package nx

  1. Overview
  2. Docs
N-dimensional arrays for OCaml

Install 

dune-project
 Dependency

github.com Readme Changelog Edit opam file Versions (4)

Authors

  1. Thibaut Mattio

Maintainers

  1. Thibaut Mattio

Sources

raven-1.0.0.alpha3.tbz
sha256=96d35ce03dfbebd2313657273e24c2e2d20f9e6c7825b8518b69bd1d6ed5870f
sha512=90c5053731d4108f37c19430e45456063e872b04b8a1bbad064c356e1b18e69222de8bfcf4ec14757e71f18164ec6e4630ba770dbcb1291665de5418827d1465

doc/nx.core/Nx_core/Shape/index.html

Module Nx_core.ShapeSource

Concrete shape operations.

Concrete tensor shapes.

A shape is an array of non-negative dimension sizes in row-major order.

Sourcetype t = int array

The type for concrete shapes.

Basic operations

Sourceval numel : t -> int

numel shape is the product of dimensions in shape.

numel [||] is 1.

Sourceval equal : t -> t -> bool

equal s0 s1 is true iff s0 and s1 are structurally equal.

Strides

Sourceval c_contiguous_strides : t -> int array

c_contiguous_strides shape is the row-major stride vector of shape.

For any zero-size dimension, strides to its left are propagated with zero according to the implementation's canonical rule.

Index conversion

Sourceval ravel_index : int array -> int array -> int

ravel_index indices strides is the linear offset sum_i (indices.(i) * strides.(i)).

Raises Invalid_argument if the array lengths differ.

Note. This function does not perform bounds checks on indices.

Sourceval unravel_index : int -> t -> int array

unravel_index k shape is the multi-index of k in a C-contiguous layout of shape.

For shape = [||], k must be 0.

For zero-size shapes, only k = 0 is accepted and the result is an array of zeros with the same rank as shape.

Raises Invalid_argument if k is out of bounds for shape.

Sourceval unravel_index_into : int -> t -> int array -> unit

unravel_index_into k shape dst is like unravel_index but writes indices into dst.

dst must have length Array.length shape.

Raises Invalid_argument if k is out of bounds for shape.

Warning. If dst has the wrong length, array access may raise Invalid_argument via OCaml's bounds checks.

Shape transformations

Sourceval resolve_neg_one : t -> int array -> t

resolve_neg_one current_shape new_spec resolves a single -1 entry in new_spec using numel current_shape.

Raises Invalid_argument if:

  • new_spec contains more than one -1.
  • The inferred size is not integral with the specified dimensions.
  • The specification is incompatible with zero-size inference rules.
Sourceval broadcast : t -> t -> t

broadcast a b is the broadcasted shape of a and b using NumPy rules (right alignment; dimensions are compatible iff equal or one is 1).

Raises Invalid_argument if the shapes are not broadcast-compatible.

Sourceval broadcast_index : int array -> t -> int array

broadcast_index target_idx source_shape maps a target index to the corresponding index in source_shape under broadcasting.

Dimensions of source_shape equal to 1 map to index 0.

Sourceval broadcast_index_into : int array -> t -> int array -> unit

broadcast_index_into target_idx source_shape dst is like broadcast_index but writes into dst.

dst must have length Array.length source_shape.

Formatting

Sourceval pp : Format.formatter -> t -> unit

pp formats shapes with the same syntax as to_string.

Sourceval to_string : t -> string

to_string shape formats shape as a bracketed comma-separated list, for example [2,3,4].