Real Dataset Paths

Small tutorial fixtures live in NN.Examples.Data.ToyPaths.

This module names the default paths used by examples that train on externally downloaded datasets prepared by scripts/datasets/download_example_data.py.

The files are intentionally not committed: CIFAR-10 and text corpora are real datasets with their own licenses/citation expectations, and CI should not download hundreds of megabytes implicitly.

def NN.Examples.Data.RealPaths.defaultDataDir : System.FilePath

Default root for user-downloaded real datasets.

def NN.Examples.Data.RealPaths.takeDataDir (args : List String) (default : System.FilePath := defaultDataDir) : Except String (System.FilePath × List String)

Parse an optional --data-dir PATH flag for real-data examples.

def NN.Examples.Data.RealPaths.cifar10Dir (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Directory containing prepared CIFAR-10 .npy arrays.

def NN.Examples.Data.RealPaths.cifar10TrainX (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Prepared CIFAR-10 training images, shape (N, 3, 32, 32), float32 in [0, 1].

def NN.Examples.Data.RealPaths.cifar10TrainY (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Prepared CIFAR-10 training labels, shape (N,), float32 integer labels 0..9.

def NN.Examples.Data.RealPaths.cifar10TestX (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Prepared CIFAR-10 test images, shape (N, 3, 32, 32), float32 in [0, 1].

def NN.Examples.Data.RealPaths.cifar10TestY (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Prepared CIFAR-10 test labels, shape (N,), float32 integer labels 0..9.

def NN.Examples.Data.RealPaths.imagenet64Dir (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Directory containing a user-prepared ImageNet-style 64x64 subset.

TorchLean does not redistribute ImageNet. Users point scripts/datasets/torchlean_data_convert.py image-folder at their own ImageNet/ILSVRC, ImageNet-compatible, or Tiny-ImageNet-style directory tree and write the converted arrays here.

def NN.Examples.Data.RealPaths.imagenet64TrainX (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Prepared ImageNet-style training images, shape (N, 3, 64, 64), float32 in [0, 1].

def NN.Examples.Data.RealPaths.imagenet64TrainY (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Prepared ImageNet-style training labels, shape (N,), float32 integer class ids.

The converter assigns ids by sorted subdirectory name when --labels-from-dirs is used.

def NN.Examples.Data.RealPaths.householdPowerDir (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Directory containing prepared UCI household-power forecasting windows.

def NN.Examples.Data.RealPaths.householdPowerX (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Prepared UCI household-power inputs, shape (N, 24, 1), float32 normalized to [0, 1].

Each row is a 24-hour window of hourly mean Global_active_power.

def NN.Examples.Data.RealPaths.householdPowerY (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Prepared UCI household-power targets, shape (N, 24, 1), float32 normalized to [0, 1].

Each target row is the corresponding input window shifted by one hour.

def NN.Examples.Data.RealPaths.autoMpgDir (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Directory containing the prepared UCI Auto MPG tabular regression CSV.

def NN.Examples.Data.RealPaths.autoMpgCsv (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Prepared UCI Auto MPG CSV with normalized columns x1..x7,y.

def NN.Examples.Data.RealPaths.textDir (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Directory containing downloaded text corpora.

def NN.Examples.Data.RealPaths.tinyShakespeare (dataDir : System.FilePath := defaultDataDir) : System.FilePath

Karpathy tiny-shakespeare corpus.

def NN.Examples.Data.RealPaths.tinyStoriesValid (dataDir : System.FilePath := defaultDataDir) : System.FilePath

TinyStories validation split, useful for small local language-model smoke training.