|
| 1 | +--- |
| 2 | +categories: |
| 3 | +- docs |
| 4 | +- develop |
| 5 | +- stack |
| 6 | +- oss |
| 7 | +- rs |
| 8 | +- rc |
| 9 | +- oss |
| 10 | +- kubernetes |
| 11 | +- clients |
| 12 | +description: Learn how to index and query vector embeddings with Redis |
| 13 | +linkTitle: Index and query vectors |
| 14 | +title: Index and query vectors |
| 15 | +weight: 30 |
| 16 | +--- |
| 17 | + |
| 18 | +[Redis Query Engine]({{< relref "/develop/interact/search-and-query" >}}) |
| 19 | +lets you index vector fields in [hash]({{< relref "/develop/data-types/hashes" >}}) |
| 20 | +or [JSON]({{< relref "/develop/data-types/json" >}}) objects (see the |
| 21 | +[Vectors]({{< relref "/develop/interact/search-and-query/advanced-concepts/vectors" >}}) |
| 22 | +reference page for more information). |
| 23 | +Among other things, vector fields can store *text embeddings*, which are AI-generated vector |
| 24 | +representations of the semantic information in pieces of text. The |
| 25 | +[vector distance]({{< relref "/develop/interact/search-and-query/advanced-concepts/vectors#distance-metrics" >}}) |
| 26 | +between two embeddings indicates how similar they are semantically. By comparing the |
| 27 | +similarity of an embedding generated from some query text with embeddings stored in hash |
| 28 | +or JSON fields, Redis can retrieve documents that closely match the query in terms |
| 29 | +of their meaning. |
| 30 | + |
| 31 | +The example below uses the [HuggingFace](https://huggingface.co/) model |
| 32 | +[`all-MiniLM-L6-v2`](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) |
| 33 | +to generate the vector embeddings to store and index with Redis Query Engine. |
| 34 | + |
| 35 | +## Initialize |
| 36 | + |
| 37 | +You can use the [TransformersPHP](https://transformers.codewithkyrian.com/) |
| 38 | +library to create the vector embeddings. Install the library with the following |
| 39 | +command: |
| 40 | + |
| 41 | +```bash |
| 42 | +composer require codewithkyrian/transformers |
| 43 | +``` |
| 44 | + |
| 45 | +## Import dependencies |
| 46 | + |
| 47 | +Import the following classes and function in your source file: |
| 48 | + |
| 49 | +```php |
| 50 | +<?php |
| 51 | + |
| 52 | +require 'vendor/autoload.php'; |
| 53 | + |
| 54 | +// TransformersPHP |
| 55 | +use function Codewithkyrian\Transformers\Pipelines\pipeline; |
| 56 | + |
| 57 | +// Redis client and query engine classes. |
| 58 | +use Predis\Client; |
| 59 | +use Predis\Command\Argument\Search\CreateArguments; |
| 60 | +use Predis\Command\Argument\Search\SearchArguments; |
| 61 | +use Predis\Command\Argument\Search\SchemaFields\TextField; |
| 62 | +use Predis\Command\Argument\Search\SchemaFields\TagField; |
| 63 | +use Predis\Command\Argument\Search\SchemaFields\VectorField; |
| 64 | +``` |
| 65 | + |
| 66 | +## Create a tokenizer instance |
| 67 | + |
| 68 | +The code below shows how to use the |
| 69 | +[`all-MiniLM-L6-v2`](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) |
| 70 | +tokenizer to generate the embeddings. The vectors that represent the |
| 71 | +embeddings have 384 dimensions, regardless of the length of the input |
| 72 | +text. Here, the `pipeline()` call creates the `$extractor` function that |
| 73 | +generates embeddings from text: |
| 74 | + |
| 75 | +```php |
| 76 | +$extractor = pipeline('embeddings', 'Xenova/all-MiniLM-L6-v2'); |
| 77 | +``` |
| 78 | + |
| 79 | +## Create the index |
| 80 | + |
| 81 | +Connect to Redis and delete any index previously created with the |
| 82 | +name `vector_idx`. (The |
| 83 | +[`ftdropindex()`]({{< relref "/commands/ft.dropindex" >}}) |
| 84 | +call throws an exception if the index doesn't already exist, which is |
| 85 | +why you need the `try...catch` block.) |
| 86 | + |
| 87 | +```php |
| 88 | + $client = new Predis\Client([ |
| 89 | + 'host' => 'localhost', |
| 90 | + 'port' => 6379, |
| 91 | +]); |
| 92 | + |
| 93 | +try { |
| 94 | + $client->ftdropindex("vector_idx"); |
| 95 | +} catch (Exception $e){} |
| 96 | +``` |
| 97 | + |
| 98 | +Next, create the index. |
