model_stack_map.md

Model stack map

This document is the quick reference for how library components couple together.

Maintenance rule

Always update this file when any of the following changes:

• new model
• new loss
• new optimizer
• new layer
• new activation
• dependency changes between existing components

Current map (who uses what)

LinearRegression

• file: lib/ruby_ml/models/linear_regression.rb
• uses loss: RubyML::Losses::MSE
• uses optimizer: RubyML::Optimizers::SGD
• uses layers/activations: no

LogisticRegression

• file: lib/ruby_ml/models/logistic_regression.rb
• uses loss: RubyML::Losses::BinaryCrossEntropy
• uses optimizer: RubyML::Optimizers::SGD
• uses layers/activations: no separate modules; sigmoid lives inside the model

KNN

• file: lib/ruby_ml/models/knn.rb
• uses loss: no
• uses optimizer: no
• uses layers/activations: no

KMeans

• file: lib/ruby_ml/models/kmeans.rb
• uses loss: no (minimizes inertia internally)
• uses optimizer: no
• uses layers/activations: no

Perceptron

• file: lib/ruby_ml/models/perceptron.rb
• uses loss: no (classification error implicit in the update rule)
• uses optimizer: no (internal perceptron update)
• uses layers/activations: no (uses internal step function)

NeuralNetwork

• file: lib/ruby_ml/models/neural_network.rb
• uses layers: RubyML::Layers::Dense
• uses activations: RubyML::Activations::ReLU, RubyML::Activations::Sigmoid
• uses loss: RubyML::Losses::CrossEntropy
• uses optimizer: RubyML::Optimizers::SGD

Require hub

• file: lib/ruby_ml.rb
• role: aggregate requires for losses, optimizers, layers, activations, and models.