fixes

notebooks/explore.ipynb

@@ -1045,7 +1045,7 @@
     "- k-nearest neighbors\n",
     "- bond energy minimization (via Metropolis algorithm)\n",
     "\n",
-    "These strategies can also be used to generate entirely novel grids. They follow the `numpy` and `scipy` conventions in exposing two primary methods, `fit()` and `predict()`. Each model\n",
+    "These strategies can also be used to generate entirely novel grids. They follow the `numpy` and `scipy` conventions in exposing two primary methods, `fit()` and `predict()`.\n",
     "\n",
     "### Random neighbor selection\n",
     "\n",
@@ -1325,7 +1325,7 @@
    "source": [
     "### K-nearest neighbors\n",
     "\n",
-    "The k-nearest neighbors model attempts to find cell neighborhoods similar to the location to be predicted, ranks them according to similarity, and selects the most common central cell value from a subset of the most similar neighborhoods (ties are broken randomly)."
+    "The k-nearest neighbors model attempts to predict cell values based on neighborhoods similar to the location to be predicted. Neighborhood similarity is evaluated by mapping each neighborhood to a fixed-length string, in which each neighboring location corresponds to a particular character, and computing the Hamming distance between strings. The algorithm then ranks neighborhoods according to similarity to the neighborhood of the cell to be predicted, retaining a small subset of the most similar neighborhoods, then selects the most common central cell value from the subset (ties are broken randomly)."
    ],
    "metadata": {
     "collapsed": false
@@ -1483,7 +1483,7 @@
   {
    "cell_type": "markdown",
    "source": [
-    "KNN can also be used to generate entirely novel grids, but this is subject to the same limitation faced by RNS: when a cell has no known neighbors, the cell value is randomly selected according to the known class distribution (preserving no spatial from the training set)."
+    "KNN can also be used to generate entirely novel grids, but this is subject to the same limitation faced by RNS: when a cell has no known neighbors, the cell value is randomly selected according to the known class distribution (preserving no spatial information from the training set)."
    ],
    "metadata": {
     "collapsed": false