Modernize to use keras3

.Rbuildignore

@@ -1,2 +1,6 @@
 ^.*\.Rproj$
 ^\.Rproj\.user$
+^\.claude$
+^README\.Rmd$
+^\.github$
+^Makefile$

.github/.gitignore

@@ -0,0 +1 @@
+*.html

.github/workflows/R-CMD-check.yaml

@@ -0,0 +1,51 @@
+# Workflow derived from https://github.com/r-lib/actions/tree/v2/examples
+# Need help debugging build failures? Start at https://github.com/r-lib/actions#where-to-find-help
+on:
+  push:
+    branches: [main, master]
+  pull_request:
+
+name: R-CMD-check.yaml
+
+permissions: read-all
+
+jobs:
+  R-CMD-check:
+    runs-on: ${{ matrix.config.os }}
+
+    name: ${{ matrix.config.os }} (${{ matrix.config.r }})
+
+    strategy:
+      fail-fast: false
+      matrix:
+        config:
+          - {os: macos-latest,   r: 'release'}
+          - {os: windows-latest, r: 'release'}
+          - {os: ubuntu-latest,   r: 'devel', http-user-agent: 'release'}
+          - {os: ubuntu-latest,   r: 'release'}
+          - {os: ubuntu-latest,   r: 'oldrel-1'}
+
+    env:
+      GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }}
+      R_KEEP_PKG_SOURCE: yes
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: r-lib/actions/setup-pandoc@v2
+
+      - uses: r-lib/actions/setup-r@v2
+        with:
+          r-version: ${{ matrix.config.r }}
+          http-user-agent: ${{ matrix.config.http-user-agent }}
+          use-public-rspm: true
+
+      - uses: r-lib/actions/setup-r-dependencies@v2
+        with:
+          extra-packages: any::rcmdcheck
+          needs: check
+
+      - uses: r-lib/actions/check-r-package@v2
+        with:
+          upload-snapshots: true
+          build_args: 'c("--no-manual","--compact-vignettes=gs+qpdf")'

.gitignore

@@ -3,3 +3,7 @@
 .RData
 .Ruserdata
 *.Rproj
+
+/.quarto/
+scratch.R
+README.html

DESCRIPTION

@@ -1,17 +1,20 @@
-Package: survnet
 Type: Package
-Title: Artificial neural networks for survival analysis
-Version: 0.0.5
-Date: 2018-11-12
-Author: Marvin N. Wright
-Maintainer: Marvin N. Wright <cran@wrig.de>
-Description: Artificial neural networks for survival analysis
+Package: survnet
+Title: Artificial Neural Networks for Survival Analysis
+Version: 0.0.7
+Authors@R: 
+    person(c("Marvin", "N."), "Wright", , "cran@wrig.de", role = c("aut", "cre"))
+Description: Fits artificial neural networks for survival analysis,
+    including competing risks, using 'keras3'.
 License: MIT + file LICENSE
 Depends:
-    survival, 
-    keras
+    keras3,
+    survival
 Imports:
-    magrittr
-Suggests: 
-    testthat
-RoxygenNote: 6.1.0
+    magrittr,
+    reticulate
+Suggests:
+    testthat (>= 3.0.0)
+Encoding: UTF-8
+RoxygenNote: 7.3.3
+Config/testthat/edition: 3

Makefile

@@ -0,0 +1,67 @@
+PKGNAME = `sed -n "s/Package: *\([^ ]*\)/\1/p" DESCRIPTION`
+PKGVERS = `sed -n "s/Version: *\([^ ]*\)/\1/p" DESCRIPTION`
+
+all: format doc README.md install check
+
+.PHONY: format
+format:
+	@echo "Formatting with $(shell air --version)"
+	@air format .
+
+.PHONY: doc
+doc: README.md
+	Rscript -e "usethis::use_tidy_description()"
+	Rscript -e "devtools::document()"
+
+.PHONY: build
+build:
+	Rscript -e "devtools::build()"
+
+.PHONY: vignettes
+vignettes:
+	Rscript -e "devtools::build_vignettes()"
+
+.PHONY: install
+install: doc
+	Rscript -e "pak::local_install()"
+
+.PHONY: deps 
+deps:
+	Rscript -e "pak::local_install_dev_deps()"
+
+# devtools::check() automatically redocuments, so no need to add doc here
+.PHONY: check
+check:
+	Rscript -e "devtools::check()"
+
+.PHONY: check-remote
+check-remote:
+	Rscript -e "devtools::check(remote = TRUE)"
+
+.PHONY: test-summary
+test-summary:
+	Rscript -e "devtools::test(reporter = 'summary')"
+
+.PHONY: test-slow
+test-slow:
+	Rscript -e "devtools::test(reporter = 'slow')"
+
+.PHONY: test
+test:
+	Rscript -e "devtools::test()"
+
+.PHONY: coverage
+coverage:
+	Rscript -e "covr::report(covr::package_coverage(\".\"), file = \"coverage.html\")"
+
+.PHONY: site
+site:
+	Rscript -e "pkgdown::build_site()"
+
+README.md: README.Rmd
+	Rscript -e "rmarkdown::render('README.Rmd')"
+	rm README.html
+
+clean:
+	rm -r docs
+	rm coverage.html

