Autograd API Reference

The autograd module provides automatic differentiation.

Tensor

The core data structure for autograd.

pub fn Tensor(comptime T: type) type

Lifecycle

init

Initialize a tensor.

pub fn init(allocator: Allocator, data: NDArray(T), requires_grad: bool) !*Self

deinit

Deinitialize the tensor.

pub fn deinit(self: *Self, allocator: Allocator) void

Gradients

backward

Compute gradients.

pub fn backward(self: *Self, allocator: Allocator) !void

Operations

add

Element-wise addition.

pub fn add(self: *Self, allocator: Allocator, other: *Self) !*Self

mul

Element-wise multiplication.

pub fn mul(self: *Self, allocator: Allocator, other: *Self) !*Self

matmul

Matrix multiplication.

pub fn matmul(self: *Self, allocator: Allocator, other: *Self) !*Self

Activation Functions

relu

Rectified Linear Unit.

pub fn relu(self: *Self, allocator: Allocator) !*Self

sigmoid

Sigmoid activation.

pub fn sigmoid(self: *Self, allocator: Allocator) !*Self

tanh

Hyperbolic tangent activation.

pub fn tanh(self: *Self, allocator: Allocator) !*Self

softmax

Softmax activation.

pub fn softmax(self: *Self, allocator: Allocator) !*Self

Loss Functions

mse_loss

Mean Squared Error loss.

pub fn mse_loss(self: *Self, allocator: Allocator, target: *Self) !*Self

cross_entropy_loss

Cross Entropy loss.

pub fn cross_entropy_loss(self: *Self, allocator: Allocator, target: *Self) !*Self