Add note about splitting in unsupervised learning notebook

examples/Unsupervised Learning.ipynb

@@ -256,7 +256,9 @@
    "source": [
     "# Unsupervised Learning\n",
     "\n",
-    "Our strategy here will be to take some per-particle descriptors, generate the minimum spanning tree of the connectivity network, and snip bonds until we have enough connected components to give an idea of some statistics for the samples.\n",
+    "Our strategy here will be to take some per-particle descriptors, generate the minimum spanning tree of the connectivity network, and snip bonds until we have enough connected components to give an idea of some statistics for the samples. \n",
+    "\n",
+    "**Important**: This notebook is set up to use the same number of components for each system. It is possible that averaging descriptors over systems of different sizes can give different distributions in the averaged descriptor values; if you are using systems with varying numbers of particles, you should instead split each system into a number of components proportional to the number of particles to remedy this.\n",
     "\n",
     "We start by generating the per-particle descriptors that will be fed into the network."
    ]