Creation

Creation

Header: include/stratax/algorithms/Creation.hpp.

Output selection

Functions are available under stratax::creation. T must satisfy DType, which includes bool.

Call	Owning result	Initialization
zeros<T>(shape) / ones<T>(shape)	Tensor	T{} / T{1}
zeros<T>(size) / ones<T>(size)	Vector	T{} / T{1}
zeros<T>(rows, cols) / ones<T>(rows, cols)	Matrix	T{} / T{1}
full<T>(shape, value)	Tensor	Copies value
full<T>(size, value)	Vector	Copies value
full<T>(rows, cols, value)	Matrix	Copies value
identity<T>(size)	Matrix	Ones on diagonal, zeros elsewhere

Invariants and failures

Shape overloads preserve all dimensions, including zero extents. Rank-zero Shape creates an empty Tensor. Identity has shape {size, size}; size zero produces an empty rank-two Matrix. All results own their storage.

Shape element/stride overflow raises Exceptions::DimensionError. Allocation can raise std::bad_alloc or std::bad_array_new_length for an unrepresentable storage byte count. Python dimensions are checked before unsigned conversion.

Example

#include <stratax.h>
#include <cassert>
#include <type_traits>
 
int main() {
    auto z = stratax::creation::zeros<double>(stratax::core::Shape{2, 3});
    auto o = stratax::creation::ones<float>(4);
    auto f = stratax::creation::full<int>(2, 2, 7);
    auto eye = stratax::creation::identity<bool>(3);
    static_assert(std::is_same_v<decltype(eye), stratax::container::Matrix<bool>>);
    assert(z.size() == 6 && o[0] == 1.0F && f(1, 1) == 7);
    assert(eye(1, 1) && !eye(1, 2));
}

Cost

Initialization takes O(n + r) time and storage for n elements and rank r. Identity takes O(size squared) element work. Python offers the same overload selection for double containers; see Python API Reference.