred notification of " << deferredObservers_.size() << " observers");
for (iterator i=deferredObservers_.begin(); i!=deferredObservers_.end(); ++i) {
try {
(*i)->update();
} catch (std::exception& e) {
successful = false;
errMsg = e.what();
} catch (...) {
successful = false;
}
}
deferredObservers_.clear();
QL_ENSURE(successful,
"could not notify one or more observers: " << errMsg);
}
}
static void serialize(storage_type& s, const boost::unordered_set<T>& o)
{
uint32_t sz = o.size();
s.serialize(sz);
for (auto it = o.begin(), it_end = o.end(); it != it_end; ++it)
s.serialize(*it);
}
Strings get_live_object_names() {
IMP::Vector<std::string> ret;
for (boost::unordered_set<Object*>::const_iterator it = live_.begin();
it != live_.end(); ++it) {
ret.push_back((*it)->get_name());
}
return ret;
}
IMPKERNEL_END_NAMESPACE
IMPKERNEL_BEGIN_INTERNAL_NAMESPACE
void check_live_objects() {
for (boost::unordered_set<Object*>::const_iterator it = live_.begin();
it != live_.end(); ++it) {
IMP_USAGE_CHECK((*it)->get_ref_count() > 0,
"Object " << (*it)->get_name() << " is not ref counted.");
}
}
IMPKERNEL_END_NAMESPACE
IMPKERNEL_BEGIN_INTERNAL_NAMESPACE
void check_live_objects() {
for (boost::unordered_set<Object*>::const_iterator it = live_.begin();
it != live_.end(); ++it) {
IMP_USAGE_CHECK((*it)->get_ref_count() > 0,
"Object " << (*it)->get_name()
<< " is alive but not ref counted - memory leakage possible."
<< " This usually happens if an owning pointer is released"
<< " without being either deleted manually or assigned to"
<< " another owning pointer.");
}
}
void save(Archive & ar, const boost::unordered_set<T> & t, unsigned int version)
{
// write the size
// TODO: should we handle bucket size as well?
typedef typename boost::unordered_set<T>::size_type size_type;
typedef typename boost::unordered_set<T>::const_iterator const_iterator;
size_type size = t.size();
ar & BOOST_SERIALIZATION_NVP(size);
unsigned int count = 0;
for (const_iterator it = t.begin(); it != t.end(); it++) {
ar & boost::serialization::make_nvp("item" + count, *it);
count++;
}
}
inline void Observable::notifyObservers() {
// check whether the notifications should be triggered
if (!settings_.updatesEnabled())
{
// if updates are only deferred, flag this for later notification
// these are held centrally by the settings singleton
if (settings_.updatesDeferred())
settings_.registerDeferredObservers(observers_);
return;
}
if (observers_.size() > 0)
{
bool successful = true;
std::string errMsg;
QL_TRACE("direct notification of " << observers_.size() << " observers");
for (iterator i=observers_.begin(); i!=observers_.end(); ++i) {
try {
(*i)->update();
} catch (std::exception& e) {
// quite a dilemma. If we don't catch the exception,
// other observers will not receive the notification
// and might be left in an incorrect state. If we do
// catch it and continue the loop (as we do here) we
// lose the exception. The least evil might be to try
// and notify all observers, while raising an
// exception if something bad happened.
successful = false;
errMsg = e.what();
} catch (...) {
successful = false;
}
}
QL_ENSURE(successful,
"could not notify one or more observers: " << errMsg);
}
}
Objects get_live_objects() {
Objects ret(live_.begin(), live_.end());
return ret;
}
static void exec(OutArcType& oarc, const boost::unordered_set<T>& vec){
serialize_iterator(oarc,
vec.begin(), vec.end(), vec.size());
}
void insert(const boost::unordered_set<key_type>& inKeys, const boost::unordered_set<value_type>& inValues)
{
for (auto it = inValues.begin(); it != inValues.end();++it)
insert(inKeys, *it);
}
/**
* Return the mirror information of the vertex in the i'th position.
*/
inline const std::vector<graph_shard_id_t> mirrors(size_t i) const {
const boost::unordered_set<graph_shard_id_t> mirrorset = shard_impl.vertex_mirrors[i];
std::vector<graph_shard_id_t> ret(mirrorset.begin(), mirrorset.end());
return ret;
}
inline void Observer::unregisterWithAll() {
for (iterator i=observables_.begin(); i!=observables_.end(); ++i)
(*i)->unregisterObserver(this);
observables_.clear();
}
inline Observer::~Observer() {
for (iterator i=observables_.begin(); i!=observables_.end(); ++i)
(*i)->unregisterObserver(this);
}
void drop(const boost::unordered_set<key_type>& inKey, const value_type& inValue)
{
for (auto it = inKey.begin(); it != inKey.end();++it)
drop(*it, inValue);
}
inline Observer::Observer(const Observer& o)
: observables_(o.observables_) {
for (iterator i=observables_.begin(); i!=observables_.end(); ++i)
(*i)->registerObserver(this);
}
inline void ObservableSettings::registerDeferredObservers(
const boost::unordered_set<Observer*>& observers) {
if (updatesDeferred()) {
deferredObservers_.insert(observers.begin(), observers.end());
}
}
void drop(const key_type& inKey, const boost::unordered_set<value_type>& inValue)
{
for (auto it = inValue.begin(); it != inValue.end();++it)
drop(inKey, *it);
}
