Skip to content

Latest commit

 

History

History
66 lines (46 loc) · 1.89 KB

README.md

File metadata and controls

66 lines (46 loc) · 1.89 KB

python NGT

日本語

Install

Python binding with pybind11 (ngtpy) is installed as follows.

pip3 install ngt

You can install the python bindings from source code. You MUST install the NGT library according to the README before installing the python bindings as follows.

cd NGT_ROOT/python
pip3 install .

Please note that the search speed of the ngtpy packages from PyPI is slower than that of the ngtpy that is built on your computer so that the package can be run on older CPUs.

Building with setup.py directly or without build isolation requires pybind11 2.12.0 or newer. This version prevents a deadlock when multiple threads first convert a search query to a NumPy array.

Tests

After installing the Python binding, run the self-contained threaded-search regression test from the repository root:

python3 -m unittest discover -s python/tests -p 'test_threaded_search.py' -v
python3 -m unittest discover -s python/tests -p 'test_ngtpy.py' -v

CI runs these self-contained tests explicitly. Other legacy scripts in python/tests/ may require the ctypes binding or prebuilt QBG indexes and are not included.

Documents

ngtpy (pybind11) reference

Simple samples

ngtpy (pybind11)

ngtpy(pybind11) can reduce the processing times than ngt(ctypes). It is more effective especially for the short search time.

  import ngtpy
  import random


  dim = 10
  nb = 100
  vectors = [[random.random() for _ in range(dim)] for _ in range(nb)]
  query = vectors[0]
  
  ngtpy.create(b"tmp", dim)
  index = ngtpy.Index(b"tmp")
  index.batch_insert(vectors)
  index.save()

  results = index.search(query, 3)
  for i, (id, distance) in enumerate(results) :
      print(str(i) + ": " + str(id) + ", " + str(distance))
      object = index.get_object(id)
      print(object)

See also sample.py.