Feat(graphql): Add vector support to graphql #9074
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…ch based on vector indexes (#48) - [Float] data type with an @Embedding directive in GraphQL is mapped to "VFLOAT" data type in dgraph - @Embedding directive is allowed only for [Float] data type - Add querySimilar<Type> query that performs similarity search based on HNSW (vector) indexes for types with fields having @Embedding directive - querySimilar<Type> accepts 3 arguments 1. id - unique id for the object to run similarity search on. 2. <predicate> - An enum of embedding predicates defined in the type to base the search on 3. <topK> - number of nearest neighbors to return
…62) Description: If topK argument in querySimilar* queries is a graphql variable dgraph reported a panic as topK was explicitly cast as int. The fix instead uses fmt.Sprintf("%v", topK) where topK is of type interface{}. This works for when topK is passed as an int literal or a graphql variable. Fixes: HYP-328
… searches (#60) Description: If the vector parameter in a similar_to function call is a variable that evaluates to NULL then we report it as a parsing (syntax) error. The fix, instead of reporting a parsing error, returns an empty result. Closes: HYP-302 --------- Co-authored-by: Bill Province <billprovince@gmail.com>
Description: Made the embedding parameeter, by, to be NonNull for `querySimilar<Type>ById `& `querySimilar<Type>ByEmbedding ` queries. This fixes the backend panic. However, the postman client fails to show the dropdown as in the input for the parameter. It shows it as a simple text input box. The client, while editing the query text in the editor pane shows valid enum values as part of the autocomplete feature. Fixes: HYP-336
…69) Description: This PR adds support for specifying @search directive with search options (vector index options) The fix is not backward compatible. Earlier searchArgs in @search directive were expected to be Enum of index types. Arguments are expected to be strings now of the following form. `@search(by: [String!])` where each string element in the array is of the form `<searchArg> := <searchType> [ ( <searchOptions> ) ]` searchOptions are optional. ``` ``` <searchOptions> := <searchOption> , <searchOption>, ... <searchOption> := <optionName><COLON><SPACE><optionValue> ``` ``` For example: `product_vector: [Float!] @hm_embedding @search(by: ["hnsw(metric: euclidian, exponent: 6)"])` Fixes: HYP-313
…_distance field was getting clobbered (#77) Description: ``` querySimilar<Type> queries defined a new derived type "<Type>WithDistance" with a new field hm_distance. However, if <Type> had any lists, hm_distance was getting clobbered by an "<ListFieldName>Aggregate" being added to <Type> The fix essentially does away with the derived type <Type>WithDistance as the resultType for querySimilar<Type> queries. Instead, we add <embeddingFieldName>Distance field for each embedding in the <Type> definition itself. This would make it easy to add support for filters on embeddings. ``` Fixes: HYP-447
…ch-filter (#79) This change adds filter agrument to querySimilarById & querySimilarByEmbedding queries. The filter argument is used as a "post-vector-serach-filter" to filter out nodes returned by the vector search. Fixes: HYP-334
…#82) Description: Computes hm_distance based on the HNSW index metric ('euclidian', 'dotproduct', 'cosine') Euclidian - (v1 - v2) dot (v1 - v2) Dot product - v1 dot v2 Cosine - (v1 dot v2) / ((v1 dot v1) * (v2 dot v2)) Fixes: HYP-477
|
|
|
|
dgraph-bot
added
area/testing
harshil-goel
changed the title
|
@harshil-goel Defining a field in graphql type as the bellow doesn't work: embedding: [Float!] @embedding @search(by: ["hnsw(exponent: 4, metric: euclidian)"]) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Adds support for vector predicate in GraphQL. Introduced new queries like similar_to() in graphql