pyflann 使用方法小结

安装

• pip install pyflann
• 如果是python3，且安装失败。
  • sudo 2to3 -w [pyflann directory in dist-packages]

使用

  建立Flann 并测试使用

###python3
import pyflann
import pickle
import numpy as np

# 生成数据
train_n = 100000
test_n = 500
feature_number = 500

train_data = np.random.rand(train_n, feature_number)
test_data = np.random.rand(test_n, feature_number)

# 设置distance type
pyflann.set_distance_type("euclidean")

# 建立FLANN类
flann = pyflann.FLANN()

# 建立索引并分类
branching = 10
params = flann.build_index(train_data, algorithm='kmeans',target_precision=0.9, branching = branching , log_level='info')

# 测试
top_k_results = 20
# sims: 500 * 20 每行代表一个test data，第k列是对应test data的TOP-k 结果
# dists: 500 * 20 每行代表一个test data， 第k列是test data 和 top-k result 的距离
sims, dists = flann.nn_index(test_data, top_k_results, checks = params['checks'])

# 保存
pickle.dump(params,open('params.pk','wb'))
flann.save_index(b'flann_index')
# Or 
# flann_filename = 'flann_index'
# flann.save_index(bytes(flann_filename, encoding='utf8'))

  读取已经建立好的FLANN， 并使用

import pickle
import pyflann
import numpy as np

# 读取上次的train_data
train_data = np.load(...)

# 从文件读取 flann
pyflann.set_distance_type("euclidean")
flann = pyflann.FLANN()

params = pickle.load(open('params.pk','rb'))
flann.load_index(b'flann_index', train_data)
# Or 
# flann_filename = 'flann_index'
# flann.load_index(bytes(flann_filename, encoding='utf8'), train_data)


# 使用
newData = np.random.rand(200,500)
topk = 20

sims, dists = flann.nn_index(newData, topk, checks=params['checks'])

 

常见错误

1. ImportError: Cannot load dynamic library. Did you compile FLANN?

出错原因：pip安装的pyflann缺少lib文件夹

解决方法：找到pyflann安装的位置（如下），在【2】中下载lib文件夹放到该目录即可

python3 -m pip show pyflann
# or
python3 -m pip show flann

 

 

REFERENCE:

【1】https://www.cs.ubc.ca/research/flann/uploads/FLANN/flann_manual-1.8.4.pdf

【2】https://github.com/primetang/pyflann