Module numpy 

NumPy arrays: .npy files, and .npz archives of them.

An .npy file is a short header, a Python dict literal giving the element type (descr), the order (fortran_order) and the shape, and then the elements packed end to end. Every element is the same size, so the data is read where it is, through the strided decoder of the binary format specs (crate::fixed_records): a map of the file, decoded only where it is shown.

A structured element type gives a column per field (nested fields as outer.inner, a subarray field as an Array column); a 1-D array is one column, named for the array; a 2-D array is a grid with a column per index, 0 to n-1. More dimensions are refused with the shape. Python objects (O), which only unpickling could read, are refused and never unpickled.

An .npz archive is a zip of .npy files. Its arrays are listed on the home screen like a directory’s files; an archive of one opens it. An array stored as it is in the archive (np.savez) is read in place from a map of the archive; a compressed one (np.savez_compressed) is decompressed once to a temporary file first.

Structs

Array

An array read in place: its records and what to say about it.

FieldDef

One field of a structured type.

Header

What a .npy header says.

Member

One array of an archive.

Plan

The columns an array is read as, their starts counted from its first element, and its rows.

Simple

A type with no fields: <f8, |S10, <M8[ns].

Enums

Dtype

An element type, as descr gives it.

Literal

A Python literal, as far as a .npy header uses them.

Open

What opening a NumPy file finds.

Constants

MAGIC

The first six bytes of every .npy file.

MAX_COLUMNS

Columns one array may make, nested fields and repeated structs flattened.

MAX_GRID

A 2-D array of more columns than this is one Array column rather than a column per index.

MAX_HEADER

The longest header read. NumPy’s own reader refuses past 10,000 bytes unless told otherwise; a structured type of many fields can be longer, and this is still small.

Functions

detail

What the Info panel’s NumPy tab says of an array.

header_len

Bytes before the header’s text, and the header’s length, from the file’s first 12 bytes; None when they are not a .npy file’s.

is_archive

Whether a file is a NumPy archive: a zip file named .npz.

looks_like

Whether head, the first bytes of a file, begins a .npy file.

members

The arrays of the archive at path, in the archive’s order.

open

Open path: a .npy file, or an archive’s array wanted names, or its one array.

open_file

The array of the .npy file at path.

open_in

The array whose .npy bytes start at bytes into bytes (the whole of a file, or a member stored in an archive) and run for len bytes, its one column named name.

parse_dtype

Parse a descr.

parse_header

Parse the header at the start of bytes, which must hold all of it.

parse_literal

Parse the Python literal text: strings, integers, True, False, None, lists, tuples and dicts, nested at most [MAX_DEPTH] deep.

plan

The columns of an array of header, named name where it is one column.

schema_preview

The columns an array would open with, from its header: a .npy file, or an archive’s member. None for one that would not open.

shape_text

The shape as Python writes it: (3,), (2, 4), ().

tables

The arrays of an archive as tables, with their columns, for the home screen.