Machine Learning API Reference

Models

Sequential

A sequential container for neural network layers.

init(allocator: Allocator) Sequential

Initializes a new empty Sequential model.

deinit(self: *Sequential, allocator: Allocator) void

Deinitializes the model and frees all associated memory.

add(self: *Sequential, allocator: Allocator, layer: Layer) !void

Adds a layer to the end of the model.

forward(self: *Sequential, allocator: Allocator, input: *Tensor(f32)) !*Tensor(f32)

Performs a forward pass through all layers in the model.

fit(self: *Sequential, allocator: Allocator, x_train: *Tensor(f32), y_train: *Tensor(f32), optimizer: anytype, loss_fn: anytype, epochs: usize) !void

Trains the model for a fixed number of epochs.

• optimizer: An optimizer instance (e.g., SGD) with an update method.
• loss_fn: A function that takes (allocator, prediction, target) and returns a loss Tensor.

train(self: *Sequential) void

Sets the model and all its layers to training mode.

eval(self: *Sequential) void

Sets the model and all its layers to evaluation mode.

save(self: *Sequential, allocator: Allocator, path: []const u8) !void

Saves the model architecture and weights to a binary file.

load(allocator: Allocator, path: []const u8) !Sequential

Loads a model from a binary file.

Layers

Dense

A fully connected layer.

init(allocator: Allocator, input_dim: usize, output_dim: usize, init_method: InitMethod) !Dense

Initializes a dense layer.

• init_method: .RandomUniform, .XavierUniform, or .HeNormal.

Layer (Union)

A tagged union representing any supported layer type.

• .Dense: Dense struct.
• .ReLU: void.
• .Sigmoid: void.
• .Tanh: void.
• .Softmax: void.
• .Dropout: Dropout struct.

Optimizers

SGD

Stochastic Gradient Descent optimizer.

init(lr: f32) SGD

Initializes SGD with a learning rate.

update(self: *SGD, param: *Tensor(f32)) void

Updates a parameter tensor using its gradient.