Input and output

np.save(file, arr)

Writes one array in NumPy's .npy format, byte for byte as NumPy writes it. Supported dtypes: bool, integers, floats and complex.

Parameters

filestring | URL | null

Path to write (the extension is appended when missing), or null to return the bytes as a Buffer.

arrArrayLike

Array to save.

Returns

Buffer when `file` is null, otherwise undefined

Example

np.save(null, [1, 2, 3]).length; // => 152
np.save(null, [1, 2, 3]).subarray(1, 6).toString(); // => "NUMPY"

np.save(file: FileLike, arr: ArrayLike): any

np.load(file, [options])

Reads a .npy file (returns an NDArray) or a .npz archive (returns an NpzFile). Reads big-endian data and Fortran-ordered arrays. Pickled object arrays and mmapMode are not supported.

Parameters

filestring | URL | Buffer | Uint8Array | ArrayBuffer

Path, or the file contents.

[options.mmapMode]null

Only null (no memory mapping).

Returns

NDArray | NpzFile

Example

np.load(np.save(null, [[1, 2], [3, 4]])); // => [[1, 2], [3, 4]]
np.load(np.save(null, np.ones(2, { dtype: "float32" }))).dtype.name; // => "float32"

np.load(file: string | URL | BytesLike, options?: LoadOptions | undefined): NDArray | NpzFile

np.savez(file, ...arrays, [named])

Writes several arrays to an uncompressed .npz zip archive. Positional arrays are named arr_0, arr_1, ...; a trailing plain object maps names to arrays (NumPy keyword arguments). The output is byte-identical to NumPy's.

Parameters

filestring | URL | null

Path to write (the extension is appended when missing), or null to return the bytes as a Buffer.

arraysArrayLike[]

Arrays to store.

[named]Record<string, ArrayLike>

Arrays stored under their keys.

Returns

Buffer when `file` is null, otherwise undefined

Example

np.load(np.savez(null, [1, 2], { w: [3] })).files; // => ["w", "arr_0"]

np.savez(file: FileLike, ...arrays: (ArrayLike | NamedArrays)[]): any

np.savezCompressed(file, ...arrays, [named])

Like savez, but the members are deflate-compressed (with Node's zlib).

Parameters

filestring | URL | null

Path to write (the extension is appended when missing), or null to return the bytes as a Buffer.

arraysArrayLike[]

Arrays to store.

[named]Record<string, ArrayLike>

Arrays stored under their keys.

Returns

Buffer when `file` is null, otherwise undefined

Example

np.load(np.savezCompressed(null, { x: np.zeros(3) })).get("x"); // => [0, 0, 0]

np.savezCompressed(file: FileLike, ...arrays: (ArrayLike | NamedArrays)[]): any

np.load(npz) → NpzFile

A loaded .npz archive. files lists the member names without .npy; get(name) decodes one member (NumPy npz[name]); also has, keys, entries, iteration over names and close().

Returns

NpzFile

Example

const z = np.load(np.savez(null, { a: [1, 2], b: [true] }));
z instanceof np.NpzFile; // => true
z.get("a.npy"); // => [1, 2]
[...z]; // => ["a", "b"]
z.has("c"); // => false

np.load(file: string | URL | BytesLike, options?: LoadOptions | undefined): NDArray | NpzFile

np.loadtxt(fname, [options])

Reads a numeric table from a text file. Blank lines and comments are skipped and every row must have the same number of columns. The result is squeezed (one row or one column gives a 1-D array), then ndmin and unpack are applied. Only numeric and bool dtypes are supported (bool cells are integers).

Parameters

fnamestring | URL | Uint8Array | string[]

Path (.gz is decompressed), the contents as bytes, or a list of lines.

[options.dtype]DTypeLike

Default float64.

[options.delimiter]string | null

One character; null (default) splits on whitespace.

[options.comments]string | string[] | null

Comment prefixes (default "#").

[options.skiprows]number

Leading lines to skip.

[options.usecols]number | number[]

Columns to read; negative indices count from the end.

[options.maxRows]number

Read at most this many rows.

[options.unpack]boolean

Transpose the result.

[options.ndmin]0 | 1 | 2

Minimum number of dimensions.

[options.quotechar]string

Quote character for fields that contain the delimiter.

Returns

NDArray

Example

np.loadtxt(["# x y", "1 2", "3 4"]); // => [[1, 2], [3, 4]]
np.loadtxt(Buffer.from("1,2\n3,4\n"), { delimiter: ",", usecols: 1, dtype: "int32" }); // => [2, 4]
np.loadtxt(["1 2 3"], { ndmin: 2 }).shape; // => [1, 3]

np.loadtxt(fname: TextSource, options?: LoadtxtOptions | undefined): NDArray

np.savetxt(fname, X, [options])

Writes a 1-D or 2-D array as text with Python %-formatting (d i u o x X e E f F g G s). A 1-D array is written as one column. With one format, complex values are written as (re+imj). A path ending in .gz is gzip-compressed.

Parameters

fnamestring | URL | null

Path to write, or null to return the text.

XArrayLike

1-D or 2-D data.

[options.fmt]string | string[]

One format, one per column, or a whole-row format. Default "%.18e".

