Call find_bmu method after last training iteration

This issue was affecting some visualizations such as BmuHitsView, which
was calling som._bmu[0] to get the number of observations allocated to
each unit as their BMU.

Tests were made which compared the new output of BmuHitsView with
som.project_data() output. They matched as expected.

Fixes #83

sompy/sompy.py

@@ -352,7 +352,8 @@ def _batchtrain(self, trainlen, radiusin, radiusfin, njob=1,
                 logging.info("nan quantization error, exit train\n")
 
                 #sys.exit("quantization error=nan, exit train")
-
+
+        bmu = self.find_bmu(data, njb=njob)
         bmu[1] = np.sqrt(bmu[1] + fixed_euclidean_x2)
         self._bmu = bmu