Loads and preprocesses the QMNIST dataset, including optional support for the NIST digit subset.
qmnist_dataset(
root = tempdir(),
split = "train",
transform = NULL,
target_transform = NULL,
download = FALSE
)(string): Root directory of dataset where
MNIST/processed/training.pt and MNIST/processed/test.pt exist.
(string, optional) Which subset to load: one of "train", "test", or "nist".
Defaults to "train". The "nist" option loads the full NIST digits set.
(callable, optional): A function/transform that takes in an
PIL image and returns a transformed version. E.g,
transform_random_crop().
(callable, optional): A function/transform that takes in the target and transforms it.
(bool, optional): If true, downloads the dataset from the internet and puts it in root directory. If dataset is already downloaded, it is not downloaded again.
An R6 dataset object compatible with the {torch} package, providing indexed access
to (image, label) pairs from the specified QMNIST subset.
This dataset is an extended version of the original MNIST, offering more samples and precise label information. It is suitable for benchmarking modern machine learning models and can serve as a drop-in replacement for MNIST in most image classification tasks.
"train": 60,000 training examples (compatible with MNIST)
"test": 60,000 test examples (extended QMNIST test set)
"nist": Entire NIST digit dataset (for advanced benchmarking)
if (FALSE) { # \dontrun{
qmnist <- qmnist_dataset(split = "train", download = TRUE)
first_item <- qmnist[1]
# image in item 1
first_item$x
# label of item 1
first_item$y
} # }