Utilities

np.mayShareMemory(a, b)

true if the two arrays may share memory, i.e. one is a view of the other or both view the same buffer.

Parameters

a, bNDArray

Arrays to compare.

Returns

boolean

Example

const a = np.arange(6);
np.mayShareMemory(a, a.reshape(2, 3)); // => true
np.mayShareMemory(a, a.copy());        // => false

np.mayShareMemory(a: NDArray, b: NDArray): boolean

np.copy(a, { order? })

A copy of a. order is "K" (default, keep the layout), "A", "C" or "F".

Parameters

aArrayLike

An NDArray, nested JS array or scalar.

orderstring

Memory order of the copy.

Returns

NDArray

Example

np.copy(np.ones([2, 3], { order: "F" })).strides; // => [8, 16]

np.copy(a: NDArray | NestedArray, options?: { order?: MemoryOrder | null | undefined; } | undefined): NDArray

np.ascontiguousarray(a, { dtype? })

a as a C-contiguous array with at least one dimension; no copy when it already is.

Parameters

aArrayLike

An NDArray, nested JS array or scalar.

Returns

NDArray

Example

np.ascontiguousarray(np.arange(6).reshape(2, 3).T).strides; // => [16, 8]

np.ascontiguousarray(a: NDArray | NestedArray, options?: ArrayOptions | undefined): NDArray

np.asfortranarray(a, { dtype? })

a as an F-contiguous (column-major) array with at least one dimension; no copy when it already is.

Parameters

aArrayLike

An NDArray, nested JS array or scalar.

Returns

NDArray

Example

np.asfortranarray(np.arange(6).reshape(2, 3)).strides; // => [8, 16]

np.asfortranarray(a: NDArray | NestedArray, options?: ArrayOptions | undefined): NDArray

np.memoryStats()

Count and total size of the native buffers currently alive. Native memory is freed when the JS objects are garbage-collected.

Returns

{ buffers: number, bytes: number }

Example

const s = np.memoryStats();
typeof s.bytes; // => "number"

np.lib.stride_tricks.asStrided(a, [shape], [strides])

View of the same memory with arbitrary shape and byte strides. It is unchecked, as in NumPy: wrong strides can read outside the buffer's logical data.

Parameters

aNDArray

Base array.

[shape]number[]

View shape.

[strides]number[]

Strides in bytes.

Returns

NDArray

Example

const a = np.arange(0, 5, 1, { dtype: "int32" });
np.lib.stride_tricks.asStrided(a, [3, 3], [4, 4]); // => [[0, 1, 2], [1, 2, 3], [2, 3, 4]]

np.ndindex(...shape)

Generator of every index of shape (dimensions as arguments or one array) in C order. ndindex() yields [] once.

Returns

Generator<number[]>

Example

[...np.ndindex(2, 2)]; // => [[0, 0], [0, 1], [1, 0], [1, 1]]

np.ndindex(...shape: (number | readonly number[])[]): Generator<number[], any, any>

np.ndenumerate(a)

Generator of [index, value] pairs in C order; values are JS scalars.

Returns

Generator<[number[], number | boolean | Complex]>

Example

[...np.ndenumerate(np.array([[1, 2], [3, 4]]))]; // => [[[0, 0], 1], [[0, 1], 2], [[1, 0], 3], [[1, 1], 4]]

np.ndenumerate(a: Operand): Generator<[number[], ScalarValue], any, any>

np.nditer(op | ops, { flags, order = "K", opFlags })

Read-only multi-dimensional iterator (NDIter). Each step yields a 0-d read-only view (an array of views for several operands, which broadcast together). "K" walks memory order. Flags: multi_index, c_index, f_index, zerosize_ok; members multiIndex, index, iterindex, itersize, shape, ndim, nop, operands, value, finished, iternext(), reset(). Buffering, external_loop and writable operands raise NotImplementedError.

Returns

NDIter

Example

const out = [];
for (const x of np.nditer(np.array([[1, 2], [3, 4]]).T)) out.push(x.item());
out; // => [1, 2, 3, 4]
const it = np.nditer([np.array([[1], [2]]), np.array([10, 20])]);
[...it].map(([x, y]) => x.item() + y.item()); // => [11, 21, 12, 22]

np.arrayRepr(a, { maxLineWidth?, precision?, suppressSmall? })

NumPy array_repr: array(...) text with shape=/dtype= added when they cannot be inferred. Float digits come from a C++ port of NumPy's Dragon4 (shortest unique digits per dtype). String(a) / a.toString() return the same text.

