Merge pull request #1454 from rstudio/keras3-1.0

Prepare CRAN release of keras3 1.0

DESCRIPTION

@@ -1,7 +1,7 @@
 Package: keras3
 Type: Package
 Title: R Interface to 'Keras'
-Version: 0.2.0.9000
+Version: 1.0.0
 Authors@R: c(
   person("Tomasz", "Kalinowski", role = c("aut", "cph", "cre"),
          email = "[email protected]"),

NEWS.md

@@ -1,10 +1,23 @@
-# keras3 (development version)
+# keras3 1.0.0
 
 - Chains of `layer_*` calls with `|>` now instantiate layers in the
   same order as `%>%` pipe chains: left-hand-side first (#1440).
 
 - `iterate()`, `iter_next()` and `as_iterator()` are now reexported from reticulate.
 
+
+User facing changes with upstream Keras v3.3.3:
+
+- new functions: `op_slogdet()`, `op_psnr()`
+
+- `clone_model()` gains new args: `call_function`, `recursive`
+  Updated example usage. 
+
+- `op_ctc_decode()` strategy argument has new default: `"greedy"`.
+  Updated docs. 
+
+- `loss_ctc()` default name fixed, changed to `"ctc"`
+
 User facing changes with upstream Keras v3.3.2:
 
 - new function: `op_ctc_decode()`

R/applications.R

@@ -2642,7 +2642,7 @@ function (input_shape = NULL, alpha = 1, include_top = TRUE,
 #'
 #' # Reference
 #' - [Searching for MobileNetV3](
-#'     https://arxiv.org/pdf/1905.02244.pdf) (ICCV 2019)
+#'     https://arxiv.org/pdf/1905.02244) (ICCV 2019)
 #'
 #' The following table describes the performance of MobileNets v3:
 #' ------------------------------------------------------------------------
@@ -2788,7 +2788,7 @@ function (input_shape = NULL, alpha = 1, minimalistic = FALSE,
 #'
 #' # Reference
 #' - [Searching for MobileNetV3](
-#'     https://arxiv.org/pdf/1905.02244.pdf) (ICCV 2019)
+#'     https://arxiv.org/pdf/1905.02244) (ICCV 2019)
 #'
 #' The following table describes the performance of MobileNets v3:
 #' ------------------------------------------------------------------------

R/losses.R

@@ -131,7 +131,7 @@ function (y_true, y_pred, from_logits = FALSE, label_smoothing = 0,
 #' Computes focal cross-entropy loss between true labels and predictions.
 #'
 #' @description
-#' According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002.pdf), it
+#' According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002), it
 #' helps to apply a focal factor to down-weight easy examples and focus more on
 #' hard examples. By default, the focal tensor is computed as follows:
 #'
@@ -157,7 +157,7 @@ function (y_true, y_pred, from_logits = FALSE, label_smoothing = 0,
 #'     when `from_logits=TRUE`) or a probability (i.e, value in `[0., 1.]` when
 #'     `from_logits=FALSE`).
 #'
-#' According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002.pdf), it
+#' According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002), it
 #' helps to apply a "focal factor" to down-weight easy examples and focus more
 #' on hard examples. By default, the focal tensor is computed as follows:
 #'
@@ -274,13 +274,13 @@ function (y_true, y_pred, from_logits = FALSE, label_smoothing = 0,
 #' @param alpha
 #' A weight balancing factor for class 1, default is `0.25` as
 #' mentioned in reference [Lin et al., 2018](
-#' https://arxiv.org/pdf/1708.02002.pdf).  The weight for class 0 is
+#' https://arxiv.org/pdf/1708.02002).  The weight for class 0 is
 #' `1.0 - alpha`.
 #'
 #' @param gamma
 #' A focusing parameter used to compute the focal factor, default is
 #' `2.0` as mentioned in the reference
-#' [Lin et al., 2018](https://arxiv.org/pdf/1708.02002.pdf).
+#' [Lin et al., 2018](https://arxiv.org/pdf/1708.02002).
 #'
 #' @param from_logits
 #' Whether to interpret `y_pred` as a tensor of
@@ -450,7 +450,7 @@ function (y_true, y_pred, from_logits = FALSE, label_smoothing = 0,
 #' `class_weights`. We expect labels to be provided in a `one_hot`
 #' representation.
 #'
-#' According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002.pdf), it
+#' According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002), it
 #' helps to apply a focal factor to down-weight easy examples and focus more on
 #' hard examples. The general formula for the focal loss (FL)
 #' is as follows:

