Struct TensorShapeInference 

pub struct TensorShapeInference { /* private fields */ }

Static shape inference engine for tensor operation graphs.

Tracks how many rules have been applied and how many errors occurred.

Implementations

impl TensorShapeInference

pub fn new() -> Self

Create a new inference engine.

pub fn infer(&mut self, rule: &InferenceRule) -> Result<TensorShape, String>

Infer the output shape for a given inference rule.

pub fn broadcast_shape( a: &TensorShape, b: &TensorShape, ) -> Result<TensorShape, String>

Compute the broadcast-compatible output shape of two tensors following NumPy broadcasting rules.

Rules:

1. Shapes are right-aligned.
2. Dimensions are compatible when they are equal, or one of them is 1.
3. The output dimension is the maximum of the two.

pub fn matmul_shape( a: &TensorShape, b: &TensorShape, ) -> Result<TensorShape, String>

Compute the output shape of a matrix multiplication.

For 2-D inputs (M, K) and (K, N), the output is (M, N). For higher-rank inputs the batch dimensions must be broadcast-compatible and the last two dimensions follow the matrix multiplication rule.

pub fn reshape_shape( input: &TensorShape, new_dims: &[usize], ) -> Result<TensorShape, String>

Verify that new_dims has the same total number of elements as input and return the reshaped tensor shape.

pub fn transpose_shape(input: &TensorShape) -> TensorShape

Return the shape with dimensions reversed (general transpose).

pub fn concat_shape( inputs: &[TensorShape], axis: usize, ) -> Result<TensorShape, String>

Concatenate inputs along axis. All dimensions except axis must match across all inputs.

pub fn slice_shape( input: &TensorShape, axis: usize, start: usize, end: usize, ) -> Result<TensorShape, String>

Slice input along axis from start (inclusive) to end (exclusive).

pub fn total_elements(shape: &TensorShape) -> usize

Total number of elements in a shape (product of dimensions). Returns 1 for a scalar (empty dims).

pub fn is_scalar(shape: &TensorShape) -> bool

Whether the shape represents a scalar: either empty dims or every dimension is 1.

pub fn stats(&self) -> ShapeInferenceStats

Return collected statistics.

Trait Implementations

impl Default for TensorShapeInference

fn default() -> Self

Returns the “default value” for a type.