Normalize cheater range for better consistency

Author: rreusser

src/traces/carpet/cheater_basis.js

@@ -35,12 +35,30 @@ module.exports = function(a, b, cheaterslope) {
         bscal = (bdata.length - 1) / (bdata[bdata.length - 1] - bdata[0]) / (nb - 1);
     }
 
+    var xval;
+    var xmin = Infinity;
+    var xmax = -Infinity;
     for(j = 0; j < nb; j++) {
         data[j] = [];
         bval = bdata ? (bdata[j] - bdata[0]) * bscal : j / (nb - 1);
         for(i = 0; i < na; i++) {
             aval = adata ? (adata[i] - adata[0]) * ascal : i / (na - 1);
-            data[j][i] = aval - bval * cheaterslope;
+            xval = aval - bval * cheaterslope;
+            xmin = Math.min(xval, xmin);
+            xmax = Math.max(xval, xmax);
+            data[j][i] = xval;
+        }
+    }
+
+    // Normalize cheater values to the 0-1 range. This comes into play when you have
+    // multiple cheater plots. After careful consideration, it seems better if cheater
+    // values are normalized to a consistent range. Otherwise one cheater affects the
+    // layout of other cheaters on the same axis.
+    var slope = 1.0 / (xmax - xmin);
+    var offset = -xmin * slope;
+    for(j = 0; j < nb; j++) {
+        for(i = 0; i < na; i++) {
+            data[j][i] = slope * data[j][i] + offset;
         }
     }