Method to compute 'TL-tsne' and 'group-TL-tsne' layouts and save it as node attributes of the focal network.
attach_layout(
metanetwork,
g = NULL,
beta = 0.1,
mode = "TL-tsne",
TL_tsne.config = TL_tsne.default,
res = NULL,
group_layout.config = group_layout.default
)
# S3 method for metanetwork
attach_layout(
metanetwork,
g = NULL,
beta = 0.1,
mode = "TL-tsne",
TL_tsne.config = TL_tsne.default,
res = NULL,
group_layout.config = group_layout.default
)object of class 'metanetwork'
character indicating the name of the network for which the 'TL-tsne' layout is computed, default is 'metaweb'
the diffusion parameter of the diffusion kernel, a positive scalar controlling the squeezing of the network, default is 0.1
'TL-tsne' or 'group-TL-tsne', default is 'TL-tsne'.
configuration list for mode 'TL-tsne', default is TL_tsne.default
resolution for the 'group-TL-tsne' layout
configuration list for mode 'group-TL-tsne', default is group_layout.default
an object of class 'metanetwork', with the computed layout added as node attribute of the considered network
NULL
The 'TL-tsne' layout is a diffusion based layout algorithm specifically designed for trophic networks.
In metanetwork, first axis is the trophic level (see compute_TL method) whereas the second axis is computed using a diffusion graph kernel (Kondor & Lafferty 2002)
and tsne dimension reduction algorithm to (see van der Maaten & Hinton (2008) and 'tsne' R package).
Let \(A\) be the adjacency matrix of the considered network and \(D\) its degree diagonal matrix.
The Laplacian matrix of the symmetrised network is defined by:
$$L = D - A - t(A)$$
The diffusion graph kernel is:
$$K = exp(-beta*L)$$
It is a similarity matrix between nodes according to a diffusion process. beta is the diffusion constant,it must be provided by the user.
beta parameter influences the layout by grouping together similar paths (see pyramid vignette).
Each node of the focal network has an attribute layout_beta_VALUE.
If this function is run several times for a given beta value, repetitions of the layout algorithm will be stored as node attributes.
The 'group-TL-tsne' layout is a variation of 'TL-tsne layout. For a focal network, it mixes 'TL-tsne' layout at the desired aggregated level
with the layout_with_graphopt function from igraph. It clusters nodes belonging to the same group.
'group-TL-tsne' layout is recommended for large networks since you only need to compute 'TL-tsne' at the aggregated network
that is much smaller than the focal network. group_layout.config allows controlling the overall size of the groups.
Kondor, R. I., & Lafferty, J. (2002, July). Diffusion kernels on graphs and other discrete structures. In Proceedings of the 19th international conference on machine learning (Vol. 2002, pp. 315-322). Van der Maaten, L., & Hinton, G. (2008). Visualizing data using t-SNE. Journal of machine learning research, 9(11).
library(metanetwork)
library(igraph)
#>
#> Attaching package: ‘igraph’
#> The following objects are masked from ‘package:stats’:
#>
#> decompose, spectrum
#> The following object is masked from ‘package:base’:
#>
#> union
# on angola dataset (metaweb)
data("meta_angola")
meta_angola = attach_layout(meta_angola,beta = 0.05)
#> attaching TL-tsne layout for metaweb_Species
#>
#> beta = 0.05
#> beta = 0.05
#> Epoch: Iteration #100 error is: 907.410775020654
#> Epoch: Iteration #200 error is: 217.110128493546
#> Epoch: Iteration #300 error is: 217.236109920637
V(meta_angola$metaweb)$layout_beta0.05
#> [1] -22.9414802 -6.0994143 4.9594423 -4.1281724 -14.2013399 15.3619913
#> [7] 30.0598299 7.2834431 20.9454738 23.0397159 11.3357604 26.4009774
#> [13] -8.3537814 -18.0145967 -26.3438149 -1.8507144 -20.3205830 2.5556471
#> [19] 0.3954457 17.8918993 -10.6969826 -31.0165226 9.3785720 13.3728007
#> [25] -15.9951749 -12.5295511 -10.1007625 19.6118919