Add EMNIST Dataset for Image Classification (issue #129)

[DerrickUnleashed]

This PR adds support for the EMNIST dataset’s Balanced split, extending the existing MNIST datasets.

• Uses the existing download and caching infrastructure for consistency and efficiency.
• Supports familiar parameters aligned with MNIST datasets.
• Includes comprehensive tests modeled after MNIST test suites.
• Updates documentation with examples and references.

[cregouby]

praise : Nice addition, thanks a lot @DerrickUnleashed , and congratulation for this first contribution !
question blocking : can we switch the parameter train=TRUE into split = c("byclass", "bymerge", "balanced", "letters", "digits", "mnist") ?
todo missing : additions to NAMESPACE and the man/emnist_dataset.Rd are missing from the P.R. Both should have been produce by running devtools::document(). Please run that function to have those two files in.
todo missing : Please add an entry in the NEWS.md file for this dataset
todo : Please add your name to the Authors@R list with role = c("ctb") in the DESCRIPTION file. We are so proud to have you as contributor !

[DerrickUnleashed]

Thanks a lot for the warm welcome!
I've made all the requested changes:

• Updated the parameter to use split = c(...) instead of train = TRUE.
• Ran devtools::document() to update the NAMESPACE and man files.
• Added an entry in NEWS.md for the EMNIST dataset.
• Added myself to the Authors@R with role = c("ctb").

Let me know if anything else needs tweaking!

[cregouby]

Your P.R currently gives me

── Failure (test-dataset-mnist.R:74:5): tests for the emnist dataset ───────────
`ds <- emnist_dataset(dir, split = split)` did not throw an error.

── Failure (test-dataset-mnist.R:74:5): tests for the emnist dataset ───────────
`ds <- emnist_dataset(dir, split = split)` did not throw an error.

── Failure (test-dataset-mnist.R:74:5): tests for the emnist dataset ───────────
`ds <- emnist_dataset(dir, split = split)` did not throw an error.

── Failure (test-dataset-mnist.R:74:5): tests for the emnist dataset ───────────
`ds <- emnist_dataset(dir, split = split)` did not throw an error.

── Failure (test-dataset-mnist.R:74:5): tests for the emnist dataset ───────────
`ds <- emnist_dataset(dir, split = split)` did not throw an error.
[ FAIL 5 | WARN 0 | SKIP 0 | PASS 99 ]

Test complete

could you fix that ?

And also inline comments.

[DerrickUnleashed]

Thanks for the clarification!

• I've fixed the test so that emnist_dataset(dir, split = split) now throws as expected when the dataset hasn't been downloaded.

• I've also removed the inline comments from the test file as requested.

Let me know if there's anything else you'd like me to update!

[cregouby]

Thanks !
Can you please resolve the merge conflicts with main that are mentioned in here ?

[DerrickUnleashed]

Merge conflicts with main have been resolved successfully. Let me know if anything else needs attention.

[cregouby]

As you change Roxygen2 content, you have to re-document the project.

Plus actually, the example fails with

Error in emnist_dataset(split = "digits", download = TRUE) : 
  argument "root" is missing, with no default

Could you please fix it ?

As the download is very long, can you add an inform user message for the start of the Download ?

[DerrickUnleashed]

I've made the necessary changes to address the issue with emnist_dataset(), including adding a default root and a user message to inform about the download.

However, while testing, I noticed similar issues in other dataset examples as well. Would you like me to go ahead and fix those too?

Error in mnist_dataset(download = TRUE) : 
  argument "root" is missing, with no default

Error in kmnist_dataset(download = TRUE) : 
  argument "root" is missing, with no default

Error in fashion_mnist_dataset(download = TRUE) : 
  argument "root" is missing, with no default

[DerrickUnleashed]

Just a quick FYI — when I run devtools::document(), it also generates changes in the following files:

• transform_resize.Rd
• transform_random_resized_crop.Rd
• transform_random_perspective.Rd
• transform_resized_crop.Rd
• transform_perspective.Rd

I'm stashing these edits.

[DerrickUnleashed]

Suggestion : Should we consider adding .RData to the .gitignore file?

[DerrickUnleashed]

I have also updated the default root as tempdir() to prevent the below errors

Error in mnist_dataset(download = TRUE) : 
  argument "root" is missing, with no default

Error in kmnist_dataset(download = TRUE) : 
  argument "root" is missing, with no default

Error in fashion_mnist_dataset(download = TRUE) : 
  argument "root" is missing, with no default

[cregouby]

praise : Thanks for the addition. We are almost there, we just need to lower the test time and footprint.

[cregouby]

Sorry, but the skip_on_cran() is missing and your last commit despite the improvement removes the code coverage on 5 different values of the "split". We need those tests to be covered. Actually, whatever the doanwload= value, the archive is downloaded only once within the test making void any optimization (like to switch to download = FALSE). Could you please fix that ?

[cregouby]

todo Sorry, but the skip_on_cran() is missing
todo your last commit despite the improvement, removes the code coverage on 5 different values of the "split =". We need those tests to be covered. Actually, whatever the download= value, the archive is downloaded only once within the test, making void any optimization (like to switch to download = FALSE). Could you please fix that ?

[DerrickUnleashed]

Thanks for pointing that out. I've just added back the skip_on_cran() (it was removed by mistake in my previous commit). I've also updated the tests to restore full coverage for the different split = values. The tests now ensure that the archive is only downloaded once. Let me know if anything else needs adjustment.

[DerrickUnleashed]

I understood my error I have forgotten to use the skip() command after using skip_on_cran() now it has less test footprint. (has reduced from ~15 mins to ~9 mins)

[DerrickUnleashed]

@cregouby I have made the requested changes...Is there anything else that needs to be updated ?