RCPP Not Compatible
lkoll opened this issue · comments
I am attempting to use the R implementation of Rsomoclu.train with the following parameters on a sparse matrix, dimensions 480k x 18k.
nEpochs<-5
nx<-70
ny<-70
som_train<-Rsomoclu.train(dfm,
nEpoch=nEpochs,
nSomX = nx,
nSomY = ny,
kernelType = 2,
radius0 = 0,
radiusN = 1,
radiusCooling = "linear",
scale0 = 1,
scaleN = .01,
scaleCooling = "linear")
When I run this it fails almost instantly with the following error:
> som_train<-Rsomoclu.train(dfm,
+ nEpoch=nEpochs,
+ nSomX = nx,
+ nSomY .... [TRUNCATED]
terminate called after throwing an instance of 'Rcpp::not_compatible'
what(): Not compatible with requested type: [type=S4; target=double].
Am I setting a parameter incorrectly?
Can you provide a MWE with the dfm?
MWE.txt
Here is the first 1000 rows of my dfm, I ran it with the same parameters I was using on the whole thing and it immediately errored out again. Not sure if you need the below or not but it's the structure of the matrix MWE.
Formal class 'dgTMatrix' [package "Matrix"] with 6 slots
..@ i : int [1:220352] 378 303 532 652 699 920 923 966 977 239 ...
..@ j : int [1:220352] 0 6 6 6 6 6 6 6 6 7 ...
..@ Dim : int [1:2] 1000 17770
..@ Dimnames:List of 2
.. ..$ : NULL
.. ..$ : NULL
..@ x : num [1:220352] 4 5 1 2 3 2 5 5 2 2 ...
..@ factors : list()
@xgdgsc
Just wanted to make sure you received my dataset, please let me know if you find anything out.
library('Rsomoclu')
data("rgbs", package = "Rsomoclu")
dfm = read.csv("~/tmp/MWE.txt",sep = " ")
dfm_m = data.matrix(dfm)
nEpochs<-5
nx<-70
ny<-70
som_train<-Rsomoclu.train(dfm_m,
nEpoch=nEpochs,
nSomX = nx,
nSomY = ny,
kernelType = 0,
radius0 = 0,
radiusN = 1,
radiusCooling = "linear",
scale0 = 1,
scaleN = .01,
scaleCooling = "linear")it' s a problem with not converting your dataframe.
and you have to remove the header:
I don't think that's my issue. When you read in the data use the "Matrix" package readMM function as follows. It is already in sparse matrix form when I run it.
dfm<-readMM("MWE.txt")
> str(dfm)
Formal class 'dgTMatrix' [package "Matrix"] with 6 slots
..@ i : int [1:220352] 378 303 532 652 699 920 923 966 977 239 ...
..@ j : int [1:220352] 0 6 6 6 6 6 6 6 6 7 ...
..@ Dim : int [1:2] 1000 17770
..@ Dimnames:List of 2
.. ..$ : NULL
.. ..$ : NULL
..@ x : num [1:220352] 4 5 1 2 3 2 5 5 2 2 ...
..@ factors : list()
Since matrix is the only type we support here. You may need to convert like:
dfm<-readMM("~/tmp/MWE.txt")
dfm_m = as.matrix(dfm)
...
It seems to take much more memory. So the recommended way of using in this case is the cmd line version.
I was under the impression that sparse matrices were supported because of the kernel type option of "Sparse CPU" in the following documentation (https://cran.r-project.org/web/packages/Rsomoclu/Rsomoclu.pdf). So there is no sparse matrix support for this package?
Edit:
The following sources also seem to indicate that there is...
https://peterwittek.com/training-emergent-self-organizing-maps-with-somoclu.html
https://arxiv.org/abs/1305.1422
The support is mainly in the cli version. We didn' t use R intensively to think about this issue.
It may be a good idea to update your R documentation in that case.
What is cli?