Merge pull request #22996 from vespa-engine/lesters/short-tensor-format-as-default

Lesters/short tensor format as default MERGEOK

8 files changed, 56 insertions, 34 deletions

diff --git a/container-search/src/main/java/com/yahoo/search/query/Presentation.java b/container-search/src/main/java/com/yahoo/search/query/Presentation.java
index 31641b5c2f0..afa87eb4a06 100644
--- a/container-search/src/main/java/com/yahoo/search/query/Presentation.java
+++ b/container-search/src/main/java/com/yahoo/search/query/Presentation.java
@@ -75,7 +75,7 @@ public class Presentation implements Cloneable {
     private boolean timing = false;
 
     /** Whether to renders tensors in short form */
-    private boolean tensorShortForm = false;
+    private boolean tensorShortForm = true;
 
     /** Set of explicitly requested summary fields, instead of summary classes */
     private Set<String> summaryFields = LazySet.newHashSet();
@@ -180,7 +180,7 @@ public class Presentation implements Cloneable {
      * Returns whether tensors should use short form in JSON and textual representations, see
      * <a href="https://docs.vespa.ai/en/reference/document-json-format.html#tensor">https://docs.vespa.ai/en/reference/document-json-format.html#tensor</a>
      * and <a href="https://docs.vespa.ai/en/reference/tensor.html#tensor-literal-form">https://docs.vespa.ai/en/reference/tensor.html#tensor-literal-form</a>.
-     * Default is false.
+     * Default is true.
      */
     public boolean getTensorShortForm() { return tensorShortForm; }
 
diff --git a/container-search/src/test/java/com/yahoo/search/rendering/JsonRendererTestCase.java b/container-search/src/test/java/com/yahoo/search/rendering/JsonRendererTestCase.java
index 118dcee9598..570b1d006de 100644
--- a/container-search/src/test/java/com/yahoo/search/rendering/JsonRendererTestCase.java
+++ b/container-search/src/test/java/com/yahoo/search/rendering/JsonRendererTestCase.java
@@ -209,14 +209,14 @@ public class JsonRendererTestCase {
                 + "                    \"object\": \"thingie\","
                 + "                    \"string\": \"stuff\","
                 + "                    \"predicate\": \"a in [b]\","
-                + "                    \"tensor1\": { \"cells\": [ { \"address\": {\"x\": \"a\"}, \"value\":2.0 } ] },"
+                + "                    \"tensor1\": { \"type\": \"tensor(x{})\", \"cells\": { \"a\":2.0 } },"
                 + "                    \"tensor2\": { \"cells\": [] },"
-                + "                    \"tensor3\": { \"cells\": [ { \"address\": {\"x\": \"a\", \"y\": \"0\"}, \"value\":2.0 }, { \"address\": {\"x\": \"a\", \"y\": \"1\"}, \"value\":-1.0 } ] },"
+                + "                    \"tensor3\": { \"type\": \"tensor(x{},y{})\", \"cells\": [ { \"address\": {\"x\": \"a\", \"y\": \"0\"}, \"value\":2.0 }, { \"address\": {\"x\": \"a\", \"y\": \"1\"}, \"value\":-1.0 } ] },"
                 + "                    \"summaryfeatures\": {"
                 + "                        \"scalar1\":1.5,"
                 + "                        \"scalar2\":2.5,"
-                + "                        \"tensor1\":{\"type\":\"tensor(x[3])\",\"cells\":[{\"address\":{\"x\":\"0\"},\"value\":1.5},{\"address\":{\"x\":\"1\"},\"value\":2.0},{\"address\":{\"x\":\"2\"},\"value\":2.5}]},"
-                + "                        \"tensor2\":{\"type\":\"tensor()\",\"cells\":[{\"address\":{},\"value\":0.5}]}"
+                + "                        \"tensor1\":{\"type\":\"tensor(x[3])\", \"values\":[1.5, 2.0, 2.5] },"
+                + "                        \"tensor2\":{\"type\":\"tensor()\", \"values\":[0.5] }"
                 + "                    },"
                 + "                    \"data\": \"Data \\\\xc3\\\\xa6 \\\\xc3\\\\xa5\""
                 + "                },"
diff --git a/model-evaluation/src/main/java/ai/vespa/models/handler/ModelsEvaluationHandler.java b/model-evaluation/src/main/java/ai/vespa/models/handler/ModelsEvaluationHandler.java
index b0e2be26f8a..2661b9c2eb2 100644
--- a/model-evaluation/src/main/java/ai/vespa/models/handler/ModelsEvaluationHandler.java
+++ b/model-evaluation/src/main/java/ai/vespa/models/handler/ModelsEvaluationHandler.java
@@ -90,13 +90,13 @@ public class ModelsEvaluationHandler extends ThreadedHttpRequestHandler {
         Tensor result = evaluator.evaluate();
 