| 99 | +The schema in the example below includes three fields: the text content to index, a |
| 100 | +[tag]({{< relref "/develop/interact/search-and-query/advanced-concepts/tags" >}}) |
| 101 | +field to represent the "genre" of the text, and the embedding vector generated from |
| 102 | +the original text content. The `embedding` field specifies |
| 103 | +[HNSW]({{< relref "/develop/interact/search-and-query/advanced-concepts/vectors#hnsw-index" >}}) |
| 104 | +indexing, the |
| 105 | +[L2]({{< relref "/develop/interact/search-and-query/advanced-concepts/vectors#distance-metrics" >}}) |
| 106 | +vector distance metric, `Float32` values to represent the vector's components, |
| 107 | +and 384 dimensions, as required by the `all-MiniLM-L6-v2` embedding model. |
| 108 | + |
| 109 | +The `CreateArguments` parameter to [`ftcreate()`]({{< relref "/commands/ft.create" >}}) |
| 110 | +specifies hash objects for storage and a prefix `doc:` that identifies the hash objects |
| 111 | +to index. |
| 112 | + |
| 113 | +```php |
| 114 | +$schema = [ |
| 115 | + new TextField("content"), |
| 116 | + new TagField("genre"), |
| 117 | + new VectorField( |
| 118 | + "embedding", |
| 119 | + "HNSW", |
| 120 | + [ |
| 121 | + "TYPE", "FLOAT32", |
| 122 | + "DIM", 384, |
| 123 | + "DISTANCE_METRIC", "L2" |
| 124 | + ] |
| 125 | + ) |
| 126 | +]; |
| 127 | + |
| 128 | +$client->ftcreate("vector_idx", $schema, |
| 129 | + (new CreateArguments()) |
| 130 | + ->on('HASH') |
| 131 | + ->prefix(["doc:"]) |
| 132 | +); |
| 133 | +``` |
| 134 | + |
| 135 | +## Add data |
| 136 | + |
| 137 | +You can now supply the data objects, which will be indexed automatically |
| 138 | +when you add them with [`hmset()`]({{< relref "/commands/hset" >}}), as long as |
| 139 | +you use the `doc:` prefix specified in the index definition. |
| 140 | + |
| 141 | +Use the `$extractor()` function as shown below to create the embedding that |
| 142 | +represents the `content` field. Note that `$extractor()` can generate multiple |
| 143 | +embeddings from multiple strings parameters at once, so it returns an array of |
| 144 | +embedding vectors. Here, there is only one embedding in the returned array. |
| 145 | +The `normalize:` and `pooling:` named parameters relate to details |
| 146 | +of the embedding model (see the |
| 147 | +[`all-MiniLM-L6-v2`](https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2) |
| 148 | +page for more information). |
| 149 | + |
| 150 | +To add an embedding as a field of a hash object, you must encode the |
| 151 | +vector array as a binary string. The built-in |
| 152 | +[`pack()`](https://www.php.net/manual/en/function.pack.php) function is a convenient |
| 153 | +way to do this in PHP, using the `g*` format specifier to denote a packed |
| 154 | +array of `float` values. Note that if you are using |
| 155 | +[JSON]({{< relref "/develop/data-types/json" >}}) |
| 156 | +objects to store your documents instead of hashes, then you should store |
| 157 | +the `float` array directly without first converting it to a binary |
| 158 | +string. |
| 159 | + |
| 160 | +```php |
| 161 | +$content = "That is a very happy person"; |
| 162 | +$emb = $extractor($content, normalize: true, pooling: 'mean'); |
| 163 | + |
| 164 | +$client->hmset("doc:0",[ |
| 165 | + "content" => $content, |
| 166 | + "genre" => "persons", |
| 167 | + "embedding" => pack('g*', ...$emb[0]) |
| 168 | +]); |
| 169 | + |
| 170 | +$content = "That is a happy dog"; |
| 171 | +$emb = $extractor($content, normalize: true, pooling: 'mean'); |
| 172 | + |
| 173 | +$client->hmset("doc:1",[ |
| 174 | + "content" => $content, |
| 175 | + "genre" => "pets", |
| 176 | + "embedding" => pack('g*', ...$emb[0]) |
| 177 | +]); |
| 178 | + |
| 179 | +$content = "Today is a sunny day"; |
| 180 | +$emb = $extractor($content, normalize: true, pooling: 'mean'); |
| 181 | + |
| 182 | +$client->hmset("doc:2",[ |
| 183 | + "content" => $content, |
| 184 | + "genre" => "weather", |
| 185 | + "embedding" => pack('g*', ...$emb[0]) |
| 186 | +]); |
| 187 | +``` |
| 188 | + |
| 189 | +## Run a query |
| 190 | + |
