Struct tract_core::analyser::types::TensorFact

pub struct TensorFact {
    pub datum_type: TypeFact,
    pub shape: ShapeFact,
    pub value: ValueFact,
}

Partial information about a tensor.

The task of the analyser is to tag every edge in the graph with information about the tensors that flow through it - specifically their datum_type, their shape and possibly their value. During the analysis, however, we might only know some of that information (say, for instance, that an edge only carries tensors of rank 4, but without knowing their precise dimension).

This is where tensor facts come in: they hold partial information about the datum_type, shape and value of tensors that might flow through an edge of the graph. The analyser will first tag each edge with a fact, starting with the most general one and specializing it at each iteration. Eventually, it will reach a fixed point that - hopefully - holds enough information.

Fields

datum_type: TypeFact

shape: ShapeFact

value: ValueFact

Methods

impl TensorFact

pub fn new() -> TensorFact

Constructs the most general tensor fact possible.

pub fn any() -> TensorFact

pub fn dt(dt: DatumType) -> TensorFact

pub fn dt_shape<S: Into<ShapeFact>>(dt: DatumType, shape: S) -> TensorFact

pub fn shape<S: Into<ShapeFact>>(shape: S) -> TensorFact

pub fn with_datum_type(self, dt: DatumType) -> TensorFact

pub fn with_shape<S: Into<ShapeFact>>(self, shape: S) -> TensorFact

pub fn with_streaming_shape<S: IntoIterator<Item = Option<usize>>>(
    self,
    shape: S
) -> TensorFact

pub fn stream_info(&self) -> TractResult<Option<StreamInfo>>

pub fn format_dt_shape(&self) -> String

Trait Implementations

impl Fact for TensorFact

type Concrete = Arc<Tensor>

fn concretize(&self) -> Option<Self::Concrete>

Tries to transform the fact into a concrete value.

fn unify(&self, other: &Self) -> TractResult<Self>

Tries to unify the fact with another fact of the same type.

fn is_concrete(&self) -> bool

Returns whether the value is fully determined.

fn unify_with(&mut self, other: &Self) -> TractResult<bool>

Tries to unify the fact with another fact of the same type and update self.

fn unify_with_mut(&mut self, other: &mut Self) -> TractResult<bool>

Tries to unify the fact with another fact of the same type and update both of them.

fn unify_all(facts: &mut [&mut Self]) -> TractResult<bool>

Tries to unify all facts in the list.

impl TensorInfo for TensorFact

fn to_tensor_fact(&self) -> TensorFact

Convert to TensorFact, the most accomoding variant of TensorInfo.

impl Clone for TensorFact

fn clone(&self) -> TensorFact

Returns a copy of the value.

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

Performs copy-assignment from source.

impl Default for TensorFact

fn default() -> TensorFact

Returns the "default value" for a type.

impl<V: Into<Arc<Tensor>>> From<V> for TensorFact

fn from(v: V) -> TensorFact

Performs the conversion.

impl<'a> From<&'a Tensor> for TensorFact

fn from(t: &'a Tensor) -> TensorFact

Performs the conversion.

impl PartialEq<TensorFact> for TensorFact

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

This method tests for self and other values to be equal, and is used by ==.

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

This method tests for !=.

impl Debug for TensorFact

fn fmt(&self, formatter: &mut Formatter) -> Result

Formats the value using the given formatter.

impl<'a> TryFrom<&'a TensorFact> for TypedTensorInfo

type Error = TractError

The type returned in the event of a conversion error.

fn try_from(fact: &TensorFact) -> TractResult<TypedTensorInfo>

Performs the conversion.

impl TryFrom<TensorFact> for TypedTensorInfo

type Error = TractError

The type returned in the event of a conversion error.

fn try_from(fact: TensorFact) -> TractResult<TypedTensorInfo>

Performs the conversion.