R/metrics.R

@@ -3,7 +3,7 @@
 #' Computes the binary focal crossentropy loss.
 #'
 #' @description
-#' According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002.pdf), it
+#' According to [Lin et al., 2018](https://arxiv.org/pdf/1708.02002), it
 #' helps to apply a focal factor to down-weight easy examples and focus more on
 #' hard examples. By default, the focal tensor is computed as follows:
 #'

R/model-training.R

@@ -269,17 +269,20 @@ function (object, x = NULL, y = NULL, ..., batch_size = NULL,
                                verbose = as_model_verbose_arg),
                           ignore = "object",
                           force = "verbose")
-    args[["return_dict"]] <- FALSE
+
+    ## return_dict=TRUE because object$metrics_names returns wrong value
+    ## (e.g., "compile_metrics" instead of "mae")
+    args[["return_dict"]] <- TRUE
 
     if(inherits(args$x, "tensorflow.python.data.ops.dataset_ops.DatasetV2") &&
        !is.null(args$batch_size))
       stop("batch_size can not be specified with a TF Dataset")
 
     result <- do.call(object$evaluate, args)
-    if (length(result) > 1L) {
-      result <- as.list(result)
-      names(result) <- object$metrics_names
-    }
+    # if (length(result) > 1L) { ## if return_dict=FALSE
+    #   result <- as.list(result)
+    #   names(result) <- object$metrics_names
+    # }
 
     tfruns::write_run_metadata("evaluation", unlist(result))
 
@@ -761,11 +764,12 @@ function (object, x, y = NULL, sample_weight = NULL, ...)
     result <- object$test_on_batch(as_array(x),
                                    as_array(y),
                                    as_array(sample_weight), ...,
-                                   return_dict = FALSE)
-    if (length(result) > 1L) {
-      result <- as.list(result)
-      names(result) <- object$metrics_names
-    } else if (is_scalar(result)) {
+                                   return_dict = TRUE)
+    # if (length(result) > 1L) {
+    #   result <- as.list(result)
+    #   names(result) <- object$metrics_names
+    # } else
+    if (is_scalar(result)) {
       result <- result[[1L]]
     }
     result
@@ -824,11 +828,12 @@ function (object, x, y = NULL, sample_weight = NULL, class_weight = NULL)
                                     as_array(y),
                                     as_array(sample_weight),
                                     class_weight = as_class_weight(class_weight),
-                                    return_dict = FALSE)
-    if (length(result) > 1L) {
-      result <- as.list(result)
-      names(result) <- object$metrics_names
-    } else if (is_scalar(result)) {
+                                    return_dict = TRUE)
+    # if (length(result) > 1L) {
+    #   result <- as.list(result)
+    #   names(result) <- object$metrics_names
+    # } else
+    if (is_scalar(result)) {
       result <- result[[1L]]
     }
 

R/optimizers-schedules.R

@@ -7,7 +7,7 @@
 #' SGDR: Stochastic Gradient Descent with Warm Restarts.
 #'
 #' For the idea of a linear warmup of our learning rate,
-#' see [Goyal et al.](https://arxiv.org/pdf/1706.02677.pdf).
+#' see [Goyal et al.](https://arxiv.org/pdf/1706.02677).
 #'
 #' When we begin training a model, we often want an initial increase in our
 #' learning rate followed by a decay. If `warmup_target` is an int, this