Quickstart

How to use deglib

Using deglib follows the following procedure:

  • Load your feature vector database

    • The following example code generates a random dataset of 10,000 samples, each with 256 dimensions.

  • Building the graph

  • A query is submitted. Feature vectors similar to this query need to be found.

  • Using the graph to find similar feature vectors in the database

The following code shows an example of how this can be implemented:

import numpy as np
import deglib

N_SAMPLES, DIMS = 10_000, 256

# generate dataset
data = np.random.random((N_SAMPLES, DIMS)).astype(np.float32)

# build index
graph = deglib.builder.build_from_data(data)

# generate query
query = np.random.random(DIMS).astype(np.float32)

# search query
indices, distances = graph.search(query, eps=0.1, k=16)

print(indices)    # data[result_indices] will show the 16 closest datapoints to "query"
print(distances)  # numpy array with 16 distances to the results

The output is a NumPy array with shape (1, k) containing k indices from the dataset and a corresponding array with the distance to the query for each result.

More Options

There are far more options to build a graph and to search for results. Look at the documentation for building graphs and the search documentation.