​
Concept

Laminar datasets can be indexed for semantic search. Every datapoint is embedded and so a semantic similarity search can be performed.

​
Indexing a dataset

​
Requirements

  • For every datapoint you want to index, the data field must be a JSON object.
  • All datapoints that you want to index must have the entry with a certain key.

​
Example

The below datapoints both have the key text in the data field. We can index on this key.

[
    {
        "data": {
            "text": "This is a good poem about flowers",
            "other_field": "some other field"
        }
    },
    {
        "data": {
            "text": "Laminar flow is opposite to turbulent flow",
            "user_tag_key": "some user tag value"
        }
    },
]

If we index on the text field, we can search over the indexed dataset by semantic similarity between the query and the text field. For example, if we search for “turbulence”, we will get the second datapoint as the most similar one.

We could also index on the user_tag_key field, but in that case the first datapoint would not be indexed, because it does not have the user_tag_key field.

​
Indexing

Go to the dataset page and click the “Index” button. Type in the key that you want to index on. Empty key will unindex all datapoints.

Once the dataset is indexed on a key, new datapoints will be indexed on that key automatically, if they have the key in data field.

You can search over indexed dataset by semantic similarity between the query and the indexed value.

​
Searching

Semantic search over datapoint can be performed in two ways:

​
API call example

import { Laminar } from "laminar";

const results = await Laminar.semanticSearch({
    query: "This is a passage about kings and queens",
    datasetId: "01234567-89ab-cdef-0123-456789abcdef",
});

console.log(results);
Adding dataIntroduction