Merge pull request #3352 from liyun95/v2.6.x

update docs

Author: liyun95

assets/scann.png

@@ -107,15 +107,17 @@ The following table lists the parameters that can be configured in `params` when
      <th><p>Value Range</p></th>
      <th><p>Tuning Suggestion</p></th>
    </tr>
+   <tr>
+     <td><p><code>nlist</code></p></td>
+     <td><p>Number of cluster units</p></td>
+     <td><p>[1, 65536]</p></td>
+     <td><p>A higher <em>nlist</em> increases pruning efficiency and typically speeds up coarse search, but partitions can get too small, which may reduce recall; a lower <em>nlist</em> scans larger clusters, improving recall but slowing search.</p></td>
+   </tr>
    <tr>
      <td><p><code>with_raw_data</code></p></td>
      <td><p>Whether to store the original vector data alongside the quantized representation. When enabled, this allows for more accurate similarity calculations during the re-ranking phase by using the original vectors instead of quantized approximations.</p></td>
-     <td><p><strong>Type</strong>: Boolean
- <strong>Range</strong>: <code>true</code>, <code>false</code></p>
-<p><strong>Default value</strong>: <code>true</code></p></td>
-     <td><p>Set to <code>true</code> for <strong>higher search accuracy</strong> and when storage space is not a primary concern. The original vector data enables more precise similarity calculations during re-ranking.
- Set to <code>false</code> to <strong>reduce storage overhead</strong> and memory usage, especially for large datasets. However, this may result in slightly lower search accuracy as the re-ranking phase will use quantized vectors.</p>
-<p><strong>Recommended</strong>: Use <code>true</code> for production applications where accuracy is critical.</p></td>
+     <td><p><strong>Type</strong>: Boolean</p><p><strong>Range</strong>: <code>true</code>, <code>false</code></p><p><strong>Default value</strong>: <code>true</code></p></td>
+     <td><p>Set to <code>true</code> for <strong>higher search accuracy</strong> and when storage space is not a primary concern. The original vector data enables more precise similarity calculations during re-ranking.</p><p>Set to <code>false</code> to <strong>reduce storage overhead</strong> and memory usage, especially for large datasets. However, this may result in slightly lower search accuracy as the re-ranking phase will use quantized vectors.</p><p><strong>Recommended</strong>: Use <code>true</code> for production applications where accuracy is critical.</p></td>
    </tr>
 </table>
 
@@ -133,13 +135,8 @@ The following table lists the parameters that can be configured in `search_param
    <tr>
      <td><p><code>reorder_k</code></p></td>
      <td><p>Controls the number of candidate vectors that are refined during the re-ranking phase. This parameter determines how many top candidates from the initial partitioning and quantization stages are re-evaluated using more precise similarity calculations.</p></td>
-     <td><p><strong>Type</strong>: Integer
- <strong>Range</strong>: [1, <em>int_max</em>]</p>
-<p><strong>Default value</strong>: None</p></td>
-     <td><p>A larger <code>reorder_k</code> generally leads to <strong>higher search accuracy</strong> as more candidates are considered during the final refinement phase. However, this also <strong>increases search time</strong> due to additional computation.
- Consider increasing <code>reorder_k</code> when achieving high recall is critical and search speed is less of a concern. A good starting point is 2-5x your desired <code>limit</code> (TopK results to return).</p>
-<p>Consider decreasing <code>reorder_k</code> to prioritize faster searches, especially in scenarios where a slight reduction in accuracy is acceptable.</p>
-<p>In most cases, we recommend you set a value within this range: [<em>limit</em>, <em>limit</em> * 5].</p></td>
+     <td><p><strong>Type</strong>: Integer</p><p><strong>Range</strong>: [1, <em>int_max</em>]</p><p><strong>Default value</strong>: None</p></td>
+     <td><p>A larger <code>reorder_k</code> generally leads to <strong>higher search accuracy</strong> as more candidates are considered during the final refinement phase. However, this also <strong>increases search time</strong> due to additional computation.</p><p>Consider increasing <code>reorder_k</code> when achieving high recall is critical and search speed is less of a concern. A good starting point is 2-5x your desired <code>limit</code> (TopK results to return).</p><p>Consider decreasing <code>reorder_k</code> to prioritize faster searches, especially in scenarios where a slight reduction in accuracy is acceptable.</p><p>In most cases, we recommend you set a value within this range: [<em>limit</em>, <em>limit</em> * 5].</p></td>
    </tr>
 </table>
 

site/en/userGuide/indexes/floating-vector/scann.md

@@ -2,6 +2,7 @@
 id: geometry-operators.md
 title: "Geometry Operators"
 summary: "Milvus supports a set of operators for spatial filtering on GEOMETRY fields, which are essential for managing and analyzing geometric data. These operators allow you to retrieve entities based on the geometric relationships between objects."
+beta: Milvus 2.6.4+
 ---
 
 # Geometry Operators
@@ -12,6 +13,24 @@ All geometry operators function by taking two geometric arguments: the name of t
 
 To learn more about `GEOMETRY` fields in Milvus, refer to [Geometry Field](geometry-field.md).
 
+## Use syntax
+
+To filter on a `GEOMETRY` field, use a geometry operator in an expression:
+
+- General: `{operator}(geo_field, '{wkt}')`
+
+- Distance-based: `ST_DWITHIN(geo_field, '{wkt}', distance)`
+
+Where:
+
+- `operator` is one of the supported geometry operators (e.g., `ST_CONTAINS`, `ST_INTERSECTS`). Operator names must be all uppercase or all lowercase. For a list of supported operators, refer to [Supported geometry operators](geometry-operators.md#Supported-geometry-operators).
+
+- `geo_field` is the name of your `GEOMETRY` field.
+
+- `'{wkt}'` is the WKT representation of the geometry to query.
+
+- `distance` is the threshold specifically for `ST_DWITHIN`.
+
 ## Supported geometry operators
 
 The following table lists the geometry operators available in Milvus.