AIM-PIbd-32-Kurbanova-A-A/aimenv/Lib/site-packages/matplotlib/mlab.pyi
2024-10-02 22:15:59 +04:00

101 lines
3.5 KiB
Python

from collections.abc import Callable
import functools
from typing import Literal
import numpy as np
from numpy.typing import ArrayLike
def window_hanning(x: ArrayLike) -> ArrayLike: ...
def window_none(x: ArrayLike) -> ArrayLike: ...
def detrend(
x: ArrayLike,
key: Literal["default", "constant", "mean", "linear", "none"]
| Callable[[ArrayLike, int | None], ArrayLike]
| None = ...,
axis: int | None = ...,
) -> ArrayLike: ...
def detrend_mean(x: ArrayLike, axis: int | None = ...) -> ArrayLike: ...
def detrend_none(x: ArrayLike, axis: int | None = ...) -> ArrayLike: ...
def detrend_linear(y: ArrayLike) -> ArrayLike: ...
def psd(
x: ArrayLike,
NFFT: int | None = ...,
Fs: float | None = ...,
detrend: Literal["none", "mean", "linear"]
| Callable[[ArrayLike, int | None], ArrayLike]
| None = ...,
window: Callable[[ArrayLike], ArrayLike] | ArrayLike | None = ...,
noverlap: int | None = ...,
pad_to: int | None = ...,
sides: Literal["default", "onesided", "twosided"] | None = ...,
scale_by_freq: bool | None = ...,
) -> tuple[ArrayLike, ArrayLike]: ...
def csd(
x: ArrayLike,
y: ArrayLike | None,
NFFT: int | None = ...,
Fs: float | None = ...,
detrend: Literal["none", "mean", "linear"]
| Callable[[ArrayLike, int | None], ArrayLike]
| None = ...,
window: Callable[[ArrayLike], ArrayLike] | ArrayLike | None = ...,
noverlap: int | None = ...,
pad_to: int | None = ...,
sides: Literal["default", "onesided", "twosided"] | None = ...,
scale_by_freq: bool | None = ...,
) -> tuple[ArrayLike, ArrayLike]: ...
complex_spectrum = functools.partial(tuple[ArrayLike, ArrayLike])
magnitude_spectrum = functools.partial(tuple[ArrayLike, ArrayLike])
angle_spectrum = functools.partial(tuple[ArrayLike, ArrayLike])
phase_spectrum = functools.partial(tuple[ArrayLike, ArrayLike])
def specgram(
x: ArrayLike,
NFFT: int | None = ...,
Fs: float | None = ...,
detrend: Literal["none", "mean", "linear"] | Callable[[ArrayLike, int | None], ArrayLike] | None = ...,
window: Callable[[ArrayLike], ArrayLike] | ArrayLike | None = ...,
noverlap: int | None = ...,
pad_to: int | None = ...,
sides: Literal["default", "onesided", "twosided"] | None = ...,
scale_by_freq: bool | None = ...,
mode: Literal["psd", "complex", "magnitude", "angle", "phase"] | None = ...,
) -> tuple[ArrayLike, ArrayLike, ArrayLike]: ...
def cohere(
x: ArrayLike,
y: ArrayLike,
NFFT: int = ...,
Fs: float = ...,
detrend: Literal["none", "mean", "linear"] | Callable[[ArrayLike, int | None], ArrayLike] = ...,
window: Callable[[ArrayLike], ArrayLike] | ArrayLike = ...,
noverlap: int = ...,
pad_to: int | None = ...,
sides: Literal["default", "onesided", "twosided"] = ...,
scale_by_freq: bool | None = ...,
) -> tuple[ArrayLike, ArrayLike]: ...
class GaussianKDE:
dataset: ArrayLike
dim: int
num_dp: int
factor: float
data_covariance: ArrayLike
data_inv_cov: ArrayLike
covariance: ArrayLike
inv_cov: ArrayLike
norm_factor: float
def __init__(
self,
dataset: ArrayLike,
bw_method: Literal["scott", "silverman"]
| float
| Callable[[GaussianKDE], float]
| None = ...,
) -> None: ...
def scotts_factor(self) -> float: ...
def silverman_factor(self) -> float: ...
def covariance_factor(self) -> float: ...
def evaluate(self, points: ArrayLike) -> np.ndarray: ...
def __call__(self, points: ArrayLike) -> np.ndarray: ...