arroy for vectors #2074
arroy for vectors #2074
Conversation
| "embedding": [1.0, 2.0], | ||
| }, | ||
| ], | ||
| }, | ||
| "vectorisedGraph": { |
There was a problem hiding this comment.
We probably want to move this api inside of graph -- which would fail if the graphs aren't vectorised
| selection.expand_entities_by_similarity("node1", 100, (20, 100)) | ||
| contents = [doc.content for doc in selection.get_documents()] | ||
| assert contents == ["node1", "edge1", "node2", "edge2", "node3"] | ||
| assert contents == ["node1", "edge1", "node2"] |
There was a problem hiding this comment.
did these have to be deleted?
| @@ -229,6 +206,7 @@ def test_default_template(): | |||
|
|
|||
| vg = g.vectorise(constant_embedding) | |||
|
|
|||
| node_docs = vg.entities_by_similarity(query="whatever", limit=10).get_documents() | |||
| @@ -27,6 +27,8 @@ use pyo3::{ | |||
| types::{PyFunction, PyList}, | |||
| }; | |||
|
|
|||
| type DynamicVectorisedGraph = VectorisedGraph<DynamicGraph>; | |||
There was a problem hiding this comment.
We want the Graph storage to have a vectorstore inside of it in the same manner that it has an index
|
|
||
| use super::embeddings::EmbeddingFunction; | ||
|
|
||
| const MAX_DISK_ITEMS: usize = 1_000_000; |
There was a problem hiding this comment.
For next PR - can we make these confirgurable?
| &self, | ||
| ctx: &Context<'_>, | ||
| query: String, | ||
| limit: usize, |
There was a problem hiding this comment.
We need to revamp how algorithms work in graphql
| .unwrap_or(Arc::new(None)); | ||
|
|
||
| GraphWithVectors::read_from_folder(folder, embedding, cache, self.create_index) | ||
| .unwrap_or_else(|| VectorCache::in_memory(openai_embedding)); // TODO: review, this is weird... |
There was a problem hiding this comment.
Wanna have another look at this?
| window: Option<Window>, | ||
| ) -> GraphResult<GqlVectorSelection> { | ||
| let vector = ctx.embed_query(query).await?; | ||
| let w = window.into_window_tuple(); |
There was a problem hiding this comment.
Can we stick some spawn blockings around these
|
|
||
| #[ResolvedObjectFields] | ||
| impl GqlVectorSelection { | ||
| async fn nodes(&self) -> Vec<Node> { |
There was a problem hiding this comment.
I think most of these will need a spawn blocking
| window: Option<Window>, | ||
| ) -> GraphResult<GqlVectorSelection> { | ||
| let vector = ctx.embed_query(query).await?; | ||
| let w = window.into_window_tuple(); |
There was a problem hiding this comment.
Note here for us to add spawn blocking when we redo algorithms
What changes were proposed in this pull request?
Why are the changes needed?
The old approach of defining custom dcument search queries in python had two problems:
Does this PR introduce any user-facing change? If yes is this documented?
How was this patch tested?
Are there any further changes required?