         Optional<String> format = property(request, "format.tensors");
-        if (format.isPresent() && format.get().equalsIgnoreCase("short")) {
-            return new Response(200, JsonFormat.encodeShortForm(result));
+        if (format.isPresent() && format.get().equalsIgnoreCase("long")) {
+            return new Response(200, JsonFormat.encode(result));
         }
         else if (format.isPresent() && format.get().equalsIgnoreCase("string")) {
             return new Response(200, result.toString().getBytes(StandardCharsets.UTF_8));
         }
-        return new Response(200, JsonFormat.encode(result));
+        return new Response(200, JsonFormat.encodeShortForm(result));
     }
 
     private HttpResponse listAllModels(HttpRequest request) {
diff --git a/model-evaluation/src/test/java/ai/vespa/models/handler/ModelsEvaluationHandlerTest.java b/model-evaluation/src/test/java/ai/vespa/models/handler/ModelsEvaluationHandlerTest.java
index 3000c83b7ec..7790f8a60d0 100644
--- a/model-evaluation/src/test/java/ai/vespa/models/handler/ModelsEvaluationHandlerTest.java
+++ b/model-evaluation/src/test/java/ai/vespa/models/handler/ModelsEvaluationHandlerTest.java
@@ -66,7 +66,7 @@ public class ModelsEvaluationHandlerTest {
     @Test
     public void testXgBoostEvaluationWithoutBindings() {
         String url = "http://localhost/model-evaluation/v1/xgboost_2_2/eval";  // only has a single function
-        String expected = "{\"cells\":[{\"address\":{},\"value\":-4.376589999999999}]}";
+        String expected = "{\"type\":\"tensor()\",\"values\":[-4.376589999999999]}";
         handler.assertResponse(url, 200, expected);
     }
 
@@ -79,7 +79,7 @@ public class ModelsEvaluationHandlerTest {
         properties.put("f109", "0.4");
         properties.put("non-existing-binding", "-1");
         String url = "http://localhost/model-evaluation/v1/xgboost_2_2/eval";
-        String expected = "{\"cells\":[{\"address\":{},\"value\":-7.936679999999999}]}";
+        String expected = "{\"type\":\"tensor()\",\"values\":[-7.936679999999999]}";
         handler.assertResponse(url, properties, 200, expected);
     }
 
@@ -91,6 +91,7 @@ public class ModelsEvaluationHandlerTest {
         properties.put("f60", "0.3");
         properties.put("f109", "0.4");
         properties.put("non-existing-binding", "-1");
+        properties.put("format.tensors", "long");
         String url = "http://localhost/model-evaluation/v1/xgboost_2_2/eval";
         String expected = "{\"cells\":[{\"address\":{},\"value\":-7.936679999999999}]}";
         handler.assertResponse(url, properties, 200, expected);
@@ -99,7 +100,7 @@ public class ModelsEvaluationHandlerTest {
     @Test
     public void testLightGBMEvaluationWithoutBindings() {
         String url = "http://localhost/model-evaluation/v1/lightgbm_regression/eval";
-        String expected = "{\"cells\":[{\"address\":{},\"value\":1.9130086820218188}]}";
+        String expected = "{\"type\":\"tensor()\",\"values\":[1.9130086820218188]}";
         handler.assertResponse(url, 200, expected);
     }
 
