DeepWalk node embeddings (alias of Deepw)

deepwk and deepw are the SAME method: truncated uniform random walks fed to skip-gram, Perozzi, Al-Rfou & Skiena (2014). Two module names for one method is exactly the duplicate this campaign is trying not to create, so this function forwards to Deepw rather than repeating it – a second copy would agree with the first at 1e-9 forever and tell nobody anything.

Usage

Deepwk(
  G,
  walk_len = 10,
  dim = 8,
  n_walks = 4,
  window = 3,
  epochs = 1,
  lr = 0.05,
  neg = 2,
  seed = 42
)

Arguments

G: An n x n adjacency matrix.
walk_len: Length of each walk.
dim: Embedding dimension.
n_walks: Walks started from each node.
window: Skip-gram context window.
epochs: Passes over the corpus.
lr: SGD step size.
neg: Negative samples per positive pair.
seed: Seed of the deterministic stream.

Value

Whatever Deepw returns.

References

Perozzi, Al-Rfou & Skiena (2014), DeepWalk, KDD 2014:701-710.

Examples

G <- matrix(c(0, 1, 0, 1, 1, 0, 1, 0, 0, 1, 0, 1, 1, 0, 1, 0), 4, 4, byrow = TRUE)
Deepwk(G, dim = 4, walk_len = 5, n_walks = 2)
#> $estimate
#> [1] 0.003179637
#> 
#> $embedding
#>             [,1]        [,2]        [,3]        [,4]
#> [1,]  0.03001782  0.06103993  0.03431656 -0.04528434
#> [2,] -0.12270361 -0.07729287  0.06482800 -0.12935236
#> [3,]  0.01029798 -0.09519040 -0.06580563  0.05562667
#> [4,]  0.03453990 -0.06140595 -0.10860119 -0.10971831
#> 
#> $walks
#> $walks[[1]]
#> [1] 1 4 1 2 1
#> 
#> $walks[[2]]
#> [1] 2 1 4 1 4
#> 
#> $walks[[3]]
#> [1] 3 2 3 2 1
#> 
#> $walks[[4]]
#> [1] 4 3 4 3 2
#> 
#> $walks[[5]]
#> [1] 1 2 1 2 3
#> 
#> $walks[[6]]
#> [1] 2 3 2 3 4
#> 
#> $walks[[7]]
#> [1] 3 2 1 4 3
#> 
#> $walks[[8]]
#> [1] 4 3 4 3 4
#> 
#> 
#> $n_walks_total
#> [1] 8
#> 
#> $degree
#> [1] 2 2 2 2
#> 
#> $n
#> [1] 4
#> 
#> $dim
#> [1] 4
#> 
#> $method
#> [1] "DeepWalk node embeddings"
#> 
#> attr(,"class")
#> [1] "morie_rich_result" "list"