Autoencoder CIFAR Example

Trains a dense 16 → 8 → 4 → 8 → 16 autoencoder with a final sigmoid. The input is the first 16 values of one flattened, channel-first CIFAR-10 image; the target is that same vector. This is a compact reconstruction exercise, not a full-image autoencoder.

The command uses Adam and mean squared error through the public Trainer, then prints a training summary and writes the selected TrainLog JSON. It does not export reconstructed image files.

def NN.Examples.Models.Generative.Autoencoder.exeName : String

CLI subcommand name used in terminal banners and error messages.

def NN.Examples.Models.Generative.Autoencoder.defaultLogPath : System.FilePath

Default JSON loss-curve path for this command.

@[reducible, inline]

abbrev NN.Examples.Models.Generative.Autoencoder.modelConfig : TorchLean.nn.models.Generative.Config

Dense autoencoder dimensions shared by the model and data boundary.

def NN.Examples.Models.Generative.Autoencoder.batchSize : ℕ

Number of image vectors loaded for each training sample.

@[reducible, inline]

abbrev NN.Examples.Models.Generative.Autoencoder.input : Spec.Shape

Input shape: a batch of flattened CIFAR image vectors.

@[reducible, inline]

abbrev NN.Examples.Models.Generative.Autoencoder.output : Spec.Shape

Target shape: the same flattened image-vector batch, because this is reconstruction.

def NN.Examples.Models.Generative.Autoencoder.model : TorchLean.nn.Builder (TorchLean.nn.Sequential input output)

Trainable dense autoencoder.

The architecture is defined in the public model API. The command chooses the dataset, optimizer, runtime options, and logging path.

def NN.Examples.Models.Generative.Autoencoder.data (flags : RealData.CifarModelTrainFlags) : TorchLean.Trainer.Dataset input output

Public singleton dataset for compact CIFAR reconstruction.

def NN.Examples.Models.Generative.Autoencoder.train (runtime : TorchLean.Runtime.Config) (flags : RealData.CifarModelTrainFlags) : IO (TorchLean.Trainer.Result input output)

Train the compact autoencoder with the public Trainer surface.

def NN.Examples.Models.Generative.Autoencoder.main (args : List String) : IO UInt32

Executable entrypoint for CIFAR reconstruction.

The command loads one real CIFAR minibatch, builds the supervised reconstruction sample x -> x, trains the autoencoder for --steps, and writes the standard TorchLean training summary/log.