NAMESPACE

@@ -4,7 +4,7 @@ S3method(predict,survnet)
 export(convert_surv_cens)
 export(loss_cif_loglik)
 export(survnet)
-import(keras)
+import(keras3)
 import(survival)
 importFrom(magrittr,freduce)
 importFrom(stats,predict)

NEWS.md

@@ -0,0 +1,11 @@
+# survnet 0.0.7
+
+* Switched from deprecated `keras` package to `keras3`.
+* Use `py_require("tensorflow")` on package load to ensure TensorFlow is available.
+* Updated loss function to use `keras3` ops (`op_sum`, `op_log`, `op_clip`) instead of legacy `backend()` API.
+* Updated `optimizer_rmsprop()` to use `learning_rate` argument (replacing deprecated `lr`).
+* Updated `regularizer_l2()` to use `l2` argument (replacing deprecated `l`).
+* Removed `CUDNN_LSTM` and `CUDNN_GRU` RNN types (regular LSTM/GRU auto-use CuDNN in modern TensorFlow).
+* Wrapped examples in `\donttest{}` and added test skips for environments without TensorFlow.
+* Fixed DESCRIPTION metadata (Authors@R, title case, encoding, complete description).
+* Used `\doi{}` instead of `\url{}` for DOI references in documentation.

R/loss_functions.R

@@ -16,23 +16,23 @@
 #' @export
 #' @references 
 #'  \itemize{
-#'   \item Jeong, J. & Fine, J. (2006). Direct parametric inference for the cumulative incidence function. J R Stat Soc Ser C Appl Stat 55:187-200. \url{https://doi.org/10.1111/j.1467-9876.2006.00532.x}.
+#'   \item Jeong, J. & Fine, J. (2006). Direct parametric inference for the cumulative incidence function. J R Stat Soc Ser C Appl Stat 55:187-200. \doi{10.1111/j.1467-9876.2006.00532.x}.
 #'   \item Lee, C., Zame, W.R., Yoon, J. & van der Shaar, M. (2018). DeepHit: A deep learning approach to survival analysis with competing risks. AAAI 2018. \url{http://medianetlab.ee.ucla.edu/papers/AAAI_2018_DeepHit}.
 #'  }
 loss_cif_loglik <- function(num_intervals, num_causes = 1){
   function(y_true, y_pred) {
-    K <- backend()
-    
+    eps <- config_epsilon()
+
     # Survival and event indicators
     S <- y_true[, 1:(num_causes * num_intervals)] # Survival
     E <- y_true[, (num_causes * num_intervals + 1):(2 * num_causes * num_intervals)] # Events
-    
+
     # Likelihood part for uncensored and censored observations (0 for censored)
-    uncens <- K$sum(E * K$log(K$clip(y_pred, K$epsilon(), NULL)), axis = -1L)
-    delta <- 1 - K$sum(E, axis = -1L)
-    cens <- delta * K$log(K$clip(1 - K$sum(S * y_pred, axis = -1L), K$epsilon(), NULL))
-    
-    # Return negative log-likelihood 
+    uncens <- op_sum(E * op_log(op_clip(y_pred, eps, 1)), axis = -1L)
+    delta <- 1 - op_sum(E, axis = -1L)
+    cens <- delta * op_log(op_clip(1 - op_sum(S * y_pred, axis = -1L), eps, 1))
+
+    # Return negative log-likelihood
     -(uncens + cens)
   }
 }

R/survnet.R

@@ -12,7 +12,7 @@
 #' @param validation_split Fraction in [0,1] of the training data to be used as validation data.
 #' @param loss Loss function. 
 #' @param activation Activation function.
-#' @param rnn_type Type of RNN layers. Either \code{"LSTM"} (default), \code{"GRU"}, \code{"CUDNN_LSTM"} or \code{"CUDNN_GRU"}.
+#' @param rnn_type Type of RNN layers. Either \code{"LSTM"} (default) or \code{"GRU"}.
 #' @param skip Add skip connection from input and RNN layers to cause-specific layers.
 #' @param dropout Vector of dropout rates after each hidden layer. Use 0 for no dropout (default).
 #' @param dropout_rnn Vector of dropout rates after each recurrent layer. Use 0 for no dropout (default).
@@ -24,22 +24,24 @@
 #' @param verbose Verbosity mode (0 = silent, 1 = progress bar, 2 = one line per epoch). 
 #' 
 #' @return Fitted model.
-#' @examples 
+#' @examples
+#' \donttest{
 #' library(survival)
 #' library(survnet)
-#' 
+#'
 #' # Survival data
 #' y <- veteran[, c(3, 4)]
 #' x <- veteran[, c(-2, -3, -4)]
 #' x <- data.frame(lapply(x, scale))
 #' breaks <- c(1, 50, 100, 200, 500, 1000)
-#' 
+#'
 #' # Fit simple model
 #' fit <- survnet(y = y, x = x, breaks = breaks)
 #' plot(fit$history)
+#' }
 #' 
 #' @export
-#' @import survival keras
+#' @import survival keras3
 #' @importFrom magrittr freduce
 survnet <- function(y, 
                     x, 
@@ -60,7 +62,7 @@ survnet <- function(y,
                     l2 = rep(0, length(units)), 
                     l2_rnn = rep(0, length(units_rnn)), 
                     l2_causes = rep(0, length(units_causes)), 
-                    optimizer = optimizer_rmsprop(lr = 0.001), 
+                    optimizer = optimizer_rmsprop(learning_rate = 0.001),
                     verbose = 2) {
 
