Pass predict additional args to the predict function

R/module.R

@@ -396,10 +396,12 @@ predict.luz_module_fitted <- function(object, newdata, ..., callbacks = list(),
   )
 
   pars <- rlang::list2(...)
-  if (is.null(pars$stack))
+  if (is.null(pars$stack)) {
     stack <- TRUE
-  else
+  } else {
     stack <- pars$stack
+    pars$stack <- NULL # pop from this list
+  }
 
   predict_fn <- if (is.null(ctx$model$predict)) ctx$model else ctx$model$predict
 
@@ -416,7 +418,7 @@ predict.luz_module_fitted <- function(object, newdata, ..., callbacks = list(),
         coro::loop(for(batch in ctx$data) {
           ctx$batch <- batch
           ctx$call_callbacks("on_predict_batch_begin")
-          ctx$pred[[length(ctx$pred) + 1]] <- do.call(predict_fn, list(ctx$input))
+          ctx$pred[[length(ctx$pred) + 1]] <- do.call(predict_fn, rlang::list2(ctx$input, !!!pars))
           ctx$call_callbacks("on_predict_batch_end")
         })
       }

tests/testthat/test-module.R

@@ -127,6 +127,34 @@ test_that("predict works for modules", {
 
 })
 
+test_that("predict passes additional arguments", {
+
+  base_model <- get_model()
+  model <- torch::nn_module(
+    inherit = base_model,
+    predict = function(x, scale = 1) {
+      self$forward(x) * scale
+    }
+  )
+  dl <- get_dl()
+
+  fitted <- model %>%
+    setup(
+      loss = torch::nn_mse_loss(),
+      optimizer = torch::optim_adam
+    ) %>%
+    set_hparams(input_size = 10, output_size = 1) %>%
+    fit(dl, verbose = FALSE)
+
+  pred <- predict(fitted, dl)
+  pred_scaled <- predict(fitted, dl, scale = 2)
+
+  expect_equal(
+    as.array(pred_scaled$to(device = "cpu")),
+    2 * as.array(pred$to(device = "cpu"))
+  )
+})
+
 test_that("predict can use a progress bar", {
 
   model <- get_model()