Facts and Dimensions

class tract.fact.InferenceFact(ptr)

Tract inference fact, to be used with InferenceModel.

It can represent partial type and shape information of a Tensor during model analysis.

dump() → str

Return a human-readable representation of the fact.

class tract.fact.Fact(ptr)

Tract-core fact, to be used with Model.

It always contains the full shape (sometimes using symbolic dimensions) and item type. In some situation it can also contain the constant value of the associated tensor.

datum_type() → int

Return the element type of the fact as a DatumType integer.

rank() → int

Return the number of dimensions (axes) of the tensor.

dim(axis: int) → Dim

Return the Dim for the given axis.

dump() → str

Return a human-readable representation of the fact.

dims()

Return a list of Dim for all axes.

class tract.dim.Dim(ptr)

A possibly symbolic dimension of a tensor.

Dimensions can be concrete integers or symbolic expressions (e.g. N, N+1). Use to_int64() to extract a concrete value, or eval() to substitute symbols.

dump() → str

Return a human-readable representation of the dimension.

eval(values: Dict[str, int]) → Dim

Substitute symbols by concrete values and return the resulting dimension.

to_int64() → int

Convert to a concrete integer. Raises if the dimension is symbolic.