0.9

project.clj

@@ -1,4 +1,4 @@
-(defproject techascent/tech.ml-base "0.9-SNAPSHOT"
+(defproject techascent/tech.ml-base "0.9"
   :description "Base concepts of the techascent ml suite"
   :url "http://github.com/tech-ascent/tech.ml-base"
   :license {:name "Eclipse Public License"

src/tech/ml/train.clj

@@ -70,12 +70,14 @@ loss-fn: (loss-fn label-sequence prediction-sequence)-> double
           {:options options
            :error (average-prediction-error
                    (partial train-fn options)
-                   (partial predict-fn options)
+                   predict-fn
                    label-key
                    loss-fn
-                   dataset-seq) }))
+                   dataset-seq)}))
        (reduce (fn [best-map {:keys [options error] :as next-map}]
                  (if (or (not best-map)
                          (< (double error)
                             (double (:error best-map))))
-                   {})))))
+                   next-map
+                   best-map))
+               nil)))

src/tech/verify/ml/train.clj

@@ -19,7 +19,8 @@
                            (take 1000))
         test-dataset (for [x (range -9.9 10 0.1)] {:x x})
         test-labels (map (comp f :x) test-dataset)
-        model (ml/train system-name [:x] :y {:model-type (or model-type :regression)} train-dataset)
+        model (ml/train system-name [:x] :y
+                        {:model-type (or model-type :regression)} train-dataset)
         test-output (ml/predict model test-dataset)
         mse (loss/mse test-output test-labels)]
     (is (< mse (double accuracy)))))
@@ -55,10 +56,35 @@
                      (take 1000))
         feature-keys [:x]
         label :y
-        train-fn (partial ml/train system-name feature-keys label {:model-type :regression})
+        train-fn (partial ml/train system-name feature-keys label
+                          {:model-type :regression})
         predict-fn ml/predict
         mse (->> dataset
                  (dataset/dataset->k-fold-datasets 10 {})
                  (train/average-prediction-error train-fn predict-fn
                                                  label loss/mse))]
     (is (< mse 0.01))))
+
+
+(defn gridsearch
+  [system-name options]
+  (let [f (partial * 2)
+        observe (fn []
+                  (let [x (- (* 20 (rand)) 10)
+                        y (f x)]
+                    {:x x :y y}))
+        dataset (->> (repeatedly observe)
+                     (take 1000))
+        feature-keys [:x]
+        label :y
+        train-fn (partial ml/train system-name feature-keys label)
+        predict-fn ml/predict
+        k-fold-ds (dataset/dataset->k-fold-datasets 5 {} dataset)
+        option-seq [(merge {:model-type :regression} options)
+                    (merge {:model-type :regression} options)]
+        {:keys [error options]} (train/find-best-options train-fn predict-fn
+                                                         label
+                                                         loss/mse {}
+                                                         option-seq k-fold-ds)
+        mse (or (:mse options) 0.01)]
+    (is (< error mse))))