Sigmoid and relu are now native functions

2 files changed, 3 insertions, 16 deletions

diff --git a/sample-apps/blog-recommendation/src/main/application/searchdefinitions/blog_post.sd b/sample-apps/blog-recommendation/src/main/application/searchdefinitions/blog_post.sd
index 9011a559f1e..5069b9302eb 100644
--- a/sample-apps/blog-recommendation/src/main/application/searchdefinitions/blog_post.sd
+++ b/sample-apps/blog-recommendation/src/main/application/searchdefinitions/blog_post.sd
@@ -55,7 +55,7 @@ search blog_post {
 
     }
 
-    rank-profile tensor {
+    rank-profile default {
         first-phase {
             expression {
                 sum(query(user_item_cf) * attribute(user_item_cf))
@@ -65,7 +65,7 @@ search blog_post {
 
     constant W_hidden {
         file: constants/W_hidden.json
-        type: tensor(input[20],hidden[40]) # user_item_cf
+        type: tensor(input[20],hidden[40])
     }
 
     constant b_hidden {
@@ -97,16 +97,6 @@ search blog_post {
         expression: sum(x * y, dim)
     }
 
-    # Not valid. Needs to be a predefined function.
-    macro relu(x) {
-        expression: max(0, x) # 0 should be a constant tensor with 0 values
-    }
-
-    # Not valid. Needs to be predefined function.
-    macro sigmoid(x) {
-        expression: 1 / (1 + exp(-x))
-    }
-
     # The input to the neural network is the concatenation of the document and query vectors
     macro nn_input() {
         expression {
diff --git a/sample-apps/blog-recommendation/src/main/java/com/yahoo/example/BlogTensorSearcher.java b/sample-apps/blog-recommendation/src/main/java/com/yahoo/example/BlogTensorSearcher.java
index 5bbd7140cb8..c142c0e0986 100644
--- a/sample-apps/blog-recommendation/src/main/java/com/yahoo/example/BlogTensorSearcher.java
+++ b/sample-apps/blog-recommendation/src/main/java/com/yahoo/example/BlogTensorSearcher.java
@@ -35,10 +35,7 @@ public class BlogTensorSearcher extends Searcher {
             }
             QueryTreeUtil.andQueryItemWithRoot(query, notItem);
 
-            // Modify the ranking by using the 'tensor' rank-profile (as defined in blog_post.sd)...
-            query.properties().set(new CompoundName("ranking"), "tensor");
-
-            // ... and setting 'query(user_item_cf)' used in that rank-profile
+            // Send the user profile to the ranking framework.
             query.getRanking().getFeatures().put("query(user_item_cf)", toTensor(userItemCfProperty));
         }