Switch from leiden to leidenbase

[alanocallaghan]

The Leiden implementation provided by leiden is absurdly slow, way more than I'd expect from the overhead of calling reticulate. This switches for the (mostly) equivalent leidenbase. Tests pass locally for me, see also the demo below comparing 3 Leiden implementations.

See #6754

library("leidenbase")
library("leidenAlg")
library("leiden")
library("igraph")
library("microbenchmark")

fpath <- system.file('testdata', 'igraph_n1500_edgelist.txt.gz', package='leidenbase')
zfp <- gzfile(fpath)
igraph <- read_graph(file = zfp, format='edgelist')

microbenchmark(
    leidenAlg::leiden.community(igraph),
    leidenbase::leiden_find_partition(igraph),
    leiden::leiden(igraph),
    times = 10
)
# Unit: milliseconds
#                           expr         min          lq        mean      median
#       leiden.community(igraph)    52.98505    55.16530    63.16430    55.95267
#  leiden_find_partition(igraph)    29.22238    30.54163    39.15174    37.57693
#                 leiden(igraph) 17140.52418 20671.12398 21172.11517 21128.94100
#           uq         max neval cld
#     75.38524    80.10432    10  a 
#     40.30238    56.91094    10  a 
#  22268.67866 24597.89747    10   b


r1 <- leiden_find_partition(igraph,
    partition_type = "RBConfigurationVertexPartition",
    seed = 1234,
    resolution_parameter = 1
    
)
r2 <- leiden(igraph,
    partition_type = "RBConfigurationVertexPartition",
    resolution_parameter = 1, 
    seed = 1234
)
table(r1$membership, r2)
#       1   2   3   4   5   6   7   8   9
#   1 295   0   0   0   0   0   0   0   0
#   2   0 209   0   0   0   0   0   0   0
#   3   0   0 201   0   0   0   0   0   0
#   4   0   0   0 191   0   0   0   0   0
#   5   0   0   0   0 175   0   0   0   0
#   6   0   0   0   0   0 170   0   0   0
#   7   0   0   0   0   0   0  93   0   0
#   8   0   0   0   0   0   0   0  85   0
#   9   0   0   0   0   0   0   0   0  81

[szhorvat]

If you are looking to reduce overhead to a minimum, verify whether you are relying on any feature not already available in igraph::cluster_leiden(), and if not, just use that function. The various "partition type" settings are reproducible by setting the vertex_weight values appropriately. See the general form of the objective function in the documentation, where $n$ represents the vertex weigths, and compare with the objective functions of the various "partition types". For example, setting $n_i = k_i / \sqrt{2m}$ where $k_i$ is the vertex degree/strength and $m$ is the edge count / total edge weight yields the modularity objective function (i.e. RBConfigurationVertexPartition).

[alanocallaghan]

It's a bit out of my wheelhouse to go through and reimplement the different objective functions to be honest, I just thought this was a useful contribution that doesn't increase the dependency count.

[alanocallaghan]

Pinging @saketkc @samuel-marsh as people who seem to be merging code:

Can you have a look at this? Feel free to say you're not interested and close it, or that you want more info/more tests showing equivalence. Either is fine

[samuel-marsh]

I will second @alanocallaghan ping that would be great to have this included!

I’m not member of dev team so my actions are limited here but adding @Gesmira from Seurat team.

Best,
Sam