| 191 | +After you have created the index and added the data, you are ready to run a query. |
| 192 | +To do this, you must create another embedding vector from your chosen query |
| 193 | +text. Redis calculates the vector distance between the query vector and each |
| 194 | +embedding vector in the index as it runs the query. You can request the results to be |
| 195 | +sorted to rank them in order of ascending distance. |
| 196 | + |
| 197 | +The code below creates the query embedding using the `$extractor()` function, as with |
| 198 | +the indexing, and passes it as a parameter when the query executes (see |
| 199 | +[Vector search]({{< relref "/develop/interact/search-and-query/query/vector-search" >}}) |
| 200 | +for more information about using query parameters with embeddings). |
| 201 | +The query is a |
| 202 | +[K nearest neighbors (KNN)]({{< relref "/develop/interact/search-and-query/advanced-concepts/vectors#knn-vector-search" >}}) |
| 203 | +search that sorts the results in order of vector distance from the query vector. |
| 204 | + |
| 205 | +The results are returned as an array with the number of results in the |
| 206 | +first element. The remaining elements are alternating pairs with the |
| 207 | +key of the returned document (for example, `doc:0`) first, followed by an array containing |
| 208 | +the fields you requested (again as alternating key-value pairs). |
| 209 | + |
| 210 | +```php |
| 211 | +$queryText = "That is a happy person"; |
| 212 | +$queryEmb = $extractor($queryText, normalize: true, pooling: 'mean'); |
| 213 | + |
| 214 | +$result = $client->ftsearch( |
| 215 | + "vector_idx", |
| 216 | + '*=>[KNN 3 @embedding $vec AS vector_distance]', |
| 217 | + new SearchArguments() |
| 218 | + ->addReturn(1, "vector_distance") |
| 219 | + ->dialect("2") |
| 220 | + ->params([ |
| 221 | + "vec", pack('g*', ...$queryEmb[0]) |
| 222 | + ]) |
| 223 | + ->sortBy("vector_distance") |
| 224 | +); |
| 225 | + |
| 226 | +$numResults = $result[0]; |
| 227 | +echo "Number of results: $numResults" . PHP_EOL; |
| 228 | +// >>> Number of results: 3 |
| 229 | + |
| 230 | +for ($i = 1; $i < ($numResults * 2 + 1); $i += 2) { |
| 231 | + $key = $result[$i]; |
| 232 | + echo "Key: $key" . PHP_EOL; |
| 233 | + $fields = $result[$i + 1]; |
| 234 | + echo "Field: {$fields[0]}, Value: {$fields[1]}" . PHP_EOL; |
| 235 | +} |
| 236 | +// >>> Key: doc:0 |
| 237 | +// >>> Field: vector_distance, Value: 3.76152896881 |
| 238 | +// >>> Key: doc:1 |
| 239 | +// >>> Field: vector_distance, Value: 18.6544265747 |
| 240 | +// >>> Key: doc:2 |
| 241 | +// >>> Field: vector_distance, Value: 44.6189727783 |
| 242 | +``` |
| 243 | + |
| 244 | +Assuming you have added the code from the steps above to your source file, |
| 245 | +it is now ready to run, but note that it may take a while to complete when |
| 246 | +you run it for the first time (which happens because the tokenizer must download the |
| 247 | +`all-MiniLM-L6-v2` model data before it can |
| 248 | +generate the embeddings). When you run the code, it outputs the following result text: |
| 249 | + |
| 250 | +``` |
| 251 | +Number of results: 3 |
| 252 | +Key: doc:0 |
| 253 | +Field: vector_distance, Value: 3.76152896881 |
| 254 | +Key: doc:1 |
| 255 | +Field: vector_distance, Value: 18.6544265747 |
| 256 | +Key: doc:2 |
| 257 | +Field: vector_distance, Value: 44.6189727783 |
| 258 | +``` |
| 259 | + |
| 260 | +Note that the results are ordered according to the value of the `distance` |
| 261 | +field, with the lowest distance indicating the greatest similarity to the query. |
| 262 | +As you would expect, the text *"That is a very happy person"* (from the `doc:0` |
| 263 | +document) |
| 264 | +is the result judged to be most similar in meaning to the query text |
| 265 | +*"That is a happy person"*. |
| 266 | + |
| 267 | +## Learn more |
| 268 | + |
| 269 | +See |
| 270 | +[Vector search]({{< relref "/develop/interact/search-and-query/query/vector-search" >}}) |
| 271 | +for more information about the indexing options, distance metrics, and query format |
| 272 | +for vectors. |
0 commit comments