[options.delimiter]string

Column separator (default " ").

[options.newline]string

Line terminator (default "\n").

[options.header]string

Text written before the data, each line prefixed by comments.

[options.footer]string

Text written after the data.

[options.comments]string

Prefix for header and footer lines (default "# ").

Returns

string when `fname` is null, otherwise undefined

Example

np.savetxt(null, [[1, 2], [3, 4]], { fmt: "%d", delimiter: "," }); // => "1,2\n3,4\n"
np.savetxt(null, [0.5, 1e16], { fmt: "%s", header: "v" }); // => "# v\n0.5\n1e+16\n"

np.savetxt(fname: string | URL | null, X: ArrayLike, options?: SavetxtOptions | undefined): string | undefined

np.genfromtxt(fname, [options])

Reads a table and fills missing or invalid cells. Cells that do not convert become filling values: NaN for floats, -1 for integers, false for bool, unless loose is false (then only missingValues are filled). Bool cells are true/false in any case. A numeric dtype is required (no type inference, names, converters or masks).

Parameters

fnamestring | URL | Uint8Array | string[]

Path, contents as bytes, or a list of lines.

[options.dtype]DTypeLike

Default float64.

[options.delimiter]string | number | number[]

Separator, a field width, or field widths. Default: whitespace.

[options.comments]string | null

Comment marker (default "#").

[options.skipHeader]number

Lines to skip at the start.

[options.skipFooter]number

Lines to drop at the end.

[options.missingValues]string | string[] | {col: value} | Map

Strings that mean missing (a string is split at ,).

[options.fillingValues]value | value[] | {col: value} | Map

Replacement for missing or invalid cells.

[options.usecols]number | number[]

Columns to read.

[options.invalidRaise]boolean

Raise on rows with a wrong column count (default true), or drop them with a warning.

[options.loose]boolean

Fill unconvertible cells (default true).

[options.autostrip]boolean

Strip spaces around fields.

[options.maxRows]number

Read at most this many rows.

[options.unpack]boolean

Transpose the result.

[options.ndmin]0 | 1 | 2

Minimum number of dimensions.

Returns

NDArray

Example

np.isnan(np.genfromtxt(["1,2", "3,"], { delimiter: "," })); // => [[false, false], [false, true]]
np.genfromtxt(["1 x", "3 4"], { dtype: "int32", fillingValues: 0 }); // => [[1, 0], [3, 4]]
np.genfromtxt(["12345"], { delimiter: [2, 3], dtype: "int32" }); // => [12, 345]

np.genfromtxt(fname: TextSource, options?: GenfromtxtOptions | undefined): NDArray

np.fromregex(file, regexp, dtype)

Every match of regexp in the text is a record, and its capture groups fill the fields of dtype. Structured arrays are not available, so the result is an object with one 1-D array per field.

Parameters

filestring | URL | Uint8Array | string[]

Path, contents as bytes, or a list of lines.

regexpRegExp | string

Pattern; the g flag is added.

dtype[name, DTypeLike][]

Field names and dtypes, one per capture group.

Returns

Record<string, NDArray>

Example

const r = np.fromregex(["a=1 b=22"], /(\w)=(\d+)/, [["key", "bool"], ["n", "int32"]]);
r.n; // => [1, 22]

np.fromregex(file: TextSource, regexp: string | RegExp, dtype: FieldList): Record<string, NDArray>

np.fromfile(file, [options])

Reads raw binary data, as written by a.tofile(), or a text file of numbers separated by sep. Binary data has no header, so you must give the dtype. A trailing partial item is ignored.

Parameters

filestring | URL | Uint8Array

Path, or the file contents.

[options.dtype]DTypeLike

Default float64.

[options.count]number

Items to read; -1 (default) reads all.

[options.sep]string

Item separator; empty (default) means binary.

[options.offset]number

Bytes to skip (binary mode only).

Returns

NDArray

Example

np.fromfile(np.array([1, 2, 3]).astype("int16").tofile(null), { dtype: "int16" }); // => [1, 2, 3]
np.fromfile(Buffer.from("1, 2, 3"), { sep: ",", count: 2 }); // => [1, 2]

np.fromfile(file: string | URL | Uint8Array<ArrayBufferLike>, options?: FromfileOptions | undefined): NDArray

a.tofile(file, [options])

Method of NDArray. Writes the items in C order as raw bytes in native byte order, or with a sep as text where each item is formatted like Python str (or with format, a %-format). No shape or dtype is stored; use save to keep them.

Parameters

filestring | URL | null

Path to write, or null to return the bytes.

[options.sep]string

Item separator; empty (default) writes binary.

[options.format]string

Format applied to each item in text mode, e.g. "%.2f".

Returns

Buffer when `file` is null, otherwise undefined

Example

np.array([[1.5, 2], [3, 4]]).tofile(null, { sep: "," }).toString(); // => "1.5,2.0,3.0,4.0"
np.array([1, 2]).tofile(null, { sep: " ", format: "%03d" }).toString(); // => "001 002"
np.arange(3).astype("uint8").tofile(null).length; // => 3

a.tofile(file: string | URL | null, options?: TofileOptions | undefined): any