NDArray

a.shape · a.ndim · a.size · a.dtype · a.strides · a.itemSize · a.nbytes · a.flags

Read-only array metadata. strides are in bytes. flags holds cContiguous, fContiguous, writeable and ownData.

Returns

number[] | number | DType | ArrayFlags

Example

const a = np.zeros([2, 3], { dtype: "float32" });
a.shape;    // => [2, 3]
a.ndim;     // => 2
a.size;     // => 6
a.itemSize; // => 4
a.nbytes;   // => 24
a.strides;  // => [12, 4]

a.shape: number[]
a.ndim: number
a.size: number
a.dtype: DType
a.strides: number[]
a.itemSize: number
a.nbytes: number
a.flags: ArrayFlags

a.toArray()

Copies the data into nested JS arrays. int64/uint64 values beyond 2^53 lose precision; use toTypedArray() for those. Complex arrays throw NotImplementedError; use toTypedArray() to read them.

Returns

NestedArray

Example

np.arange(4).reshape(2, 2).toArray(); // => [[0, 1], [2, 3]]

a.item([...index])

Returns a single element as a JS number or boolean. With no index, the array must have exactly one element.

Parameters

[...index]number[]

One integer per axis (negative counts from the end).

Returns

number | boolean

Example

const a = np.array([[1, 2], [3, 4]]);
a.item(1, 0);      // => 3
np.sum(a).item();  // => 10

a.item(...index: number[]): number | boolean | Complex

a.toTypedArray()

Copies the data into a TypedArray of the matching element type (BigInt64Array for int64). Complex arrays come out interleaved as [re, im, re, im, ...].

Returns

ArrayBufferView

Example

const t = np.array([1, 2, 3], { dtype: "float32" }).toTypedArray();
t instanceof Float32Array; // => true

a.reshape(shape) / a.reshape(...dims)

Same data with a new shape. Returns a view when the layout allows it. One dimension may be -1 and is then inferred.

Parameters

shapenumber[] | ...number

New dimensions.

Returns

NDArray

Example

np.arange(6).reshape(2, 3); // => [[0, 1, 2], [3, 4, 5]]
np.arange(6).reshape([-1, 2]).shape; // => [3, 2]

a.reshape(shape: number | Shape, ...rest: (number | OrderOptions)[]): NDArray

a.T · a.transpose([...axes])

Transposed view with no copy. transpose() reverses the axes, or permutes them in the order given.

Returns

NDArray

Example

const a = np.array([[1, 2, 3], [4, 5, 6]]);
a.T; // => [[1, 4], [2, 5], [3, 6]]
np.zeros([2, 3, 4]).transpose(1, 0, 2).shape; // => [3, 2, 4]

a.T: NDArray
a.transpose(...axes: number[] | [readonly number[]]): NDArray

a.astype(dtype)

Copy converted to another dtype. Float → int truncates toward zero, as in NumPy.

Parameters

dtypeDTypeLike

Target dtype.

Returns

NDArray

Example

np.array([1.7, -2.7]).astype("int32"); // => [1, -2]

a.astype(dt: DTypeLike, opts?: AstypeOptions | undefined): NDArray

a.copy()

Deep, C-contiguous copy of the array. ravel() flattens to a view when possible; flatten() always copies.

Returns

NDArray

Example

const a = np.array([[1, 2], [3, 4]]);
const b = a.copy();
np.mayShareMemory(a, b); // => false
a.T.flatten();           // => [1, 3, 2, 4]

a.copy(opts?: OrderOptions | undefined): NDArray

String(a)

NumPy-style representation, also used by console.log.

Returns

string

Example

String(np.array([1, 2])); // => "array([1, 2])"

a.fill(value)

Sets every element to value, cast to the array's dtype like np.array(value, {dtype}) (floats truncate for integer dtypes; out-of-range integers and NaN raise ValueError).

Parameters

valuenumber | boolean | bigint | Complex | NDArray

A scalar or a size-1 array.

Returns

void

Example

const a = np.zeros([2], { dtype: "int32" });
a.fill(7.9);
a.toArray(); // => [7, 7]

a.fill(value: number | bigint | boolean | NDArray | ComplexLike): void

a.tolist()

Nested JS arrays of the elements, the same as toArray() (NumPy name).

