Pull in changes from remote repo

Author: Matthew T. Warkentin

.github/workflows/test-coverage-pak.yaml

@@ -23,6 +23,7 @@ jobs:
       RSPM: https://packagemanager.rstudio.com/cran/__linux__/bionic/latest
       GITHUB_PAT: ${{ secrets.GITHUB_TOKEN }}
       TORCH_INSTALL: 1
+      TORCH_TEST: 1
       DEBIAN_FRONTEND: 'noninteractive'
 
     steps:

NEWS.md

@@ -1,5 +1,9 @@
 # luz (development version)
 
+* Fixed bug in CSV logger callback that was saving the logs as a space delimited file (#52, @mattwarkentin).
+* Fixed bug in the length of the progress bar for the validation dataset (#52, @mattwarkentin).
+* `ctx$data` now refers to the current in use `data` instead of always refering to `ctx$train_data`. (#54)
+
 # luz 0.1.0
 
 * Added a `NEWS.md` file to track changes to the package.

R/callbacks.R

@@ -176,7 +176,7 @@ luz_callback_progress <- luz_callback(
     show_after <- if (getOption("luz.force_progress_bar", FALSE)) 0 else 0.2
 
     format <- paste0(c(format, abbrevs), collapse = " - ")
-    total <- if (ctx$training) length(ctx$data) else length(ctx$valid_data)
+    total <- length(ctx$data) # ctx$data is the current dataset - can be the validation or training.
 
     self$pb <- progress::progress_bar$new(
       force = getOption("luz.force_progress_bar", FALSE),

R/module.R

@@ -199,7 +199,8 @@ fit.luz_module_generator <- function(
   ctx$model$ctx <- ctx
 
   ctx$optimizers <- optimizers
-  ctx$data <- data
+
+  ctx$train_data <- data
   ctx$valid_data <- valid_data
 
   if (length(epochs) == 1) epochs <- c(0, epochs)
@@ -227,8 +228,10 @@ fit.luz_module_generator <- function(
       for (epoch in seq_len(ctx$max_epochs)) {
         ctx$epoch <- epoch
         ctx$iter <- 0L
-        ctx$call_callbacks("on_epoch_begin")
 
+        ctx$data <- ctx$train_data
+
+        ctx$call_callbacks("on_epoch_begin")
         ctx$call_callbacks("on_train_begin")
 
         coro::loop(for (batch in ctx$data) {
@@ -244,11 +247,12 @@ fit.luz_module_generator <- function(
 
         if (!is.null(ctx$valid_data)) {
 
+          ctx$data <- ctx$valid_data
           ctx$call_callbacks("on_valid_begin")
 
           ctx$iter <- 0L
           torch::with_no_grad({
-            coro::loop(for (batch in ctx$valid_data) {
+            coro::loop(for (batch in ctx$data) {
               bind_batch_to_ctx(ctx, batch)
               ctx$iter <- ctx$iter + 1L
 

man/ctx.Rd

@@ -11,7 +11,9 @@ The `ctx` object is used in luz to share information between the training loop a
 +------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
 | `optimizers`     | A named list of optimizers used during training.                                                                                                                                                                |
 +------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
-| `data`           | Dataloader passed to the `data` argument in `fit`. Modified to yield data in the selected device.                                                                                                               |
+| `data`           | Current in use dataloader. When training it's `ctx$train_data`, when doing validation its `ctx$valid_data`. It can also be the prediction dataset when in `predict`.                                            |
++------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
+| `train_data`     | Dataloader passed to the `data` argument in `fit`. Modified to yield data in the selected device.                                                                                                               |
 +------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+
 | `valid_data`     | Dataloader passed to the `valid_data` argument in `fit`. Modified to yield data in the selected device.                                                                                                         |
 +------------------+-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------+

man/rmd/ctx.Rmd

@@ -182,7 +182,7 @@ test_that("csv callback", {
       luz_callback_csv_logger(tmp)
     ))
 
-  x <- read.table(tmp, header = TRUE)
+  x <- read.table(tmp, header = TRUE, sep = ",")
   expect_equal(nrow(x), 5)
   expect_equal(names(x), c("epoch", "set", "loss"))
 
@@ -192,7 +192,7 @@ test_that("csv callback", {
       luz_callback_csv_logger(tmp)
     ))
 
-  x <- read.table(tmp, header = TRUE)
+  x <- read.table(tmp, header = TRUE, sep = ",")
 
   expect_equal(nrow(x), 10)
   expect_equal(names(x), c("epoch", "set", "loss"))