1 Introduction

The BiocNeighbors package provides several algorithms for approximate neighbor searches:

  • The Annoy (Approximate Nearest Neighbors Oh Yeah) method uses C++ code from the RcppAnnoy package. It works by building a tree where a random hyperplane partitions a group of points into two child groups at each internal node. This is repeated to construct a forest of trees where the number of trees determines the accuracy of the search. Given a query data point, we identify all points in the same leaf node for each tree. We then take the union of leaf node sets across trees and search them exactly for the nearest neighbors.
  • The HNSW (Hierarchical Navigable Small Worlds) method uses C++ code from the RcppHNSW package. It works by building a series of nagivable small world graphs containing links between points across the entire data set. The algorithm walks through the graphs where each step is chosen to move closer to a given query point. Different graphs contain links of different lengths, yielding a hierarchy where earlier steps are large and later steps are small. The accuracy of the search is determined by the connectivity of the graphs and the size of the intermediate list of potential neighbors.

These methods complement the exact algorithms described previously. Again, it is straightforward to switch from one algorithm to another by simply changing the BNPARAM argument in findKNN and queryKNN.

2 Identifying nearest neighbors

We perform the k-nearest neighbors search with the Annoy algorithm by specifying BNPARAM=AnnoyParam().

nobs <- 10000
ndim <- 20
data <- matrix(runif(nobs*ndim), ncol=ndim)

fout <- findKNN(data, k=10, BNPARAM=AnnoyParam())
head(fout$index)
##      [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
## [1,] 3613 9813 4083 6358 3983 3841 9690  990 3672  9237
## [2,] 2730  898 5300 7456 1705 9050 8029 3410 1236  5852
## [3,] 8347 3664 3221 5832 9552 6442 4651 3990 4528  4338
## [4,] 8960 3145 8699 6751 5387 5205  127 2384 9399  2483
## [5,] 2152 6140 3869 7777 4876 8563 3927 5677 1768  4050
## [6,] 6756 1306 4789  257 2477 3204 8845 7856 9325  2529
head(fout$distance)
##           [,1]      [,2]      [,3]     [,4]      [,5]      [,6]      [,7]
## [1,] 1.0781144 1.1113344 1.1145079 1.129354 1.1311136 1.1343979 1.1349317
## [2,] 0.9071338 0.9250305 0.9465445 1.024742 1.0320660 1.0788873 1.0926216
## [3,] 1.0713972 1.0809000 1.0839305 1.086237 1.1044838 1.1107026 1.1273242
## [4,] 0.7429730 0.9661263 1.0163893 1.019275 1.0233577 1.0260446 1.0373743
## [5,] 0.8687162 0.9705194 1.0535698 1.066771 1.0937678 1.1508070 1.1616614
## [6,] 0.8059543 0.8987697 0.9082887 0.910625 0.9567227 0.9669381 0.9923332
##          [,8]     [,9]    [,10]
## [1,] 1.138030 1.142646 1.144375
## [2,] 1.102200 1.104858 1.105894
## [3,] 1.144813 1.163753 1.164322
## [4,] 1.043010 1.056140 1.065495
## [5,] 1.165069 1.184794 1.187664
## [6,] 0.992562 1.000768 1.005777

We can also identify the k-nearest neighbors in one dataset based on query points in another dataset.

nquery <- 1000
ndim <- 20
query <- matrix(runif(nquery*ndim), ncol=ndim)

qout <- queryKNN(data, query, k=5, BNPARAM=AnnoyParam())
head(qout$index)
##      [,1] [,2] [,3] [,4] [,5]
## [1,]  331 4323 4123 4205 3873
## [2,] 8202 5315 8944 3853 7496
## [3,] 9098 3327 8742 2371 6678
## [4,] 4017 6387 2827 8607 3822
## [5,] 3041 7477 9651 8744 1238
## [6,] 3275 8183 3086 3916 5454
head(qout$distance)
##           [,1]      [,2]      [,3]      [,4]      [,5]
## [1,] 0.9419816 1.0356504 1.0510969 1.0612359 1.0825620
## [2,] 0.9605968 0.9929906 1.0112295 1.0572425 1.0747641
## [3,] 0.8392634 0.9016435 0.9875906 1.0381438 1.0406716
## [4,] 0.7842190 0.7871433 0.8616644 0.9260836 0.9327102
## [5,] 0.8686712 0.9164358 0.9401396 0.9404123 0.9861693
## [6,] 0.9874870 1.0020382 1.0524002 1.0581913 1.0642735

It is similarly easy to use the HNSW algorithm by setting BNPARAM=HnswParam().

3 Further options

Most of the options described for the exact methods are also applicable here. For example:

  • subset to identify neighbors for a subset of points.
  • get.distance to avoid retrieving distances when unnecessary.
  • BPPARAM to parallelize the calculations across multiple workers.
  • BNINDEX to build the forest once for a given data set and re-use it across calls.

The use of a pre-built BNINDEX is illustrated below:

pre <- buildIndex(data, BNPARAM=AnnoyParam())
out1 <- findKNN(BNINDEX=pre, k=5)
out2 <- queryKNN(BNINDEX=pre, query=query, k=2)

Both Annoy and HNSW perform searches based on the Euclidean distance by default. Searching by Manhattan distance is done by simply setting distance="Manhattan" in AnnoyParam() or HnswParam().

Users are referred to the documentation of each function for specific details on the available arguments.

4 Saving the index files

Both Annoy and HNSW generate indexing structures - a forest of trees and series of graphs, respectively - that are saved to file when calling buildIndex(). By default, this file is located in tempdir()1 On HPC file systems, you can change TEMPDIR to a location that is more amenable to concurrent access. and will be removed when the session finishes.

AnnoyIndex_path(pre)
## [1] "/tmp/Rtmpoen9na/file66a048a0e4a6.idx"

If the index is to persist across sessions, the path of the index file can be directly specified in buildIndex. This can be used to construct an index object directly using the relevant constructors, e.g., AnnoyIndex(), HnswIndex(). However, it becomes the responsibility of the user to clean up any temporary indexing files after calculations are complete.

5 Session information

sessionInfo()
## R version 4.0.3 (2020-10-10)
## Platform: x86_64-pc-linux-gnu (64-bit)
## Running under: Ubuntu 18.04.5 LTS
## 
## Matrix products: default
## BLAS:   /home/biocbuild/bbs-3.12-bioc/R/lib/libRblas.so
## LAPACK: /home/biocbuild/bbs-3.12-bioc/R/lib/libRlapack.so
## 
## locale:
##  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
##  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=C              
##  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
##  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
##  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
## [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
## [1] BiocNeighbors_1.8.2 knitr_1.30          BiocStyle_2.18.1   
## 
## loaded via a namespace (and not attached):
##  [1] Rcpp_1.0.5          bookdown_0.21       lattice_0.20-41    
##  [4] digest_0.6.27       grid_4.0.3          stats4_4.0.3       
##  [7] magrittr_2.0.1      evaluate_0.14       rlang_0.4.9        
## [10] stringi_1.5.3       S4Vectors_0.28.0    Matrix_1.2-18      
## [13] rmarkdown_2.5       BiocParallel_1.24.1 tools_4.0.3        
## [16] stringr_1.4.0       parallel_4.0.3      xfun_0.19          
## [19] yaml_2.2.1          compiler_4.0.3      BiocGenerics_0.36.0
## [22] BiocManager_1.30.10 htmltools_0.5.0