From 7e2ab6aaac94dfaa39f85e4cbe18118b9b48a0f4 Mon Sep 17 00:00:00 2001 From: vankichi Date: Fri, 20 May 2022 18:04:53 +0900 Subject: [PATCH] :pencil: fix grammar Signed-off-by: vankichi --- docs/user-guides/search-config.md | 65 ++++++++++++++++--------------- 1 file changed, 33 insertions(+), 32 deletions(-) diff --git a/docs/user-guides/search-config.md b/docs/user-guides/search-config.md index d70d5cb489..65db985632 100644 --- a/docs/user-guides/search-config.md +++ b/docs/user-guides/search-config.md @@ -4,37 +4,37 @@ -Vald provides the two types of search service. +Vald provides the two types of search services. 1. Search - `Search` is just the `ANN(Approximate Nearest Neighbor)` search with query vector. - It is the fast search even the vector consists large dimension. - The search duration is fast, but less accurate than `LinearSearch`. - Search algorithm depends on each core algorithm. + It is a fast search even though the vector consists large dimension. + The search duration is quick but less accurate than `LinearSearch`. + The search algorithm depends on each core algorithm. 1. SearchById - - `SearchById` is just the `ANN(Approximate Nearest Neighbor)` search with vector's id. - The id should be already indexed before search process. - The sarch algorithm is the same as `Search`. + - `SearchById` is just the `ANN(Approximate Nearest Neighbor)` search with the vector's id. + The id should already exist in the NGT indexes before the search process. + The search algorithm is the same as `Search`. 1. LinearSearch - - `LinearSearch` is the basic search algorithm with query vector. + - `LinearSearch` is the primary search algorithm with a query vector. It searches all indexed vectors and calculates the distance between the query. Its accuracy is exact, but the search time requires more than `Search` (ANN search) and increases the amount of indexed vector. 1. LinearSearchById - - `LinearSearchById` is the basic search algorithm with vector's id. - The id should be already indexed before search process. - The sarch algorithm is the same as `LinearSearch`. + - `LinearSearchById` is the primary search algorithm with the vector's id. + The id should already exist in the NGT indexes before the search process. + The search algorithm is the same as `LinearSearch`.
Linear Search service is available from Vald v1.4 or later.
-For more details, please refer to [Search API document](../api/search.md) +For more details, please refer to [the Search API document](../api/search.md). ## Configuration @@ -67,50 +67,51 @@ message Config { ### request_id -`request_id` is unique request ID. +`request_id` is a unique request ID. It is **NOT** indexed vector's id. -It can be used for e.g., error handling process. +Users can use it for, e.g., the error handling process. ### num -`num` is the maximum number of search result which you'd like to get. -`num` should be positive integer. +`num` is the maximum number of search results you'd like to get. +`num` should be a positive integer. ### radius `radius`, the specific parameter for NGT, specifies the search range centered on the query vector in terms of the radius of a circle. The number of search target vectors increases along with the radius is large. -There is the trade-off between accuracy and search speed. -In many cases, it is hard to set it due to depending on dataset. +There is a trade-off between accuracy and search speed. +In many cases, it is hard to set it depending on dataset. -The default value is infinite circle. -When setting negative number as `radius`, `NGT` applies the radius as infinite circle. +The default value is the infinite circle. +When setting a negative number as `radius`, `NGT` applies the radius as an infinite circle.
-NGT will self-update radius during search process. +NGT will self-update the radius during the search process.
### epsilon -`epsilon`, the specific parameter for NGT, specifies the magnification coefficient (epsilon) of the search range. +`epsilon`, the specific parameter for NGT, specifies the search range's magnification coefficient (epsilon). NGT will use `(1+epsilon)*radius` as the search range. -The number of search target vectors increases along with the epsilon is large. +The number of search target vectors increases along with the epsilon being large. -The default value (recommend value) is `0.1`. -While it is desirable to adjust this value within the range of 0 - 0.3, a negative value (over than `-1`) may also be specified. + +The default value (recommend value) is 0.1. +While it is desirable to adjust this value within 0 - 0.3, it can also set a negative value (over than -1). ### ingress_filters -`ingress_filters` is required when using ingress filter component. -It requires ingress filter component's hostname and port. +`ingress_filters` is required when using the ingress filter component. +It requires the ingress filter component's hostname and port. ### egress_filters -`egress_filters` is required when using egress filter component. -It requires egress filter component's hostname and port. +`egress_filters` is required when using the egress filter component. +It requires the egress filter component's hostname and port. ### min_num -`min_num` is the minimum number of search result which you'd like to get at least. -It helps you to avoid getting the timeout error when the search process requires more time. -`min_num` should be positive integer and smaller than `num`. +`min_num` is the minimum number of search result you'd like to get at least. +It helps you avoid getting the timeout error when the search process requires more time. +`min_num` should be a positive integer and smaller than `num`.