Returns

string

Example

np.arrayRepr(np.array([1, 2])); // => "array([1, 2])"
np.arrayRepr(np.array([1.5, -2, NaN])); // => "array([ 1.5, -2. ,  nan])"
np.arrayRepr(np.array([-1, 10], { dtype: "int8" })); // => "array([-1, 10], dtype=int8)"

np.arrayRepr(a: AnyArray, opts?: ArrayReprOptions | undefined): string

np.arrayStr(a, { maxLineWidth?, precision?, suppressSmall? })

NumPy array_str (Python str(a)): elements separated by spaces; a 0-d array prints like a NumPy scalar.

Returns

string

Example

np.arrayStr(np.array([1, 2, 3])); // => "[1 2 3]"
np.arrayStr(np.array(1e16)); // => "1e+16"

np.arrayStr(a: AnyArray, opts?: ArrayReprOptions | undefined): string

np.array2string(a, { maxLineWidth?, precision?, suppressSmall?, separator?, prefix?, suffix?, formatter?, threshold?, edgeitems?, sign?, floatmode?, legacy? })

NumPy array2string. formatter maps NumPy formatter names (all, bool, int, float, complexfloat, int_kind, float_kind, complex_kind) to JS callbacks that get each element and return a string. Only legacy: false is supported (others raise NotImplementedError).

Returns

string

Example

np.array2string(np.array([1, 2, 3]), { separator: "," }); // => "[1,2,3]"
np.array2string(np.array([1.123, 2]), { floatmode: "fixed", precision: 2 }); // => "[1.12 2.00]"

np.array2string(a: AnyArray, opts?: Array2StringOptions | undefined): string

np.formatFloatPositional(x, { precision?, unique?, fractional?, trim?, sign?, padLeft?, padRight?, minDigits? })

NumPy format_float_positional (Dragon4). x is a JS number (float64) or a size-1 NDArray, formatted in its own float dtype (ints/bool as float64; complex raises DTypeError).

Returns

string

Example

np.formatFloatPositional(1); // => "1."
np.formatFloatPositional(0.3, { minDigits: 20 }); // => "0.29999999999999998890"
np.formatFloatPositional(np.array(0.1, { dtype: "float32" }), { unique: false, precision: 10 }); // => "0.1000000015"

np.formatFloatPositional(x: number | boolean | NDArray, opts?: FormatFloatPositionalOptions | undefined): string

np.formatFloatScientific(x, { precision?, unique?, trim?, sign?, padLeft?, expDigits?, minDigits? })

NumPy format_float_scientific (Dragon4), with the same argument rules as formatFloatPositional.

Returns

string

Example

np.formatFloatScientific(123.456, { precision: 2 }); // => "1.23e+02"
np.formatFloatScientific(1e100, { expDigits: 4 }); // => "1.e+0100"

np.formatFloatScientific(x: number | boolean | NDArray, opts?: FormatFloatScientificOptions | undefined): string

np.setPrintoptions({ precision?, threshold?, edgeitems?, linewidth?, suppress?, nanstr?, infstr?, sign?, floatmode?, formatter?, legacy?, overrideRepr? })

NumPy set_printoptions: module-wide print options. Omitted options stay unchanged, except formatter and overrideRepr, which reset on every call (like NumPy).

Returns

void

Example

np.setPrintoptions({ precision: 2 });
String(np.array([1 / 3])); // => "array([0.33])"
np.setPrintoptions({ precision: 8 });

np.setPrintoptions(opts?: PrintOptionsInput | undefined): void

np.getPrintoptions()

NumPy get_printoptions: a copy of the current print options (camelCase keys).

Returns

PrintOptions

Example

np.getPrintoptions().threshold; // => 1000

np.getPrintoptions(): PrintOptions

np.printoptions(opts, fn)

NumPy's printoptions context manager as a callback: opts apply while fn() runs and are restored afterwards, also if it throws. Returns fn's result.

Returns

T

Example

np.printoptions({ precision: 2 }, () => String(np.array([2 / 3]))); // => "array([0.67])"

np.printoptions<T>(opts: PrintOptionsInput, fn: () => T): T

np.baseRepr(number, [base], [padding])

String of an integer in base (2 to 36, default 2), with padding zeros added on the left. Negative numbers get a minus sign. Accepts number, bigint or a 0-d integer array.

Parameters

numbernumber | bigint | NDArray

Integer to convert.

[base]number

Base, 2 to 36 (default 2).

[padding]number

Zeros to prepend (default 0).

Returns

string

Example

np.baseRepr(255, 16); // => "FF"
np.baseRepr(-7, 2, 3); // => "-000111"

np.baseRepr(number: IntegerLike, base?: number | undefined, padding?: number | undefined): string

np.binaryRepr(num, [options])

Binary string of an integer. Without width, negative numbers get a minus sign; with width, they are written in two's complement. A width that is too small raises ValueError.

