|
| 1 | +--- |
| 2 | +categories: |
| 3 | +- docs |
| 4 | +- develop |
| 5 | +- stack |
| 6 | +- oss |
| 7 | +- rs |
| 8 | +- rc |
| 9 | +- oss |
| 10 | +- kubernetes |
| 11 | +- clients |
| 12 | +description: Index and query embeddings with Redis vector sets |
| 13 | +linkTitle: Vector set embeddings |
| 14 | +title: Vector set embeddings |
| 15 | +weight: 40 |
| 16 | +bannerText: Vector set is a new data type that is currently in preview and may be subject to change. |
| 17 | +bannerChildren: true |
| 18 | +--- |
| 19 | + |
| 20 | +A Redis [vector set]({{< relref "/develop/data-types/vector-sets" >}}) lets |
| 21 | +you store a set of unique keys, each with its own associated vector. |
| 22 | +You can then retrieve keys from the set according to the similarity between |
| 23 | +their stored vectors and a query vector that you specify. |
| 24 | + |
| 25 | +You can use vector sets to store any type of numeric vector but they are |
| 26 | +particularly optimized to work with text embedding vectors (see |
| 27 | +[Redis for AI]({{< relref "/develop/ai" >}}) to learn more about text |
| 28 | +embeddings). The example below shows how to use the |
| 29 | +[`Microsoft.ML`](https://dotnet.microsoft.com/en-us/apps/ai/ml-dotnet) |
| 30 | +library to generate vector embeddings and then |
| 31 | +store and retrieve them using a vector set with `StackExchange.Redis`. |
| 32 | + |
| 33 | +## Initialize |
| 34 | + |
| 35 | +Start by installing `StackExchange.Redis` with the following |
| 36 | +command (version 2.9.17 or later is required for vector sets): |
| 37 | + |
| 38 | +```bash |
| 39 | +dotnet add package StackExchange.Redis --version 2.9.17 |
| 40 | +``` |
| 41 | + |
| 42 | +Also, install `Microsoft.ML`: |
| 43 | + |
| 44 | +```bash |
| 45 | +dotnet add package Microsoft.ML |
| 46 | +``` |
| 47 | + |
| 48 | +In a new C# file, import the required classes. Note that the `#pragma` |
| 49 | +directive suppresses warnings about the experimental status of the vector set API: |
| 50 | + |
| 51 | +{{< clients-example set="home_vecsets" step="import" lang_filter="C#" >}} |
| 52 | +{{< /clients-example >}} |
| 53 | + |
| 54 | +## Access the model |
| 55 | + |
| 56 | +Use the `GetPredictionEngine()` helper function declared in the example below to load the model that creates the embeddings: |
| 57 | + |
| 58 | +{{< clients-example set="home_vecsets" step="model" lang_filter="C#" >}} |
| 59 | +{{< /clients-example >}} |
| 60 | + |
| 61 | +The `GetPredictionEngine()` function uses two classes, `TextData` and `TransformedTextData`, |
| 62 | +to specify the `PredictionEngine` model. These have a very simple definition |
| 63 | +and are required because the model expects the input and output to be |
| 64 | +passed in named object fields: |
| 65 | + |
| 66 | +{{< clients-example set="home_vecsets" step="data_classes" lang_filter="C#" >}} |
| 67 | +{{< /clients-example >}} |
| 68 | + |
| 69 | +Note that you must declare these classes at the end of the source file |
| 70 | +if you are using a console app without a main class. |
| 71 | + |
| 72 | +The `GetEmbedding()` function declared below can then use this model to |
| 73 | +generate an embedding from a section of text and return it as a `float[]` array, |
| 74 | +which is the format required by the vector set API: |
| 75 | + |
| 76 | +{{< clients-example set="home_vecsets" step="get_embedding" lang_filter="C#" >}} |
| 77 | +{{< /clients-example >}} |
| 78 | + |
| 79 | +## Create the data |
| 80 | + |
| 81 | +The example data is contained a `Dictionary` object with some brief |
| 82 | +descriptions of famous people: |
| 83 | + |
| 84 | +{{< clients-example set="home_vecsets" step="data" lang_filter="C#" >}} |
| 85 | +{{< /clients-example >}} |
| 86 | + |
| 87 | +## Add the data to a vector set |
| 88 | + |
| 89 | +The next step is to connect to Redis and add the data to a new vector set. |
| 90 | + |
| 91 | +The code below iterates through `peopleData` and adds corresponding |
| 92 | +elements to a vector set called `famousPeople`. |
| 93 | + |
| 94 | +Use the `GetEmbedding()` function declared above to generate the |
| 95 | +embedding as a `byte` array that you can pass to the |
| 96 | +[`VectorSetAdd()`]({{< relref "/commands/vadd" >}}) command to set the embedding. |
| 97 | + |
| 98 | +The call to `VectorSetAdd()` also adds the `born` and `died` values from the |
| 99 | +original dictionary as attribute data. You can access this during a query |
| 100 | +or by using the [`VectorSetGetAttributesJson()`]({{< relref "/commands/vgetattr" >}}) method. |
| 101 | + |
| 102 | +{{< clients-example set="home_vecsets" step="add_data" lang_filter="C#" >}} |
| 103 | +{{< /clients-example >}} |
| 104 | + |
