Struct Tensor 

pub struct Tensor {
    pub shape: Vec<usize>,
    pub data: Vec<f32>,
}

An N-dimensional tensor stored in row-major order.

Fields

shape: Vec<usize>

data: Vec<f32>

Implementations

impl Tensor

pub fn zeros(shape: Vec<usize>) -> Self

pub fn ones(shape: Vec<usize>) -> Self

pub fn rand(shape: Vec<usize>, rng: u64) -> Self

Pseudo-random tensor using a simple xorshift seeded from rng.

pub fn from_vec(data: Vec<f32>, shape: Vec<usize>) -> Self

pub fn scalar(v: f32) -> Self

Scalar tensor.

pub fn get(&self, indices: &[usize]) -> f32

pub fn set(&mut self, indices: &[usize], val: f32)

pub fn slice(&self, ranges: &[Range<usize>]) -> Tensor

Slice along each axis with the given ranges. Produces a new tensor whose shape matches the range extents.

pub fn add(&self, other: &Tensor) -> Tensor

pub fn sub(&self, other: &Tensor) -> Tensor

pub fn mul(&self, other: &Tensor) -> Tensor

pub fn scale(&self, s: f32) -> Tensor

pub fn matmul(a: &Tensor, b: &Tensor) -> Tensor

2-D matrix multiply: (M, K) x (K, N) -> (M, N).

pub fn transpose(&self) -> Tensor

Transpose the last two dimensions. For 2-D tensors this is the standard matrix transpose.

pub fn sum(&self) -> f32

pub fn mean(&self) -> f32

pub fn max(&self) -> f32

pub fn min(&self) -> f32

pub fn argmax(&self, axis: usize) -> Tensor

Argmax along a given axis, returning a tensor with that axis removed.

pub fn reshape(&self, new_shape: Vec<usize>) -> Tensor

pub fn flatten(&self) -> Tensor

pub fn squeeze(&self) -> Tensor

Remove all size-1 dimensions.

pub fn unsqueeze(&self, dim: usize) -> Tensor

Insert a size-1 dimension at dim.

pub fn broadcast_to(&self, target: &[usize]) -> Tensor

Broadcast this tensor to the target shape, repeating data as needed.

pub fn relu(&self) -> Tensor

pub fn sigmoid(&self) -> Tensor

pub fn tanh_act(&self) -> Tensor

pub fn softmax(&self, axis: usize) -> Tensor

Softmax along axis.

pub fn gelu(&self) -> Tensor

GELU activation: x * 0.5 * (1 + tanh(sqrt(2/pi) * (x + 0.044715 * x^3)))

pub fn conv2d(&self, kernel: &Tensor, stride: usize, padding: usize) -> Tensor

2-D convolution. Input shape: (C_in, H, W). Kernel shape: (C_out, C_in, kH, kW). Returns shape (C_out, H_out, W_out).

pub fn max_pool2d(&self, kernel_size: usize, stride: usize) -> Tensor

Max pooling 2-D. Input shape: (C, H, W).

pub fn avg_pool2d(&self, kernel_size: usize, stride: usize) -> Tensor

Average pooling 2-D. Input shape: (C, H, W).

pub fn batch_norm( &self, mean: &Tensor, var: &Tensor, gamma: &Tensor, beta: &Tensor, eps: f32, ) -> Tensor

Batch normalization: y = gamma * (x - mean) / sqrt(var + eps) + beta. All parameter tensors must have the same total length as self.

pub fn layer_norm(&self, axis: usize, eps: f32) -> Tensor

Layer normalization along the last n dimensions starting from axis.

pub fn dropout(&self, p: f32, rng: u64, training: bool) -> Tensor

Dropout: randomly zero elements with probability p during training.

pub fn concat(tensors: &[Tensor], axis: usize) -> Tensor

Concatenate tensors along an axis.

pub fn stack(tensors: &[Tensor], axis: usize) -> Tensor

Stack tensors along a new axis.

Trait Implementations

impl Clone for Tensor

fn clone(&self) -> Tensor

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for Tensor

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl PartialEq for Tensor

fn eq(&self, other: &Tensor) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for Tensor