R version 3.4.1 rkward_0.6.5 This other software outside of R is required: - Python - TensorFlow - Keras code example: library(keras) layer_input(shape=c(100)) > Tensor("input_9:0", shape=(?, 100), dtype=float32) input_1 <- layer_input(shape=c(100)) > Error: (list) object cannot be coerced to type 'integer' > Warning in .rk.get.structure.global("input_1") : > failure to get object input_1 In a plain R terminal this works flawlessly.
Created attachment 107929 [details] saved object from R terminal
Hi! Thanks for reporting, and sorry about the terrible delay. This works for me in the current development version of RKWard (KF5-based; version number 0.6.9something), both with R 3.3.1 and R 3.4.4. Do you have a chance to test this? Note that I did not load the tensor lib (in fact I did not even install it). Is that required to reproduce the bug?
Thank you for solving the problem. I use Ubuntu 14.04. In the RKWard-ppa the new version is not available for Trusty (which is still supported as a LTS-version of Ubuntu). Can the new version of RKWard provided for this version of Ubuntu so that I can test it?
As much as I would like to give you a better answer, I am afraid RKWard 0.7.0 will not become available for Ubuntu 14.04. As far as I can see, there are simply no KF5 libraries available for Trusty. What you _could_ do is to try the daily builds on the kde4 branch at https://launchpad.net/~rkward-devel/+archive/ubuntu/rkward-dailys-cran . These have few fixes on top of 0.6.5 (*). We will probably create another bugfix release based on this, shortly after R 3.5.0 is out (for users like you, who are still stuck with KDE4). However, the KDE4 branch is not going to see any significant development beyond this, as we simply do not have the resources for that. --- (*) To be honest, I'll be surprised, if these solve your issue. However, I have just tested, and could not reproduce the issue. Can you confirm that I should be seeing the bug when loading your .Rdata file, with library(tensorflow) loaded?
I was able to test the bugfix on another system with Ubuntu 16.04. Unfortunately, the error still occurs. Loading library(tensorflow) and my attached Rdata-file doesn't reproduce the bug. These commands are necessary to generate the bug: library(keras) input_1 <- layer_input(shape=c(100))
Git commit c97ab062227cd414578bc67d93832ca8e957dcda by Thomas Friedrichsmeier. Committed on 17/04/2018 at 09:15. Pushed by tfry into branch 'master'. Do not get confused by non-numeric dim(). M +1 -0 ChangeLog M +1 -1 rkward/rbackend/rkstructuregetter.cpp https://commits.kde.org/rkward/c97ab062227cd414578bc67d93832ca8e957dcda
Git commit 9ff8e2fa96bb6e42c58a86160ceb89e3c79f7bde by Thomas Friedrichsmeier. Committed on 17/04/2018 at 09:17. Pushed by tfry into branch 'kde4'. Do not get confused by non-numeric dim(). M +1 -0 ChangeLog M +1 -1 rkward/rbackend/rkstructuregetter.cpp https://commits.kde.org/rkward/9ff8e2fa96bb6e42c58a86160ceb89e3c79f7bde
Thanks. I had tried to shortcut installing keras, but that turned out to be needed to reproduce. This should now be fixed both for the KF5 and the KDE4 version (but not in any official release, yet).