Merge pull request #889 from dfalbel/prepare-release

Prepare 2.2.5.0 release
rstudio · Oct 8, 2019 · d357c68 · d357c68
2 parents b76a0e2 + 07a7406
commit d357c68
Show file tree

Hide file tree

Showing 21 changed files with 93 additions and 23 deletions.
diff --git a/.Rbuildignore b/.Rbuildignore
@@ -34,3 +34,4 @@
 ^logo.png$
 
 ^appveyor\.yml$
+^CRAN-RELEASE$
diff --git a/DESCRIPTION b/DESCRIPTION
@@ -1,7 +1,7 @@
 Package: keras
 Type: Package
 Title: R Interface to 'Keras'
-Version: 2.2.4.1.9001
+Version: 2.2.5.0
 Authors@R: c(
   person("Daniel", "Falbel", role = c("ctb", "cph", "cre"), email = "[email protected]"),
   person("JJ", "Allaire", role = c("aut", "cph")),
@@ -28,7 +28,7 @@ Depends:
 Imports:
     generics (>= 0.0.1),
     reticulate (>= 1.10), 
-    tensorflow (>= 1.13.1),
+    tensorflow (>= 2.0.0),
     tfruns (>= 1.0),
     magrittr, 
     zeallot,
@@ -45,7 +45,4 @@ SystemRequirements: Keras >= 2.0 (https://keras.io)
 Roxygen: list(markdown = TRUE)
 RoxygenNote: 6.1.1
 VignetteBuilder: knitr
-Remotes:
-  rstudio/reticulate,
-  rstudio/tensorflow
 
diff --git a/NAMESPACE b/NAMESPACE
@@ -362,6 +362,7 @@ export(layer_zero_padding_1d)
 export(layer_zero_padding_2d)
 export(layer_zero_padding_3d)
 export(load_model_hdf5)
+export(load_model_tf)
 export(load_model_weights_hdf5)
 export(load_model_weights_tf)
 export(load_text_tokenizer)
@@ -431,6 +432,7 @@ export(regularizer_l2)
 export(reset_states)
 export(run_dir)
 export(save_model_hdf5)
+export(save_model_tf)
 export(save_model_weights_hdf5)
 export(save_model_weights_tf)
 export(save_text_tokenizer)

diff --git a/R/initializers.R b/R/initializers.R
@@ -227,7 +227,7 @@ initializer_identity <- function(gain = 1.0) {
 #' 
 #' @section References:
 #'  - [Self-Normalizing Neural Networks](https://arxiv.org/abs/1706.02515) 
-#'  - [Efficient Backprop](http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf)
+#'  - Efficient Backprop, \cite{LeCun, Yann et al. 1998}
 #'
 #' @family initializers
 #'
@@ -356,7 +356,6 @@ initializer_he_uniform <- function(seed = NULL) {
 #' @inheritParams initializer_random_normal
 #'   
 #' @section References: LeCun 98, Efficient Backprop,
-#'   http://yann.lecun.com/exdb/publis/pdf/lecun-98b.pdf
 #'   
 #' @family initializers
 #'   

diff --git a/man/get_weights.Rd b/man/get_weights.Rd
diff --git a/man/initializer_lecun_normal.Rd b/man/initializer_lecun_normal.Rd
diff --git a/man/initializer_lecun_uniform.Rd b/man/initializer_lecun_uniform.Rd
diff --git a/man/model_to_json.Rd b/man/model_to_json.Rd
diff --git a/man/model_to_yaml.Rd b/man/model_to_yaml.Rd
diff --git a/man/save_model_hdf5.Rd b/man/save_model_hdf5.Rd
diff --git a/man/save_model_tf.Rd b/man/save_model_tf.Rd
diff --git a/man/save_model_weights_hdf5.Rd b/man/save_model_weights_hdf5.Rd
diff --git a/man/serialize_model.Rd b/man/serialize_model.Rd
diff --git a/tests/testthat.R b/tests/testthat.R
@@ -5,4 +5,6 @@ if (Sys.getenv("TENSORFLOW_EAGER") == "TRUE")
 
 library(keras)
 
-test_check("keras")
+if (identical(Sys.getenv("NOT_CRAN"), "true")) {
+  test_check("keras") 
+}
diff --git a/tests/testthat/test-applications.R b/tests/testthat/test-applications.R
@@ -34,7 +34,7 @@ test_succeeds("keras pre-built models can be instantiated", {
   application_inception_v3()
 })
 
-test_that("can use any input_shape", {
+test_succeeds("can use any input_shape", {
   x <- application_vgg16(weights = NULL, input_shape = shape(NULL, NULL, 3), include_top = FALSE)
   expect_s3_class(x, "keras.engine.training.Model")
 })

diff --git a/tests/testthat/test-generators.R b/tests/testthat/test-generators.R
@@ -124,7 +124,8 @@ test_succeeds("R function can be used as custom generator with multiple inputs",
   input1 <- layer_input(shape = 1)
   input2 <- layer_input(shape = 1)
 
