Struct dfdx::nn::tensor_collection::TensorOptions

#[non_exhaustive]
pub struct TensorOptions<S: Shape, E: Dtype, D: Device<E>> {
    pub do_gradient_update: bool,
    pub reset: fn(_: &mut Tensor<S, E, D>) -> Result<(), D::Err>,
    pub shape: S,
}

Options to change behavior of ModuleVisitor

Fields (Non-exhaustive)

Non-exhaustive structs could have additional fields added in future. Therefore, non-exhaustive structs cannot be constructed in external crates using the traditional Struct { .. } syntax; cannot be matched against without a wildcard ..; and struct update syntax will not work.

do_gradient_update: bool

Whether the tensor should be updated with gradients

reset: fn(_: &mut Tensor<S, E, D>) -> Result<(), D::Err>

How to reset the tensor in the future.

shape: S

The Shape that BuildModule uses to construct the tensor

Implementations

impl<S: Shape, E: Dtype, D: Device<E>> TensorOptions<S, E, D>

pub fn reset_to_zeros() -> Selfwhere D: ZeroFillStorage<E>, S: ConstShape,

A tensor that should be updated with gradients & reset to 0

pub fn reset_to_ones() -> Selfwhere D: OneFillStorage<E>, S: ConstShape,

A tensor that should be updated with gradients & reset to 1

pub fn reset_with( reset: fn(_: &mut Tensor<S, E, D>) -> Result<(), D::Err> ) -> Selfwhere S: ConstShape,

A tensor that should be updated with gradients & reset with the fn passed in

pub fn detached( reset: fn(_: &mut Tensor<S, E, D>) -> Result<(), D::Err> ) -> Selfwhere S: ConstShape,

A tensor that should NOT be updated with gradients & reset with the fn passed in