Downloads the MNIST training dataset or fetches it locally.

Returns a tuple of format:

{{images_binary, images_type, images_shape},
 {labels_binary, labels_type, labels_shape}}

If you want to one-hot encode the labels, you can:

labels_binary
|> Nx.from_binary(labels_type)
|> Nx.new_axis(-1)
|> Nx.equal(Nx.tensor(Enum.to_list(0..9)))

options

Options.

• :base_url - Dataset base URL.

  Defaults to "https://storage.googleapis.com/cvdf-datasets/mnist/"

• :train_image_file - Training set image filename.

  Defaults to "train-images-idx3-ubyte.gz"

• :train_label_file - Training set label filename.

  Defaults to "train-images-idx1-ubyte.gz"

• :cache_dir - Cache directory.

  Defaults to System.tmp_dir!()

Downloads the MNIST test dataset or fetches it locally.

options

Options.

• :base_url - Dataset base URL.

  Defaults to "https://storage.googleapis.com/cvdf-datasets/mnist/"

• :train_image_file - Training set image filename.

  Defaults to "train-images-idx3-ubyte.gz"

• :train_label_file - Training set label filename.

  Defaults to "train-images-idx1-ubyte.gz"

• :test_image_file - Test set image filename.

  Defaults to "t10k-images-idx3-ubyte.gz"

• :test_label_file - Test set label filename.

  Defaults to "t10k-labels-idx1-ubyte.gz"

• :cache_dir - Cache directory.

  Defaults to System.tmp_dir!()