Issue with length of progress bar for validation loop

[mattwarkentin]

The length of the progress bars for both train and validation loops are only based on the number of batches for the training data loader.

I think there also needs to be a line break after the progress bar so that the metrics summary isn't on the same line.

Example:

library(luz)
library(torch)

ds <- 
  dataset(
    initialize = function(n) {
      self$x <- 
        torch_tensor(matrix(rnorm(n * 100), nrow = n))
      self$y <- 
        torch_tensor(matrix(sample(0:1, n, TRUE), nrow = n))$to(torch_float())
    },
    .length = function() {
      dim(self$x)[[1]]
    },
    .getitem = function(i) {
      list(
        self$x[i, ],
        self$y[i]
      )
    }
  )

train_dl <- dataloader(ds(1000), batch_size = 32, shuffle = TRUE)
test_dl <- dataloader(ds(500), batch_size = 32)

net <- nn_module(
  "Net",
  initialize = function() {
    self$fc1 = nn_linear(100, 10)
    self$out = nn_linear(10, 1)
  },
  forward = function(x) {
    x %>% 
      self$fc1() %>% 
      nnf_relu() %>% 
      self$out()
  }
)

fitted <- 
  net %>%
  setup(
    loss = nn_bce_with_logits_loss(),
    optimizer = optim_adam,
    metrics = list(
      luz_metric_binary_accuracy_with_logits()
    )
  ) %>%
  fit(
    data = train_dl, 
    epochs = 1, 
    valid_data = test_dl
  )

Epoch 1/1
16/32 [==========>-----------] - ETA:  1s - Loss: 0.7024  - Acc: 0.472 Train metrics: Loss: 0.6997 - Acc: 0.497
Valid metrics: Loss: 0.7024 - Acc: 0.472