Parameters

numnumber | bigint | NDArray

Integer to convert.

[options.width]number

Output length (zero-padded, or two's complement for negatives).

Returns

string

Example

np.binaryRepr(5); // => "101"
np.binaryRepr(-5); // => "-101"
np.binaryRepr(-5, { width: 8 }); // => "11111011"

np.binaryRepr(num: IntegerLike, options?: BinaryReprOptions | undefined): string

np.pi

Mathematical constant π ≈ 3.14159…

Returns

number

Example

np.pi; // => 3.141592653589793

np.pi: number

np.e

Euler's number, the base of natural logarithms, ≈ 2.71828…

Returns

number

Example

np.e; // => 2.718281828459045

np.e: number

np.inf

IEEE 754 positive infinity. Aliases: np.PINF, np.Inf, np.Infinity.

Returns

number

Example

np.inf > 1e308; // => true

np.inf: number

np.nan

IEEE 754 not-a-number. Alias: np.NaN.

Returns

number

Example

np.nan !== np.nan; // => true

np.nan: number

np.euler_gamma

Euler–Mascheroni constant γ ≈ 0.5772156649…

Returns

number

Example

Math.abs(np.euler_gamma - 0.5772156649015329) < 1e-15; // => true

np.euler_gamma: number

np.True_

Python True (JS true). Provided for NumPy API parity.

Returns

boolean

Example

np.True_ === true; // => true

np.True_: true

np.False_

Python False (JS false). Provided for NumPy API parity.

Returns

boolean

Example

np.False_ === false; // => true

np.False_: false

np.PINF

Positive infinity (alias of np.inf).

Returns

number

Example

np.PINF === Infinity; // => true

np.NINF

Negative infinity.

Returns

number

Example

np.NINF === -Infinity; // => true

np.PZERO

Positive zero.

Returns

number

Example

1 / np.PZERO === Infinity; // => true

np.NZERO

Negative zero.

Returns

number

Example

1 / np.NZERO === -Infinity; // => true

np.vectorize(fn, [options])

Wraps a JS callback so it is applied element-wise over its inputs, which are broadcast to a common shape. Returns a callable with a pyfunc property. otypes[0] forces the output dtype; otherwise it is inferred from the first call. signature must be null (scalar-in / scalar-out only).

Parameters

fn(...args: unknown[]) => unknown

The scalar function to apply element-wise.

[options.otypes]DTypeLike[]

Output dtype(s); only the first element is used.

[options.signature]null

Must be null or omitted (generalised ufunc signatures are not supported).

Returns

VectorizedFn

Example

const vf = np.vectorize((x, y) => x + y);
vf([1, 2], [10, 20]).toArray(); // => [11, 22]
vf.pyfunc(3, 4); // => 7

np.vectorize(fn: (...args: unknown[]) => unknown, opts?: VectorizeOptions | undefined): VectorizedFn

np.shares_memory(a, b, [options])

Returns true if arrays a and b share any underlying memory. Delegates to native buffer identity and byte-range overlap. maxWork is accepted for NumPy API compatibility but ignored.

Parameters

aNDArray

First array.

bNDArray

Second array.

[options.maxWork]number

Accepted but not used.

Returns

boolean

Example

const a = np.array([1, 2, 3]);
const b = a.slice([[0, 2]]);
np.shares_memory(a, b); // => true
np.shares_memory(a, np.array([1, 2, 3])); // => false

np.cumsum(a, [axis], [options])

Cumulative sum of array elements along the given axis. If axis is omitted or null, the array is flattened first.

Parameters

aArrayLike

Input array.

[axis]number | null

Axis to accumulate along; null flattens first.

[options.dtype]DTypeLike

Accumulator dtype.

Returns

NDArray

Example

np.cumsum([1, 2, 3]).toArray(); // => [1, 3, 6]
np.cumsum([[1, 2], [3, 4]], 0).toArray(); // => [[1, 2], [4, 6]]

np.cumsum(a: NDArray | NestedArray, axis?: number | null | undefined, opts?: { dtype?: DTypeLike | null | undefined; } | undefined): NDArray

np.cumprod(a, [axis], [options])

Cumulative product of array elements along the given axis. If axis is omitted or null, the array is flattened first.

Parameters

aArrayLike

Input array.

[axis]number | null

Axis to accumulate along; null flattens first.

[options.dtype]DTypeLike

Accumulator dtype.

Returns

NDArray

Example

np.cumprod([1, 2, 3, 4]).toArray(); // => [1, 2, 6, 24]
np.cumprod([[1, 2], [3, 4]], 1).toArray(); // => [[1, 2], [3, 12]]

np.cumprod(a: NDArray | NestedArray, axis?: number | null | undefined, opts?: { dtype?: DTypeLike | null | undefined; } | undefined): NDArray

np.nancumsum(a, opts?) · np.nancumprod(a, opts?)

