Skip to content

API reference

fiery.xtensor makes names a first-class citizen of torch.Tensor, in the spirit of xarray. Most of its public surface lives on one class, XTensor (also available lowercase as xtensor) — an xarray-like DataArray over a live torch.Tensor, carrying self-managed named dimensions and, optionally, per-dimension coordinate labels. The reference is split by what each group of methods/functions does, rather than one page for the whole package:

  • Core — XTensor itself: names, axis descriptors, coordinates, renaming, and dtype/device conversion; plus ExtendedTensor, the generic name-aware tensor-subclass base it's built on.
  • Factories — build an XTensor directly (xzeros, xones, xarange, ...), or coerce something else into one (as_xtensor, is_xtensor).
  • Reductions — sums, means, extrema, sorts, and their nan-aware / cumulative variants.
  • Combining — concatenation/stacking and the matrix-product family.
  • Pointwise operations — elementwise arithmetic, comparison, logical, and trigonometric operations.
  • Gather, scatter & indexing — sel/isel/interp, PyTorch's gather/scatter family, and slicing/splitting.
  • Shape — reshaping, (un)squeezing, transposing/permuting, and broadcasting.
  • Data units — the physical unit of a tensor's values (see also the Data units guide).
  • Options — library-wide behaviour switches.

Every method lives on exactly one of these pages (XTensor itself is introduced on Core and referenced from the others).