@@ -112,7 +113,7 @@ public class ModelsEvaluationHandlerTest {
         properties.put("categorical_2", "i");
         properties.put("non-existing-binding", "-1");
         String url = "http://localhost/model-evaluation/v1/lightgbm_regression/eval";
-        String expected = "{\"cells\":[{\"address\":{},\"value\":2.054697758469921}]}";
+        String expected = "{\"type\":\"tensor()\",\"values\":[2.054697758469921]}";
         handler.assertResponse(url, properties, 200, expected);
     }
 
@@ -125,7 +126,7 @@ public class ModelsEvaluationHandlerTest {
         properties.put("categorical_2", "j");
         properties.put("non-existing-binding", "-1");
         String url = "http://localhost/model-evaluation/v1/lightgbm_regression/eval";
-        String expected = "{\"cells\":[{\"address\":{},\"value\":2.0745534018208094}]}";
+        String expected = "{\"type\":\"tensor()\",\"values\":[2.0745534018208094]}";
         handler.assertResponse(url, properties, 200, expected);
     }
 
@@ -162,7 +163,7 @@ public class ModelsEvaluationHandlerTest {
         Map<String, String> properties = new HashMap<>();
         properties.put("Placeholder", inputTensor());
         String url = "http://localhost/model-evaluation/v1/mnist_softmax/eval";
-        String expected = "{\"cells\":[{\"address\":{\"d0\":\"0\",\"d1\":\"0\"},\"value\":-0.3546536862850189},{\"address\":{\"d0\":\"0\",\"d1\":\"1\"},\"value\":0.3759574592113495},{\"address\":{\"d0\":\"0\",\"d1\":\"2\"},\"value\":0.06054411828517914},{\"address\":{\"d0\":\"0\",\"d1\":\"3\"},\"value\":-0.251544713973999},{\"address\":{\"d0\":\"0\",\"d1\":\"4\"},\"value\":0.017951013520359993},{\"address\":{\"d0\":\"0\",\"d1\":\"5\"},\"value\":1.2899067401885986},{\"address\":{\"d0\":\"0\",\"d1\":\"6\"},\"value\":-0.10389615595340729},{\"address\":{\"d0\":\"0\",\"d1\":\"7\"},\"value\":0.6367976665496826},{\"address\":{\"d0\":\"0\",\"d1\":\"8\"},\"value\":-1.4136744737625122},{\"address\":{\"d0\":\"0\",\"d1\":\"9\"},\"value\":-0.2573896050453186}]}";
+        String expected = "{\"type\":\"tensor(d0[],d1[10])\",\"values\":[[-0.3546536862850189,0.3759574592113495,0.06054411828517914,-0.251544713973999,0.017951013520359993,1.2899067401885986,-0.10389615595340729,0.6367976665496826,-1.4136744737625122,-0.2573896050453186]]}";
         handler.assertResponse(url, properties, 200, expected);
     }
 
@@ -171,7 +172,7 @@ public class ModelsEvaluationHandlerTest {
         Map<String, String> properties = new HashMap<>();
         properties.put("Placeholder", inputTensor());
         String url = "http://localhost/model-evaluation/v1/mnist_softmax/default.add/eval";
-        String expected = "{\"cells\":[{\"address\":{\"d0\":\"0\",\"d1\":\"0\"},\"value\":-0.3546536862850189},{\"address\":{\"d0\":\"0\",\"d1\":\"1\"},\"value\":0.3759574592113495},{\"address\":{\"d0\":\"0\",\"d1\":\"2\"},\"value\":0.06054411828517914},{\"address\":{\"d0\":\"0\",\"d1\":\"3\"},\"value\":-0.251544713973999},{\"address\":{\"d0\":\"0\",\"d1\":\"4\"},\"value\":0.017951013520359993},{\"address\":{\"d0\":\"0\",\"d1\":\"5\"},\"value\":1.2899067401885986},{\"address\":{\"d0\":\"0\",\"d1\":\"6\"},\"value\":-0.10389615595340729},{\"address\":{\"d0\":\"0\",\"d1\":\"7\"},\"value\":0.6367976665496826},{\"address\":{\"d0\":\"0\",\"d1\":\"8\"},\"value\":-1.4136744737625122},{\"address\":{\"d0\":\"0\",\"d1\":\"9\"},\"value\":-0.2573896050453186}]}";
+        String expected = "{\"type\":\"tensor(d0[],d1[10])\",\"values\":[[-0.3546536862850189,0.3759574592113495,0.06054411828517914,-0.251544713973999,0.017951013520359993,1.2899067401885986,-0.10389615595340729,0.6367976665496826,-1.4136744737625122,-0.2573896050453186]]}";
         handler.assertResponse(url, properties, 200, expected);
     }
 