Like cumsum/cumprod but NaN values are treated as 0 / 1 respectively (i.e. they are skipped).

Parameters

aArrayLike

Input array.

[opts.axis]number | null

Axis; default flattened.

[opts.dtype]DTypeLike

Accumulator dtype.

Returns

NDArray

Example

np.nancumsum([1, NaN, 2]).toArray(); // => [1, 1, 3]
np.nancumprod([2, NaN, 3]).toArray(); // => [2, 2, 6]

np.nancumsum(a: ArrayLike, opts?: CumsumOptions | undefined): NDArray
np.nancumprod(a: ArrayLike, opts?: CumsumOptions | undefined): NDArray

new np.busdaycalendar([options])

A business-day calendar combining a weekmask (which weekdays are business days) and optional holidays. Used with is_busday, busday_count, and busday_offset.

Parameters

[options.weekmask]string | boolean[]

7-character bit string "1111100", a space-separated list of day names like "Mon Tue Wed Thu Fri", or a boolean array. Defaults to Mon–Fri.

[options.holidays](string | number | bigint)[]

List of holiday dates as ISO-8601 strings or epoch-day integers.

Returns

busdaycalendar

Example

new np.busdaycalendar({ weekmask: "Mon Tue Wed Thu Fri" }).weekmask[0]; // => true

np.is_busday(dates, [options])

Return true for each date that is a business day (not a weekend and not a holiday).

Parameters

datesstring | number | DatetimeArray | Array

One or more dates.

[options.weekmask]string | boolean[]

Which weekdays count as business days. Defaults to Mon–Fri.

[options.holidays](string | number | bigint)[]

Holidays to exclude.

[options.busdaycal]busdaycalendar

Pre-built calendar; overrides weekmask/holidays.

Returns

boolean | boolean[]

Example

np.is_busday("2023-01-16"); // => true

np.is_busday(dates: string | number | bigint | DatetimeArray | (string | number | bigint | DatetimeArray)[], options?: IsBusdayOptions | undefined): boolean | boolean[]

np.busday_count(begindates, enddates, [options])

Count the number of business days in [begindates, enddates). Negative when end < begin.

Parameters

begindatesstring | number | DatetimeArray

Start date(s), inclusive.

enddatesstring | number | DatetimeArray

End date(s), exclusive.

[options.weekmask]string | boolean[]

Which weekdays count. Defaults to Mon–Fri.

[options.holidays](string | number | bigint)[]

Holidays to exclude.

[options.busdaycal]busdaycalendar

Pre-built calendar.

Returns

number | number[]

Example

np.busday_count("2023-01-01", "2023-01-08"); // => 5

np.busday_count(begindates: string | number | bigint | DatetimeArray, enddates: string | number | bigint | DatetimeArray, options?: BusdayCountOptions | undefined): number | number[]

np.busday_offset(dates, offsets, [options])

Shift dates by offsets business days. The roll option controls what to do when a date falls on a non-business day before stepping.

Parameters

datesstring | number | DatetimeArray | Array

Starting date(s).

offsetsnumber | number[]

Number of business days to advance (positive) or retreat (negative).

[options.roll]string

"raise" (default), "nat", "forward", "following", "backward", "preceding", "modifiedfollowing", "modifiedpreceding".

[options.weekmask]string | boolean[]

Which weekdays count. Defaults to Mon–Fri.

[options.holidays](string | number | bigint)[]

Holidays to exclude.

[options.busdaycal]busdaycalendar

Pre-built calendar.

Returns

DatetimeArray | DatetimeArray[]

Example

np.busday_offset("2023-01-16", 5).toString(); // => "2023-01-23"

np.busday_offset(dates: string | number | bigint | DatetimeArray | (string | number | bigint | DatetimeArray)[], offsets: number | number[], options?: BusdayOffsetOptions | undefined): DatetimeArray | DatetimeArray[] | null

np.sharesMemory(a, b, [options])

Return true if a and b share the same underlying memory buffer (NumPy np.shares_memory). Unlike mayShareMemory, this is always an exact check.

Parameters

aNDArray

First array.

bNDArray

Second array.

[options.maxWork]number

Ignored (always exact).

Returns

boolean

Example

const a = np.arange(6);
const b = a.reshape([2, 3]);
np.sharesMemory(a, b); // => true
np.sharesMemory(a, np.array([1, 2])); // => false

np.sharesMemory(a: NDArray, b: NDArray, _opts?: SharesMemoryOptions | undefined): boolean