   # Force evaluation of dependent arguments
@@ -176,24 +178,18 @@ survnet <- function(y,
         return_sequences <- TRUE
       }
       if (l2_rnn[i] > 0) {
-        kernel_regularizer <- regularizer_l2(l = l2_rnn[i])
+        kernel_regularizer <- regularizer_l2(l2 = l2_rnn[i])
       } else {
         kernel_regularizer <- NULL
       }
       if (rnn_type == "LSTM") {
-        layer_lstm(units = units_rnn[i], activation = activation, return_sequences = return_sequences, 
+        layer_lstm(units = units_rnn[i], activation = activation, return_sequences = return_sequences,
                    kernel_regularizer = kernel_regularizer, name = paste0("rnn_", i))
       } else if (rnn_type == "GRU") {
-        layer_gru(units = units_rnn[i], activation = activation, return_sequences = return_sequences, 
+        layer_gru(units = units_rnn[i], activation = activation, return_sequences = return_sequences,
                    kernel_regularizer = kernel_regularizer, name = paste0("rnn_", i))
-      } else if (rnn_type == "CUDNN_LSTM") {
-        layer_cudnn_lstm(units = units_rnn[i], return_sequences = return_sequences, 
-                  kernel_regularizer = kernel_regularizer, name = paste0("rnn_", i))
-      } else if (rnn_type == "CUDNN_GRU") {
-        layer_cudnn_gru(units = units_rnn[i], return_sequences = return_sequences, 
-                         kernel_regularizer = kernel_regularizer, name = paste0("rnn_", i))
       } else {
-        stop("Unknown rnn_type.")
+        stop("Unknown rnn_type. Use 'LSTM' or 'GRU'.")
       }
 
     })
@@ -208,7 +204,7 @@ survnet <- function(y,
   # non-RNN layers
   dense_layers <- lapply(1:length(units), function(i) {
     if (l2[i] > 0) {
-      kernel_regularizer <- regularizer_l2(l = l2[i])
+      kernel_regularizer <- regularizer_l2(l2 = l2[i])
     } else {
       kernel_regularizer <- NULL
     }
@@ -257,7 +253,7 @@ survnet <- function(y,
       # Cause-specific layers
       layers <- lapply(1:length(units_causes[[i]]), function(j) {
         if (l2_causes[[i]][j] > 0) {
-          kernel_regularizer <- regularizer_l2(l = l2_causes[[i]][j])
+          kernel_regularizer <- regularizer_l2(l2 = l2_causes[[i]][j])
         } else {
           kernel_regularizer <- NULL
         }

R/zzz.R

@@ -0,0 +1,3 @@
+.onLoad <- function(libname, pkgname) {
+  reticulate::py_require("tensorflow")
+}

README.Rmd

@@ -0,0 +1,44 @@
+---
+title: "survnet: Artificial neural networks for survival analysis"
+output: github_document
+---
+
+<!-- README.md is generated from README.Rmd. Please edit that file -->
+
+```{r, include = FALSE}
+knitr::opts_chunk$set(
+  collapse = TRUE,
+  comment = "#>",
+  fig.path = "man/figures/README-",
+  out.width = "100%"
+)
+```
+
+Marvin N. Wright
+
+<!-- badges: start -->
+[![R-CMD-check](https://github.com/bips-hb/survnet/actions/workflows/R-CMD-check.yaml/badge.svghttps://github.com/bips-hb/survnet/actions/workflows/R-CMD-check.yaml/badge.svghttps://github.com/bips-hb/survnet/actions/workflows/R-CMD-check.yaml/badge.svg)](https://github.com/bips-hb/survnet/actions/workflows/R-CMD-check.yaml)
+<!-- badges: end -->
+
+
+## Installation
+
+```r
+pak::pak("bips-hb/survnet")
+```
+
+## Example
+
+```r
+library(survnet)
+
+# Survival data
+y <- veteran[, c(3, 4)]
+x <- veteran[, c(-2, -3, -4)]
+x <- data.frame(lapply(x, scale))
+breaks <- c(1, 50, 100, 200, 500, 1000)
+
+# Fit simple model
+fit <- survnet(y = y, x = x, breaks = breaks)
+plot(fit$history)
+```