@@ -179,6 +180,7 @@ public class ModelsEvaluationHandlerTest {
     public void testMnistSoftmaxEvaluateSpecificFunctionWithBindingsShortForm() {
         Map<String, String> properties = new HashMap<>();
         properties.put("Placeholder", inputTensorShortForm());
+        properties.put("format.tensors", "long");
         String url = "http://localhost/model-evaluation/v1/mnist_softmax/default.add/eval";
         String expected = "{\"cells\":[{\"address\":{\"d0\":\"0\",\"d1\":\"0\"},\"value\":-0.3546536862850189},{\"address\":{\"d0\":\"0\",\"d1\":\"1\"},\"value\":0.3759574592113495},{\"address\":{\"d0\":\"0\",\"d1\":\"2\"},\"value\":0.06054411828517914},{\"address\":{\"d0\":\"0\",\"d1\":\"3\"},\"value\":-0.251544713973999},{\"address\":{\"d0\":\"0\",\"d1\":\"4\"},\"value\":0.017951013520359993},{\"address\":{\"d0\":\"0\",\"d1\":\"5\"},\"value\":1.2899067401885986},{\"address\":{\"d0\":\"0\",\"d1\":\"6\"},\"value\":-0.10389615595340729},{\"address\":{\"d0\":\"0\",\"d1\":\"7\"},\"value\":0.6367976665496826},{\"address\":{\"d0\":\"0\",\"d1\":\"8\"},\"value\":-1.4136744737625122},{\"address\":{\"d0\":\"0\",\"d1\":\"9\"},\"value\":-0.2573896050453186}]}";
         handler.assertResponse(url, properties, 200, expected);
diff --git a/vespaclient-container-plugin/src/main/java/com/yahoo/document/restapi/resource/DocumentV1ApiHandler.java b/vespaclient-container-plugin/src/main/java/com/yahoo/document/restapi/resource/DocumentV1ApiHandler.java
index 8c2e39d595e..626d4f57c58 100644
--- a/vespaclient-container-plugin/src/main/java/com/yahoo/document/restapi/resource/DocumentV1ApiHandler.java
+++ b/vespaclient-container-plugin/src/main/java/com/yahoo/document/restapi/resource/DocumentV1ApiHandler.java
@@ -713,12 +713,17 @@ public class DocumentV1ApiHandler extends AbstractRequestHandler {
             }
         }
 
-        synchronized void writeSingleDocument(Document document) throws IOException {
-            boolean tensorShortForm = false;
-            if (request != null && request.parameters().containsKey("format.tensors")) {
-                tensorShortForm = request.parameters().get("format.tensors").contains("short");
+        private boolean tensorShortForm() {
+            if (request != null &&
+                request.parameters().containsKey("format.tensors") &&
+                request.parameters().get("format.tensors").contains("long")) {
+                return false;
             }
-            new JsonWriter(json, tensorShortForm).writeFields(document);
+            return true;  // default
+        }
+
+        synchronized void writeSingleDocument(Document document) throws IOException {
+            new JsonWriter(json, tensorShortForm()).writeFields(document);
         }
 
         synchronized void writeDocumentsArrayStart() throws IOException {
@@ -737,7 +742,7 @@ public class DocumentV1ApiHandler extends AbstractRequestHandler {
             ByteArrayOutputStream myOut = new ByteArrayOutputStream(1);
             myOut.write(','); // Prepend rather than append, to avoid double memory copying.
             try (JsonGenerator myJson = jsonFactory.createGenerator(myOut)) {
-                new JsonWriter(myJson).write(document);
+                new JsonWriter(myJson, tensorShortForm()).write(document);
             }
             docs.add(myOut);
 
