TM getLeastUsedCells_() return 1st least used cell

from those "least". Not random. Slight optimization

src/examples/hello/HelloSPTP.cpp

@@ -213,12 +213,12 @@ EPOCHS = 2; // make test faster in Debug
 
       SDR goldTM({COLS});
       const SDR_sparse_t deterministicTM{
-        72, 85, 102, 114, 122, 126, 287, 308, 337, 339, 542, 920, 939, 952, 1268, 1507, 1508, 1518, 1546, 1547, 1626, 1627, 1633, 1668, 1727, 1804, 1805, 1827, 1832, 1844, 1859, 1862, 1918, 1920, 1924, 1931, 1933, 1945, 1961, 1965, 1966, 1968, 1970, 1973, 1975, 1976, 1977, 1979, 1986, 1987, 1991, 1992, 1996, 1998, 2002, 2006, 2008, 2012, 2042, 2045
+        36, 85, 118, 263, 287, 303, 308, 322, 336, 337, 339, 370, 432, 1115, 1147, 1214, 1508, 1512, 1518, 1523, 1626, 1668, 1691, 1694, 1729, 1781, 1797, 1798, 1803, 1827, 1832, 1844, 1858, 1859, 1860, 1861, 1862, 1917, 1929, 1936, 1939, 1941, 1943, 1947, 1950, 1953, 1956, 1958, 1964, 1965, 1967, 1971, 1973, 1976, 1984, 1985, 1987, 1994, 1996, 1997, 1998, 1999, 2002,2006, 2012, 2027, 2040, 2042
       };
       goldTM.setSparse(deterministicTM);
 
-      const float goldAn    = 0.637255f; //Note: this value is for a (randomly picked) datapoint, it does not have to improve (decrease) with better algorithms
-      const float goldAnAvg = 0.40804f; // ...the averaged value, on the other hand, should improve/decrease. 
+      const float goldAn    = 0.745098f; //Note: this value is for a (randomly picked) datapoint, it does not have to improve (decrease) with better algorithms
+      const float goldAnAvg = 0.408207f; // ...the averaged value, on the other hand, should improve/decrease. 
 
 #ifdef _ARCH_DETERMINISTIC
       if(e+1 == 5000) {

src/htm/algorithms/TemporalMemory.cpp

@@ -141,19 +141,16 @@ void TemporalMemory::initialize(
   reset();
 }
 
-CellIdx TemporalMemory::getLeastUsedCell_(const CellIdx column) {
+CellIdx TemporalMemory::getLeastUsedCell_(const CellIdx column) const {
   if(cellsPerColumn_ == 1) return column;
 
   vector<CellIdx> cells = cellsForColumn(column);
-
-  //TODO: decide if we need to choose randomly from the "least used" cells, or if 1st is fine. 
-  //In that case the line below is not needed, and this method can become const, deterministic results in tests need to be updated
-  //un/comment line below: 
-  rng_.shuffle(cells.begin(), cells.end()); //as min_element selects 1st minimal element, and we want to randomly choose 1 from the minimals. //TODO return the 1st
+  //Note: from the found "least used cells" (if there are more), choose just 1st, not randomly  
+  // or un/comment line below: 
+  //rng_.shuffle(cells.begin(), cells.end()); //as min_element selects 1st minimal element, and we want to randomly choose 1 from the minimals.
 
   const auto compareByNumSegments = [&](const CellIdx a, const CellIdx b) {
-    if(connections.numSegments(a) == connections.numSegments(b)) 
-      return a < b; //TODO rm? 
+    if(connections.numSegments(a) == connections.numSegments(b)) return a < b; //TODO rm? 
     else return connections.numSegments(a) < connections.numSegments(b);
   };
   return *std::min_element(cells.begin(), cells.end(), compareByNumSegments);
@@ -171,12 +168,13 @@ void TemporalMemory::growSynapses_(
   const size_t nActual = std::min(static_cast<size_t>(nDesiredNewSynapses) + (size_t)connections.numSynapses(segment), (size_t)maxSynapsesPerSegment_); //even with the new additions, synapses fit to segment's limit
 
   // Pick nActual cells randomly.
-  rng_.shuffle(candidates.begin(), candidates.end()); //TODO use sample()
+  rng_.shuffle(candidates.begin(), candidates.end()); 
   for (const auto syn : candidates) {
     // #COND: this loop finishes two folds: a) we ran out of candidates (above), b) we grew the desired number of new synapses (below)
     if(connections.numSynapses(segment) == nActual) break; //this break is also used because conn.createSynapse() can "exit early" if a syn already exists, this IF handles that case too.
     connections_.createSynapse(segment, syn, initialPermanence_); //TODO createSynapse consider creating a vector of new synapses at once?
   }
+  NTA_ASSERT(connections.numSynapses(segment) == nActual);
 }
 
 

src/htm/algorithms/TemporalMemory.hpp

@@ -660,7 +660,7 @@ class TemporalMemory : public Serializable
 		     const SynapseIdx nDesiredNewSynapses,
 		     const vector<CellIdx> &prevWinnerCells);
 
-  CellIdx getLeastUsedCell_(const CellIdx column);
+  CellIdx getLeastUsedCell_(const CellIdx column) const;
 
   void calculateAnomalyScore_(const SDR &activeColumns);