From 024c61251bf11f13e5555d1b073a54445aafc4f5 Mon Sep 17 00:00:00 2001
From: Sigrid Keydana <keydana@gmx.de>
Date: Thu, 27 Sep 2018 17:04:29 +0200
Subject: [PATCH] small changes in guide

---
 vignettes/eager_guide.Rmd | 65 +++++++++++++++++++++++++--------------
 vignettes/guide_keras.Rmd |  4 +--
 2 files changed, 43 insertions(+), 26 deletions(-)

diff --git a/vignettes/eager_guide.Rmd b/vignettes/eager_guide.Rmd
index 302f7eab5..b3611dbf6 100644
--- a/vignettes/eager_guide.Rmd
+++ b/vignettes/eager_guide.Rmd
@@ -12,7 +12,7 @@ menu:
     name: "Keras with eager execution"
     identifier: "keras-eager"
     parent: "keras-using-keras"
-    weight: 10
+    weight: 15
     
 ---
 
@@ -98,12 +98,18 @@ model <- iris_regression_model()
 ```
 
 At this point, the shapes of the model's weights are still unknown (note how no `input_shape` has been defined for its first layer).
-You can, however, already call the model on some test data:
+You can, however, already call the model on some dummy data:
 
 ```{r}
 model(k_constant(matrix(1:6, nrow = 2, ncol = 3)))
 ```
 
+```
+tf.Tensor(
+[[-1.1474639]
+ [-1.0472134]], shape=(2, 1), dtype=float32)
+ ```
+
 After that call, you can inspect the model's weights using 
 
 ```{r}
@@ -114,14 +120,11 @@ This will not just display the tensor shapes, but the actual weight values.
 
 ## Losses and optimizers
 
-
-> JJA: We should explain why we need to use the tf loss and optimizer here
-
 An appropriate loss function for a regression task like this is mean squared error:
 
 ```{r}
 mse_loss <- function(y_true, y_pred, x) {
-  # can't use Keras loss_mean_squared_error() here
+  # it's required to use a TensorFlow function here, not loss_mean_squared_error() from Keras
   mse <- tf$losses$mean_squared_error(y_true, y_pred)
   # here you could compute and add other losses 
   mse
@@ -131,7 +134,7 @@ mse_loss <- function(y_true, y_pred, x) {
 Note how we have to use loss functions from TensorFlow, not the Keras equivalents. In the same vein, we need to use an optimizer from the `tf$train` module.
 
 ```{r}
-# can't use Keras' optimizer_adam()
+# have to use an optimizer from tf$train, not Keras
 optimizer <- tf$train$AdamOptimizer()
 ```
 
@@ -182,8 +185,33 @@ Datasets are available in non-eager (graph) execution as well. However, in eager
 batch
 ```
 
-> JJA: show output for simple data inspection (another case up above)
-
+```
+[[1]]
+tf.Tensor(
+[[1.4 5.1 0.2]
+ [1.4 4.9 0.2]
+ [1.3 4.7 0.2]
+ [1.5 4.6 0.2]
+ [1.4 5.  0.2]
+ [1.7 5.4 0.4]
+ [1.4 4.6 0.3]
+ [1.5 5.  0.2]
+ [1.4 4.4 0.2]
+ [1.5 4.9 0.1]], shape=(10, 3), dtype=float32)
+
+[[2]]
+tf.Tensor(
+[[3.5]
+ [3. ]
+ [3.2]
+ [3.1]
+ [3.6]
+ [3.9]
+ [3.4]
+ [3.4]
+ [2.9]
+ [3.1]], shape=(10, 1), dtype=float32)
+```
 
 ## Training loop
 
@@ -212,7 +240,7 @@ for (i in seq_len(n_epochs)) {
     
   })
   
-  cat("Total loss (epoch): ", i, ": ", total_loss$numpy(), "\n")
+  cat("Total loss (epoch): ", i, ": ", as.numeric(total_loss), "\n")
 }
 ```
 
@@ -284,8 +312,6 @@ Getting predictions on the test set is just a call to `model`, just like trainin
 model(x_test)
 ```
 
-> JJA: show output?
-
 
 ## Saving and restoring model weights
 
@@ -317,14 +343,11 @@ This call saves model weights only, not the complete graph. Thus on restore, you
 checkpoint$restore(tf$train$latest_checkpoint(checkpoint_dir))
 ```
 
-To get predictions from a freshly restored model, you need to pass in the test data as a tensor.
-Two ways to achieve this are shown below: using `k_constant` on `x_test` directly or using an iterator, like we do in the training loop.
+You can then obtain predictions from the restored model, on the test set as a whole or batch-wise, using an iterator.
 
 ```{r}
-# way 1 
-model(k_constant(x_test))
+model(x_test)
 
-# way 2
 iter <- make_iterator_one_shot(test_dataset)
 until_out_of_range({
   batch <-  iterator_get_next(iter)
@@ -339,8 +362,6 @@ until_out_of_range({
 Here is the complete example.
 
 ```{r}
-reticulate::use_condaenv("tf-10-gpu", required = TRUE)
-
 library(keras)
 use_implementation("tensorflow")
 
@@ -441,7 +462,7 @@ if (!restore) {
       
     })
     
-    cat("Total loss (epoch): ", i, ": ", total_loss$numpy(), "\n")
+    cat("Total loss (epoch): ", i, ": ", as.numeric(total_loss), "\n")
     
     checkpoint$save(file_prefix = checkpoint_prefix)
   }
@@ -452,10 +473,8 @@ if (!restore) {
 
 # Get model predictions on test set ---------------------------------------
 
-# way 1 
-model(k_constant(x_test))
+model(x_test)
 
-# way 2
 iter <- make_iterator_one_shot(test_dataset)
 until_out_of_range({
   batch <-  iterator_get_next(iter)
diff --git a/vignettes/guide_keras.Rmd b/vignettes/guide_keras.Rmd
index ec0d4e349..2c80eddd4 100644
--- a/vignettes/guide_keras.Rmd
+++ b/vignettes/guide_keras.Rmd
@@ -491,7 +491,7 @@ model <- load_model_hdf5('my_model.h5')
 
 ## Eager execution
 
-TensorFlow [Eager execution](https://www.tensorflow.org/guide/eager) is an imperative programming environment that evaluates operations immediately. This is not required for Keras,
+[Eager execution](eager_guide.Rmd) is an imperative programming environment that evaluates operations immediately. This is not required for Keras,
 but is supported by the TensorFlow backend and useful for inspecting your program
 and debugging.
 
@@ -506,8 +506,6 @@ use_implementation("tensorflow")
 When using the TensorFlow implementation,  all model-building APIs are compatible with eager
 execution. 
 
-Consult the [RStudio TensorFlow blog](https://blogs.rstudio.com/tensorflow/) for examples
-using eager execution.
 
 ## Distribution