Struct engram::nn::Dataset

pub struct Dataset {
    pub inputs: Tensor,
    pub targets: Tensor,
}

Fields

inputs: Tensor

targets: Tensor

Implementations

impl Dataset

pub fn new(inputs: Tensor, targets: Tensor) -> Self

Create a new dataset from the given input and target tensors. The number of samples in the input and target tensors must be the same.

Examples

let inputs = tensor![[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]];
let targets = tensor![[0.0], [1.0], [0.0]];
let dataset = Dataset::new(inputs, targets);

pub fn batches(&self, batch_size: usize) -> Vec<(Tensor, Tensor)>

Create batches of input and target tensors with the given batch size. The last batch may be smaller if the number of samples is not divisible by the batch size.

Examples

let inputs = tensor![[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]];
let targets = tensor![[0.0], [1.0], [0.0]];
let dataset = Dataset::new(inputs, targets);
let batches = dataset.batches(2);

pub fn batch(&self, start: usize, end: usize) -> (Tensor, Tensor)

Get a batch of input and target tensors [start..end].

Examples

let inputs = tensor![[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]];
let targets = tensor![[0.0], [1.0], [0.0]];
let dataset = Dataset::new(inputs, targets);
let (input_batch, target_batch) = dataset.batch(0, 2);