nove_optimizer 0.1.2

An easy-to-use, lightweight deep learning library wrapped around Candle Tensor.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34

use nove_tensor::TensorError;
use thiserror::Error;

mod sgd;
pub use sgd::Sgd;

#[derive(Debug, Error)]
pub enum OptimizerError {
    /// Tensor errors from the `nove_tensor` crate.
    #[error(transparent)]
    TensorError(#[from] TensorError),

    /// Other errors.
    #[error("{0}")]
    OtherError(String),
}

pub trait Optimizer {
    type StepOutput;

    /// Performs an optimization step.
    ///
    /// # Returns
    /// * `Ok(Self::StepOutput)` - The output of the optimization step.
    /// * `Err(OptimizerError)` - The error when performing the optimization step.
    fn step(&mut self) -> Result<Self::StepOutput, OptimizerError>;

    /// Zeros the gradients of all parameters.
    ///
    /// # Returns
    /// * `Ok(())` - The gradients of all parameters are zeroed.
    /// * `Err(OptimizerError)` - The error when zeroing the gradients.
    fn zero_grad(&mut self) -> Result<(), OptimizerError>;
}