feat: add vector() method to SourceDocument interface - Add vector() …

…method with default null implementation - Update vector document generators to use vector() with string fallback - Addresses inefficient vector-to-string conversion pattern - See: castorini#2661

src/main/java/io/anserini/collection/SourceDocument.java

@@ -50,4 +50,15 @@ public interface SourceDocument {
    * @return <code>true</code> if this document is meant to be indexed
    */
   boolean indexable();
+
+  /**
+   * Optional method to directly access a document's vector representation without string parsing.
+   * Added to avoid the inefficient pattern of converting vectors to/from strings.
+   * Implementations can override this to provide direct vector access.
+   *
+   * @return float array containing the vector representation, or null if not implemented
+   */
+  default float[] vector() {
+    return null;
+  }
 }

src/main/java/io/anserini/index/generator/JsonDenseVectorDocumentGenerator.java

@@ -58,10 +58,15 @@ public Document createDocument(T src) throws InvalidDocumentException {
     String id = src.id();
     float[] contents;
 
-    try {
-      contents = convertJsonArray(src.contents());
-    } catch (Exception e) {
-      throw new InvalidDocumentException();
+    // Try direct vector access first for efficiency, fall back to string parsing for backward compatibility
+    contents = src.vector();
+
+    if (contents == null) {
+      try {
+        contents = convertJsonArray(src.contents());
+      } catch (Exception e) {
+        throw new InvalidDocumentException();
+      }
     }
 
     // Make a new, empty document.

src/main/java/io/anserini/index/generator/JsonInvertedDenseVectorDocumentGenerator.java

@@ -56,10 +56,15 @@ public Document createDocument(T src) throws InvalidDocumentException {
     String id = src.id();
     float[] contents;
 
-    try {
-      contents = convertJsonArray(src.contents());
-    } catch (Exception e) {
-      throw new InvalidDocumentException();
+    // Try direct vector access first for efficiency, fall back to string parsing for backward compatibility
+    contents = src.vector();
+
+    if (contents == null) {
+      try {
+        contents = convertJsonArray(src.contents());
+      } catch (Exception e) {
+        throw new InvalidDocumentException();
+      }
     }
 
     StringBuilder sb = new StringBuilder();

src/main/java/io/anserini/index/generator/ParquetDenseVectorDocumentGenerator.java

@@ -47,10 +47,16 @@ public class ParquetDenseVectorDocumentGenerator<T extends SourceDocument> imple
    */
   @Override
   public Document createDocument(T src) throws InvalidDocumentException {
-
     try {
-      // Parse vector data from document contents
-      float[] contents = parseVectorFromString(src.contents());
+      float[] contents;
+
+      // Try direct vector access first for efficiency, fall back to string parsing for backward compatibility
+      contents = src.vector();
+
+      if (contents == null) {
+        contents = parseVectorFromString(src.contents());
+      }
+
       if (contents == null || contents.length == 0) {
         LOG.error("Vector data is null or empty for document ID: " + src.id());
         throw new InvalidDocumentException();