diff --git a/vespaclient-container-plugin/src/test/cfg/music.sd b/vespaclient-container-plugin/src/test/cfg/music.sd
index 75a1918de5b..86597379dd3 100644
--- a/vespaclient-container-plugin/src/test/cfg/music.sd
+++ b/vespaclient-container-plugin/src/test/cfg/music.sd
@@ -2,5 +2,6 @@
 search music {
     document music {
         field artist type string { }
+        field embedding type tensor(x[3]) {}
     }
 }
diff --git a/vespaclient-container-plugin/src/test/java/com/yahoo/document/restapi/resource/DocumentV1ApiTest.java b/vespaclient-container-plugin/src/test/java/com/yahoo/document/restapi/resource/DocumentV1ApiTest.java
index bdb6c6d5580..74a86b6a7b7 100644
--- a/vespaclient-container-plugin/src/test/java/com/yahoo/document/restapi/resource/DocumentV1ApiTest.java
+++ b/vespaclient-container-plugin/src/test/java/com/yahoo/document/restapi/resource/DocumentV1ApiTest.java
@@ -14,6 +14,7 @@ import com.yahoo.document.FixedBucketSpaces;
 import com.yahoo.document.TestAndSetCondition;
 import com.yahoo.document.config.DocumentmanagerConfig;
 import com.yahoo.document.datatypes.StringFieldValue;
+import com.yahoo.document.datatypes.TensorFieldValue;
 import com.yahoo.document.restapi.DocumentOperationExecutorConfig;
 import com.yahoo.document.restapi.resource.DocumentV1ApiHandler.StorageCluster;
 import com.yahoo.document.update.FieldUpdate;
@@ -51,6 +52,7 @@ import com.yahoo.schema.derived.Deriver;
 import com.yahoo.slime.Inspector;
 import com.yahoo.slime.JsonFormat;
 import com.yahoo.slime.SlimeUtils;
+import com.yahoo.tensor.Tensor;
 import com.yahoo.test.ManualClock;
 import com.yahoo.vdslib.VisitorStatistics;
 import com.yahoo.vespa.config.content.AllClustersBucketSpacesConfig;
@@ -118,7 +120,9 @@ public class DocumentV1ApiTest {
     final Document doc3 = new Document(manager.getDocumentType("music"), "id:space:music:g=a:three");
     {
         doc1.setFieldValue("artist", "Tom Waits");
+        doc1.setFieldValue("embedding", new TensorFieldValue(Tensor.from("tensor(x[3]):[1,2,3]")));
         doc2.setFieldValue("artist", "Asa-Chan & Jun-Ray");
+        doc2.setFieldValue("embedding", new TensorFieldValue(Tensor.from("tensor(x[3]):[4,5,6]")));
     }
 
