Update linear regression to train/predict non-log

src/com/owoga/prhyme/util/math.clj

@@ -134,13 +134,15 @@
   estimate. It defines significantly different as exceeding 1.65 times the
   standard deviation of the Turing estimate."
   [xs ys]
-  (let [n (count xs)
+  (let [log-xs (map #(Math/log %) xs)
-        sum-x (apply + xs)
+        log-ys (map #(Math/log %) ys)
-        sum-y (apply + ys)
+        n (count xs)
+        sum-x (apply + log-xs)
+        sum-y (apply + log-ys)
         mean-x (/ sum-x n)
         mean-y (/ sum-y n)
-        err-x (map #(- % mean-x) xs)
+        err-x (map #(- % mean-x) log-xs)
-        err-y (map #(- % mean-y) ys)
+        err-y (map #(- % mean-y) log-ys)
         err-x-sqr (map #(* % %) err-x)
         m (/ (apply + (map #(apply * %) (map vector err-x err-y)))
              (apply + err-x-sqr))

test/com/owoga/prhyme/util/math_test.clj

@@ -41,10 +41,8 @@
   (t/testing "The results of the linear regression model are accurate"
     (let [r-coll  [1  2  3 5 10]
           zr-coll [20 10 5 1 2] ;; not really smoothed, but smoothing isn't under test
-          log-r (map #(Math/log %) r-coll)
+          linear-model (math/least-squares-linear-regression r-coll zr-coll)
-          log-zr (map #(Math/log %) zr-coll)
+          linear-results (map linear-model r-coll)]
-          linear-model (math/least-squares-linear-regression log-r log-zr)
-          linear-results (map linear-model (map #(Math/log %) r-coll))]
       (t/is (every?
              (fn [[expected predicted]]
                (approx= expected predicted 0.01))
@@ -55,7 +53,7 @@
                 4.81
                 2.59
                 1.12)
-              (map #(Math/pow Math/E %) linear-results)))))))
+              linear-results))))))
 
 ;; The below passes a sanity check in that each r* is slightly less than r.
 #_(t/deftest turing-estimation