Convolutional Classifier

The classifier accepts any batchShape and number of spatial axes. Each stage independently configures its convolution, activation, optional dropout, and pooling. The linear head consumes the final feature map; an empty stage list applies the head directly to the flattened input.

structure TorchLean.nn.models.CNN.Config (d : ℕ) : Type

Configuration for a convolutional classifier over d spatial axes.

• inputChannels : ℕ

  Number of channels in each input sample.

• spatial : Tensor ℕ [d]

  Extent of each spatial axis.

• stages : List (ConvPoolBlock.Config d)

  Ordered convolution/activation/pooling stages.

• classCount : ℕ

  Number of classifier logits per sample.

def TorchLean.nn.models.CNN.Config.validate {d : ℕ} (config : Config d) : Except String Unit

Validate the complete classifier geometry before allocating convolution or head parameters.

def TorchLean.nn.models.CNN.Config.validate.validateStages {d : ℕ} (channels : ℕ) (spatial : Tensor ℕ [d]) : List (ConvPoolBlock.Config d) → Except String Unit

@[reducible, inline]

abbrev TorchLean.nn.models.CNN.Config.inputShape {d : ℕ} (config : Config d) (batchShape : Shape := []) : Shape

Input tensor shape after prepending an arbitrary batch shape.

@[reducible, inline]

abbrev TorchLean.nn.models.CNN.Config.outputShape {d : ℕ} (config : Config d) (batchShape : Shape := []) : Shape

Classifier output shape with the same batch shape as the input.

def TorchLean.nn.models.cnn {d : ℕ} (config : CNN.Config d) (batchShape : Shape := []) : Builder (Sequential (config.inputShape batchShape) (config.outputShape batchShape))

Build the configured feature stages, flatten their output, and apply the linear head.

def TorchLean.nn.models.cnn.buildStages {d : ℕ} (config : CNN.Config d) (batchShape : Shape) (channels : ℕ) (spatial : Tensor ℕ [d]) (stages : List (ConvPoolBlock.Config d)) : Builder (Sequential (batchShape.concat ((spatial.to Shape).prependDim channels)) (config.outputShape batchShape))