Module `Kaun_datasets`

Dataset loaders for kaun.

Each loader returns plain Nx tensors — (train_x, train_y, test_x, test_y) with axis 0 indexing examples — ready for Kaun.Data.batches2:

let train_x, train_y, test_x, test_y = Kaun_datasets.mnist () in
let batches = Data.batches2 ~shuffle:true ~batch_size:128 (train_x, train_y)

Datasets are downloaded on demand and cached locally under $RAVEN_CACHE_ROOT/datasets/ (or $XDG_CACHE_HOME/raven/datasets/).

val mnist : 
  ?fashion:bool ->
  ?normalize:bool ->
  ?data_format:[ `NCHW | `NHWC ] ->
  unit ->
  Nx.float32_t * Nx.int32_t * Nx.float32_t * Nx.int32_t

mnist () is (train_x, train_y, test_x, test_y): 60,000 training and 10,000 test examples.

Images are float32 in [0;1] (when normalize is true, the default). Labels are int32 class indices.

fashion selects Fashion-MNIST when true. Defaults to false. data_format defaults to `NCHW.

Tensor shapes:

`NCHW: images [N; 1; 28; 28], labels [N]
`NHWC: images [N; 28; 28; 1], labels [N]

Raises Failure on download or parsing errors.

val cifar10 : 
  ?normalize:bool ->
  ?data_format:[ `NCHW | `NHWC ] ->
  unit ->
  Nx.float32_t * Nx.int32_t * Nx.float32_t * Nx.int32_t

cifar10 () is (train_x, train_y, test_x, test_y): 50,000 training and 10,000 test examples.

Images are float32 in [0;1] (when normalize is true, the default). Labels are int32 class indices (0--9: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck).

data_format defaults to `NCHW.

Tensor shapes:

`NCHW: images [N; 3; 32; 32], labels [N]
`NHWC: images [N; 32; 32; 3], labels [N]

Raises Failure on download, extraction, or parsing errors.