Incubator-mxnet: graph.viz broken - returns Error in symbol$as.json : $ operator is invalid for atomic vectors

[dockstreet]

For bugs or installation issues, please provide the following information.
The more information you provide, the more likely people will be able to help you.

Environment info

Operating System:Windows 7 / RHEL 7

Compiler:g++

Package used (Python/R/Scala/Julia):R

MXNet version:0.9.4

Or if installed from source:

MXNet commit hash (git rev-parse HEAD):

If you are using python package, please provide

Python version and distribution:

If you are using R package, please provide

R sessionInfo():R version 3.3.1 (2016-06-21)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 7 x64 (build 7601) Service Pack 1

locale:
[1] LC_COLLATE=English_United States.1252 LC_CTYPE=English_United States.1252 LC_MONETARY=English_United States.1252
[4] LC_NUMERIC=C LC_TIME=English_United States.1252

attached base packages:
[1] stats graphics grDevices utils datasets methods base

other attached packages:
[1] mxnet_0.9.4 mlbench_2.1-1

loaded via a namespace (and not attached):
[1] htmlwidgets_0.8 magrittr_1.5 rsconnect_0.7 htmltools_0.3.5 visNetwork_1.0.4 tools_3.3.1 Rcpp_0.12.9
[8] codetools_0.2-15 stringi_1.1.2 digest_0.6.12 stringr_1.2.0 jsonlite_1.3

Error Message:

Please paste the full error message, including stack trace.
Error in symbol$as.json : $ operator is invalid for atomic vectors

Minimum reproducible example

I try :
http://mxnet.io/tutorials/r/fiveMinutesNeuralNetwork.html

require(mlbench)

Loading required package: mlbench

require(mxnet)

Loading required package: mxnet

Loading required package: methods

data(Sonar, package="mlbench")

Sonar[,61] = as.numeric(Sonar[,61])-1
train.ind = c(1:50, 100:150)
train.x = data.matrix(Sonar[train.ind, 1:60])
train.y = Sonar[train.ind, 61]
test.x = data.matrix(Sonar[-train.ind, 1:60])
test.y = Sonar[-train.ind, 61]

mx.set.seed(0)
model <- mx.mlp(train.x, train.y, hidden_node=10, out_node=2, out_activation="softmax",
num.round=20, array.batch.size=15, learning.rate=0.07, momentum=0.9,
eval.metric=mx.metric.accuracy)

graph.viz(model$symbol$as.json())

Steps to reproduce

or if you are running standard examples, please provide the commands you have run that lead to the error.

graph.viz(model$symbol$as.json())

What have you tried to solve it?

1. different parameters and saving the variable out, but no luck

[jeremiedb]

Hi, the graph.viz now takes the symbol as argument. You can try graph.viz(model$symbol). This allowed to add the option to display the shapes of the outputs following each operator. However, I noticed DiagrammeR was not loaded by the NAMESPACE, so I'm not sure if you're having the latest version.

[dockstreet]

Wow ! thanks so much ! changing it to graph.viz(model$symbol) worked!

[dockstreet]

one more question - thanks again for the help; I was able to use shiny with visNetworkOutput in 0.9.3 on windows with graph.viz because that was what I had for binaries installed.

I also had 0.9.4 compiled from source on Red Hat, because I couldn't get the binaries otherwise ; and noticed some differences. I do see a new type paramameter for vis in 0.9.4 ,

I'm hoping vis.js will be there going forward as I use visNetwork frequently with shiny? I noticed it wasn't the default in the new version. DiagrammerR uses visNetwork, but still....

[jeremiedb]

Hello, I can't make any promise regarding the graph.viz features as I'm not an oficial maintainer of the package, I actually just added some features to it following bugs with an update to DiagrammeR. That being said, I personnally appreciate having both options, since I find cleaner the DiagrammeR rendering of LSTM/RNN models while the vis.js can be helpful for very deep networks.

You can get the latest pre-compiled package for Windows (CPU only) from here, by the time Drat is updated:

install.packages("https://github.com/jeremiedb/mxnet_winbin/raw/master/mxnet.zip", repos = NULL)

[dockstreet]

perfect thanks again!