     final Map<String, StorageCluster> clusters = Map.of("content", new StorageCluster("content",
@@ -229,13 +233,15 @@ public class DocumentV1ApiTest {
                        "    {" +
                        "      \"id\": \"id:space:music::one\"," +
                        "      \"fields\": {" +
-                       "        \"artist\": \"Tom Waits\"" +
+                       "        \"artist\": \"Tom Waits\", " +
+                       "        \"embedding\": { \"type\": \"tensor(x[3])\", \"values\": [1.0,2.0,3.0] } " +
                        "      }" +
                        "    }," +
                        "    {" +
                        "      \"id\": \"id:space:music:n=1:two\"," +
                        "      \"fields\": {" +
-                       "        \"artist\": \"Asa-Chan & Jun-Ray\"" +
+                       "        \"artist\": \"Asa-Chan & Jun-Ray\", " +
+                       "        \"embedding\": { \"type\": \"tensor(x[3])\", \"values\": [4.0,5.0,6.0] } " +
                        "      }" +
                        "    }," +
                        "    {" +
@@ -278,13 +284,15 @@ public class DocumentV1ApiTest {
                        "    {" +
                        "      \"id\": \"id:space:music::one\"," +
                        "      \"fields\": {" +
-                       "        \"artist\": \"Tom Waits\"" +
+                       "        \"artist\": \"Tom Waits\"," +
+                       "        \"embedding\": { \"type\": \"tensor(x[3])\", \"values\": [1.0,2.0,3.0] } " +
                        "      }" +
                        "    }," +
                        "    {" +
                        "      \"id\": \"id:space:music:n=1:two\"," +
                        "      \"fields\": {" +
-                       "        \"artist\": \"Asa-Chan & Jun-Ray\"" +
+                       "        \"artist\": \"Asa-Chan & Jun-Ray\"," +
+                       "        \"embedding\": { \"type\": \"tensor(x[3])\", \"values\": [4.0,5.0,6.0] } " +
                        "      }" +
                        "    }" +
                        "  ]," +
@@ -491,12 +499,13 @@ public class DocumentV1ApiTest {
             parameters.responseHandler().get().handleResponse(new DocumentResponse(0, doc1));
             return new Result();
         });
-        response = driver.sendRequest("http://localhost/document/v1/space/music/docid/one?");
+        response = driver.sendRequest("http://localhost/document/v1/space/music/docid/one?format.tensors=long");
         assertSameJson("{" +
                        "  \"pathId\": \"/document/v1/space/music/docid/one\"," +
                        "  \"id\": \"id:space:music::one\"," +
                        "  \"fields\": {" +
-                       "    \"artist\": \"Tom Waits\"" +
+                       "    \"artist\": \"Tom Waits\"," +
+                       "    \"embedding\": { \"cells\": [{\"address\":{\"x\":\"0\"},\"value\":1.0},{\"address\":{\"x\":\"1\"},\"value\": 2.0},{\"address\":{\"x\":\"2\"},\"value\": 3.0}]}" +
                        "  }" +
                        "}", response.readAll());
         assertEquals(200, response.getStatus());
@@ -532,7 +541,8 @@ public class DocumentV1ApiTest {
         response = driver.sendRequest("http://localhost/document/v1/space/music/number/1/two?condition=test%20it&tracelevel=9", POST,
                                       "{" +
                                       "  \"fields\": {" +
-                                      "    \"artist\": \"Asa-Chan & Jun-Ray\"" +
+                                      "    \"artist\": \"Asa-Chan & Jun-Ray\"," +
+                                      "    \"embedding\": { \"values\": [4.0,5.0,6.0] } " +
                                       "  }" +
                                       "}");
         assertSameJson("{" +
diff --git a/vespajlib/src/main/java/com/yahoo/tensor/serialization/JsonFormat.java b/vespajlib/src/main/java/com/yahoo/tensor/serialization/JsonFormat.java
index d7758f309db..b683519988c 100644
--- a/vespajlib/src/main/java/com/yahoo/tensor/serialization/JsonFormat.java
+++ b/vespajlib/src/main/java/com/yahoo/tensor/serialization/JsonFormat.java
@@ -109,11 +109,15 @@ public class JsonFormat {
 
     private static void encodeValues(IndexedTensor tensor, Cursor cursor, long[] indexes, int dimension) {
         DimensionSizes sizes = tensor.dimensionSizes();
-        for (indexes[dimension] = 0; indexes[dimension] < sizes.size(dimension); ++indexes[dimension]) {
-            if (dimension < (sizes.dimensions() - 1)) {
-                encodeValues(tensor, cursor.addArray(), indexes, dimension + 1);
-            } else {
-                cursor.addDouble(tensor.get(indexes));
+        if (indexes.length == 0) {
+            cursor.addDouble(tensor.get(0));
+        } else {
+            for (indexes[dimension] = 0; indexes[dimension] < sizes.size(dimension); ++indexes[dimension]) {
+                if (dimension < (sizes.dimensions() - 1)) {
+                    encodeValues(tensor, cursor.addArray(), indexes, dimension + 1);
+                } else {
+                    cursor.addDouble(tensor.get(indexes));
+                }
             }
         }
     }