| 105 | +## Query the vector set |
| 106 | + |
| 107 | +You can now query the data in the set. The basic approach is to use the |
| 108 | +`GetEmbedding()` function to generate another embedding vector for the query text. |
| 109 | +(This is the same method used to add the elements to the set.) Then, pass |
| 110 | +the query vector to [`VectorSetSimilaritySearch()`]({{< relref "/commands/vsim" >}}) to |
| 111 | +return elements of the set, ranked in order of similarity to the query. |
| 112 | + |
| 113 | +Start with a simple query for "actors": |
| 114 | + |
| 115 | +{{< clients-example set="home_vecsets" step="basic_query" lang_filter="C#" >}} |
| 116 | +{{< /clients-example >}} |
| 117 | + |
| 118 | +This returns the following list of elements (formatted slightly for clarity): |
| 119 | + |
| 120 | +``` |
| 121 | +'actors': ['Masako Natsume', 'Chaim Topol', 'Linus Pauling', |
| 122 | +'Marie Fredriksson', 'Maryam Mirzakhani', 'Marie Curie', |
| 123 | +'Freddie Mercury', 'Paul Erdos'] |
| 124 | +``` |
| 125 | + |
| 126 | +The first two people in the list are the two actors, as expected, but none of the |
| 127 | +people from Linus Pauling onward was especially well-known for acting (and there certainly |
| 128 | +isn't any information about that in the short description text). |
| 129 | +As it stands, the search attempts to rank all the elements in the set, based |
| 130 | +on the information contained in the embedding model. |
| 131 | +You can use the `Count` property of `VectorSetSimilaritySearchRequest` to limit the |
| 132 | +list of elements to just the most relevant few items: |
| 133 | + |
| 134 | +{{< clients-example set="home_vecsets" step="limited_query" lang_filter="C#" >}} |
| 135 | +{{< /clients-example >}} |
| 136 | + |
| 137 | +The reason for using text embeddings rather than simple text search |
| 138 | +is that the embeddings represent semantic information. This allows a query |
| 139 | +to find elements with a similar meaning even if the text is |
| 140 | +different. For example, the word "entertainer" doesn't appear in any of the |
| 141 | +descriptions but if you use it as a query, the actors and musicians are ranked |
| 142 | +highest in the results list: |
| 143 | + |
| 144 | +{{< clients-example set="home_vecsets" step="entertainer_query" lang_filter="C#" >}} |
| 145 | +{{< /clients-example >}} |
| 146 | + |
| 147 | +Similarly, if you use "science" as a query, you get the following results: |
| 148 | + |
| 149 | +``` |
| 150 | +'science': ['Marie Curie', 'Linus Pauling', 'Maryam Mirzakhani', |
| 151 | +'Paul Erdos', 'Marie Fredriksson', 'Freddie Mercury', 'Masako Natsume', |
| 152 | +'Chaim Topol'] |
| 153 | +``` |
| 154 | + |
| 155 | +The scientists are ranked highest but they are then followed by the |
| 156 | +mathematicians. This seems reasonable given the connection between mathematics |
| 157 | +and science. |
| 158 | + |
| 159 | +You can also use |
| 160 | +[filter expressions]({{< relref "/develop/data-types/vector-sets/filtered-search" >}}) |
| 161 | +with `VectorSetSimilaritySearch()` to restrict the search further. For example, |
| 162 | +repeat the "science" query, but this time limit the results to people |
| 163 | +who died before the year 2000: |
| 164 | + |
| 165 | +{{< clients-example set="home_vecsets" step="filtered_query" lang_filter="C#" >}} |
| 166 | +{{< /clients-example >}} |
| 167 | + |
| 168 | +Note that the boolean filter expression is applied to items in the list |
| 169 | +before the vector distance calculation is performed. Items that don't |
| 170 | +pass the filter test are removed from the results completely, rather |
| 171 | +than just reduced in rank. This can help to improve the performance of the |
| 172 | +search because there is no need to calculate the vector distance for |
| 173 | +elements that have already been filtered out of the search. |
| 174 | + |
| 175 | +## More information |
| 176 | + |
| 177 | +See the [vector sets]({{< relref "/develop/data-types/vector-sets" >}}) |
| 178 | +docs for more information and code examples. See the |
| 179 | +[Redis for AI]({{< relref "/develop/ai" >}}) section for more details |
| 180 | +about text embeddings and other AI techniques you can use with Redis. |
| 181 | + |
| 182 | +You may also be interested in |
| 183 | +[vector search]({{< relref "/develop/clients/dotnet/vecsearch" >}}). |
| 184 | +This is a feature of the |
| 185 | +[Redis query engine]({{< relref "/develop/ai/search-and-query" >}}) |
| 186 | +that lets you retrieve |
| 187 | +[JSON]({{< relref "/develop/data-types/json" >}}) and |
| 188 | +[hash]({{< relref "/develop/data-types/hashes" >}}) documents based on |
| 189 | +vector data stored in their fields. |
0 commit comments