Returns

NestedArray

Example

np.array([[1, 2], [3, 4]]).tolist(); // => [[1, 2], [3, 4]]

a.tolist(): NestedArray

a.tobytes({ order })

A new Uint8Array with the raw bytes of the elements, in "C" (default and "K"), "F" or "A" order.

Returns

Uint8Array

Example

const a = np.array([[0, 1], [2, 3]], { dtype: "uint8" });
Array.from(a.tobytes());                 // => [0, 1, 2, 3]
Array.from(a.tobytes({ order: "F" }));   // => [0, 2, 1, 3]

a.tobytes(opts?: OrderOptions | undefined): Uint8Array<ArrayBufferLike>

a.view([dtype])

A new view of the same memory, optionally reinterpreted as another dtype. With a different item size the last axis is rescaled; it must be contiguous and its byte size divisible by the new item size (otherwise ValueError, as NumPy).

Returns

NDArray

Example

const a = np.array([1, 2], { dtype: "int32" });
a.view("uint8").toArray(); // => [1, 0, 0, 0, 2, 0, 0, 0]
a.view().base === a;       // => true

a.view(dt?: DTypeLike | undefined): NDArray

a.byteswap({ inplace = false })

Reverses the bytes of every element (each component for complex). Returns a copy, or swaps in place and returns a.

Returns

NDArray

Example

np.array([1, 2], { dtype: "int16" }).byteswap(); // => [256, 512]

a.byteswap(opts?: { inplace?: boolean | undefined; } | undefined): NDArray

a.setflags({ write })

Sets the writeable flag. write: true raises ValueError when the array that owns the memory is read-only. align/uic are not supported.

Returns

void

Example

const a = np.array([1, 2]);
a.setflags({ write: false });
a.flags.writeable; // => false

a.setflags(opts?: { write?: boolean | null | undefined; align?: boolean | null | undefined; uic?: boolean | null | undefined; } | undefined): void

a.base

The array that owns the memory this array views, or null if it owns its data (NumPy base, same object identity).

Returns

NDArray | null

Example

const a = np.array([1, 2, 3]);
a.base;               // => null
a.get([1, 3]).base === a; // => true

a.base: NDArray | null

a.mT

View with the last two axes swapped (matrix transpose of a stack). Arrays with fewer than 2 dimensions raise ValueError.

Returns

NDArray

Example

np.zeros([5, 2, 3]).mT.shape; // => [5, 3, 2]

a.mT: NDArray

a.flat

A FlatIter: a 1-D, C-order view of the elements. flat.get(i) takes an integer (returns a JS scalar), a slice tuple, an integer index array or a boolean mask (returns a new array); flat.set(i, value) repeats value cyclically over the selection; a.flat = v assigns to every element. It is also iterable, and has base, index, coords, length and copy().

Returns

FlatIter

Example

const a = np.array([[1, 2], [3, 4]]);
a.T.flat.get(1);      // => 3
a.flat.get([1, 4]);   // => [2, 3, 4]
a.flat.set(np.array([0, 3]), [9, 8]);
a.toArray();          // => [[9, 2], [3, 8]]
[...a.T.flat];        // => [9, 3, 2, 8]

a.flat: FlatIter

a.astype(dtype, { order, copy, casting })

Converted copy. casting (default "unsafe") is checked with canCast first and a disallowed cast raises DTypeError; copy: false returns a itself when nothing would change; order defaults to "K".

Returns

NDArray

Example

np.array([1.7, -2.2]).astype("int32"); // => [1, -2]
const f = np.array([1.5]);
f.astype("float64", { copy: false }) === f; // => true

a.astype(dt: DTypeLike, opts?: AstypeOptions | undefined): NDArray

a.dot(b)

Same as np.dot(a, b).

Parameters

bArrayLike

An NDArray or nested JS array.

Returns

NDArray

Example

np.array([[1, 2], [3, 4]]).dot([1, 1]); // => [3, 7]

a.dot(b: ArrayLike): NDArray