-  out <- layer_add(list(input1, input2))
+  out <- layer_add(list(input1, input2)) %>%
+    layer_dense(units = 1)
 
   model <- keras_model(list(input1, input2), out)
 
@@ -144,7 +145,8 @@ test_succeeds("Fixed validation_data instead of generator with fit_generator", {
   input1 <- layer_input(shape = 1)
   input2 <- layer_input(shape = 1)
 
-  out <- layer_add(list(input1, input2))
+  out <- layer_add(list(input1, input2)) %>% 
+    layer_dense(units = 1)
 
   model <- keras_model(list(input1, input2), out)
 

diff --git a/tests/testthat/test-model-persistence.R b/tests/testthat/test-model-persistence.R
@@ -125,6 +125,11 @@ test_succeeds("model can be exported to saved model format", {
   if (!tensorflow::tf_version() >= "1.14") skip("Needs TF >= 1.14")
 
   model <- define_and_compile_model()
+  data <- matrix(rexp(1000*784), nrow = 1000, ncol = 784)
+  labels <- matrix(round(runif(1000*10, min = 0, max = 9)), nrow = 1000, ncol = 10)
+
+  model %>% fit(data, labels, epochs = 2, verbose = 0)
+
   model_dir <- tempfile()
   dir.create(model_dir)
 

diff --git a/tests/testthat/test-model.R b/tests/testthat/test-model.R
@@ -85,16 +85,24 @@ test_succeeds("can call model with R objects", {
   model <- keras_model_sequential() %>% 
     layer_dense(units = 1, input_shape = 1)
 
-  model(matrix(runif(10), ncol = 1))
+  model(
+    tensorflow::tf$convert_to_tensor(
+      matrix(runif(10), ncol = 1), 
+      dtype = tensorflow::tf$float32
+    )
+  )
 
   input1 <- layer_input(shape = 1)
   input2 <- layer_input(shape = 1)
 
   output <- layer_concatenate(list(input1, input2))
 
   model <- keras_model(list(input1, input2), output)
-
-  model(list(matrix(runif(10), ncol = 1), matrix(runif(10), ncol = 1)))
+  l <- lapply(
+    list(matrix(runif(10), ncol = 1), matrix(runif(10), ncol = 1)), 
+    function(x) tensorflow::tf$convert_to_tensor(x, dtype = tensorflow::tf$float32)
+  )
+  model(l)
 })
 
 

diff --git a/tests/testthat/utils.R b/tests/testthat/utils.R
@@ -7,13 +7,18 @@ skip_if_no_keras <- function(required_version = NULL) {
 
 
 test_succeeds <- function(desc, expr, required_version = NULL) {
-  IPython <- reticulate::import("IPython")
-  py_capture_output <- IPython$utils$capture$capture_output
+
+  py_capture_output <- NULL
+  if (reticulate::py_module_available("IPython")) {
+    IPython <- reticulate::import("IPython")
+    py_capture_output <- IPython$utils$capture$capture_output
+  }
+
   invisible(
     capture.output({
-      with(py_capture_output(), {
-        test_that(desc, {
-          skip_if_no_keras(required_version)
+      test_that(desc, {
+        skip_if_no_keras(required_version)
+        with(py_capture_output(), {
           expect_error(force(expr), NA)
         })
       })

diff --git a/vignettes/about_keras_models.Rmd b/vignettes/about_keras_models.Rmd
@@ -65,7 +65,7 @@ predictions <- layer_concatenate(c(encoded_a, encoded_b), axis=-1) %>%
 model <- keras_model(inputs = c(tweet_a, tweet_b), outputs = predictions)
 ```
 
-Learn more by reading the [Guide to the Functional API](unctional_api.html).
+Learn more by reading the [Guide to the Functional API](functional_api.html).
 
 ### Custom
 

diff --git a/vignettes/why_use_keras.Rmd b/vignettes/why_use_keras.Rmd
@@ -90,7 +90,7 @@ As such, your Keras model can be trained on a number of different hardware platf
 
 - Keras has [built-in support for multi-GPU data parallelism](https://keras.rstudio.com/reference/multi_gpu_model.html)
 - [Horovod](https://github.com/uber/horovod), from Uber, has first-class support for Keras models
-- Keras models [can be turned into TensorFlow Estimators](https://www.tensorflow.org/versions/master/api_docs/python/tf/keras/estimator/model_to_estimator) and trained on [clusters of GPUs on Google Cloud](https://cloud.google.com/solutions/running-distributed-tensorflow-on-compute-engine)
+- Keras models can be turned into TensorFlow Estimators and trained on [clusters of GPUs on Google Cloud](https://cloud.google.com/solutions/running-distributed-tensorflow-on-compute-engine)
 - Keras can be run on Spark via [Dist-Keras](https://github.com/cerndb/dist-keras) (from CERN) and [Elephas](https://github.com/maxpumperla/elephas)
 
 ---
Original file line number	Diff line number	Diff line change
Expand Up		@@ -34,3 +34,4 @@
		^logo.png$

		^appveyor\.yml$
		^CRAN-RELEASE$