Error in py_call_impl(callable, dots$args, dots$keywords)...

[faltinl]

The following program has been run flawlessly with my previous setup using keras 2.1.2., kerasR 0.6.1. and tensorflow 1.4.3. :

library(kerasR)
library(keras)
model <- keras_model_sequential() %>%
  layer_dense(units = 32, activation = "relu", input_shape = c(64)) %>%
  layer_dense(units = 32, activation = "relu") %>%
  layer_dense(units = 10, activation = "softmax")

model %>% compile(
  optimizer = "rmsprop",
  loss = "categorical_crossentropy",
  metrics = c("accuracy")
)

x_train <- array(runif(1000 * 64), dim = c(1000, 64))
y_train <- array(runif(1000 * 10), dim = c(1000, 10))

model %>% fit(
  x_train,
  y_train,
  epochs = 10, # exchange to <10 to initiate a real-time display in viewer pane!
  batch_size = 128,
  verbose = 2,
  view_metrics = "auto",
  validation_split = 0.2)

After updating to tensorflow 1.5.0.9000 via devtools::install_github("rstudio/tensorflow") without noticing any problems, followed by

library(tensorflow)
install_tensorflow()

the short test with

sess = tf$Session()
hello <- tf$constant('Hello, TensorFlow!')
sess$run(hello)

was passed ok. Upon running the above NN program the following error message , shown with the 2 uppermost traceback items was obtained:

 Error in py_call_impl(callable, dots$args, dots$keywords) : 
  TypeError: update() takes from 2 to 3 positional arguments but 4 were given 

12. | stop(structure(list(message = "TypeError: update() takes from 2 to 3 positional arguments but 4 were
 given",      call = py_call_impl(callable, dots$args, dots$keywords),      cppstack = structure(list(file = "",
 line = -1L, stack = "C++ stack not available on this system"), .Names = c("file",      "line", "stack"), class =
 "Rcpp_stack_trace")), .Names = c("message",  ...
11.
(structure(function (...) 
{
    dots <- py_resolve_dots(list(...))
    result <- py_call_impl(callable, dots$args, dots$keywords) ... 
etc.

Several completely different models (using a very diverse choice of architectures) which were running flawlessly with my previous tensorflow-installation all failed at the same point of the fit()-process, showing almost the same message text in all cases.

In the meantime I have removed tensorflow 1.5.0., then repeated the installation process - with identical results.

Any ideas what could be wrong? => Tnx in advance!

[jjallaire]

This was an issue introduced in Keras 2.1.4 which we've since resolved in the R keras package, see rstudio/keras3#285

So if you update the R keras package to the latest version from CRAN then the error should go away.

[faltinl]

Tnx for info. However, when trying to install.packages("keras") as indicated at CRAN, it is installing v2.1.2. again. If, however, I try to install.packages("keras 2.1.5") I keep getting

Warning in install.packages :
  package ‘keras 2.1.5’ is not available (for R version 3.4.3)

...which is, indeed, my version of R I have installed presently.

Finally, using devtools::install_github("rstudio/keras") led to installation of keras 2.1.5.9001 - and now both the sample program above as well as one of the other programms mentioned worked smoothly. Thanks for all the hints!

[jjallaire]

The current version of keras on CRAN is v2.1.5: https://cran.r-project.org/web/packages/keras/index.html

You must be using a CRAN mirror that has an older snapshot of the package.

Glad you figured out the devtools workaround!

[faltinl]

Well - I used https://mran.microsoft.com/package/keras, where version 2.1.5, published 2018-03-25 is announced. However, upon installing as indicated, the whole process as described in my last post starts again. (I just repeated it, alas...). Only from devtools::install_github("rstudio/keras") I am able to get a working installation.
Thanks anyway!