Fix keras cache creation and restore

[javierluraschi]

~/.keras was not being cached properly, now it should, validated with:

library(keras)

message("~/.keras BEFORE: ", paste(dir("~/.keras"), collapse = ","))

conv_base <- application_vgg16(
  weights = "imagenet",
  include_top = FALSE,
  input_shape = c(150, 150, 3)
)

message("~/.keras AFTER: ", paste(dir("~/.keras"), collapse = ","))

with output:

I  master-replica-0 ~/.keras BEFORE: keras.json,models master-replica-0 
I  master-replica-0 > FLAGS <- flags(flag_numeric("dropout_rate", 0.4)) master-replica-0 
I  master-replica-0 > conv_base <- application_vgg16(weights = "imagenet",  master-replica-0 
I  master-replica-0 +     include_top = FALSE, input_shape = c(150, 150, 3)) master-replica-0 
I  master-replica-0 Using TensorFlow backend. master-replica-0 
I  master-replica-0 2017-12-13 20:49:14.584777: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use AVX2 instructions, but these are available on your machine and could speed up CPU computations. master-replica-0 
I  master-replica-0 2017-12-13 20:49:14.584829: W tensorflow/core/platform/cpu_feature_guard.cc:45] The TensorFlow library wasn't compiled to use FMA instructions, but these are available on your machine and could speed up CPU computations. master-replica-0 
I  master-replica-0 > message("~/.keras AFTER: ", paste(dir("~/.keras"),  master-replica-0 
I  master-replica-0 +     collapse = ",")) master-replica-0 
I  master-replica-0 ~/.keras AFTER: keras.json,models master-replica-0