dspk.ops.math

Per-channel statistics on the device, and window designs.


`channel_mad`	Median absolute deviation of every channel, `median(\|x - median(x)\|)`.
`channel_mean_std`	Mean and population standard deviation of every channel.
`channel_noise_std`	Robust noise level of every channel.
`window`	A symmetric window of `length` values.

Functions

`channel_mad`

channel_mad(data, *, runtime: str | None = None) -> np.ndarray

Median absolute deviation of every channel, median(|x - median(x)|).

Unscaled: divide by MAD_TO_SIGMA for a standard deviation (SpikeInterface’s get_noise_levels(method="mad") does). Medians of an even count are NumPy’s.

Parameters

data : array_like
Signal (channels, samples). float32, int8, int16, uint16 and int32 arrays go to the device as they are and are converted there; other dtypes and lists become float32 first.
runtime : str, default None
A runtime from runtime.available ("cuda", "hip", "wgpu", "cpu") for this call. By default: the one chosen with runtime.set, else DSPK_RUNTIME, else the first available (GPUs first).

Returns

np.ndarray — One value per channel (channels,).

Examples

>>> import numpy as np
>>> import dspk
>>> x = np.random.default_rng(0).standard_normal((4, 30_000)).astype(np.float32)
>>> mad = dspk.ops.math.channel_mad(x)

`channel_mean_std`

channel_mean_std(
    data,
    *,
    runtime: str | None = None,
) -> tuple[np.ndarray, np.ndarray]

Mean and population standard deviation of every channel.

Parameters

data : array_like
Signal (channels, samples). float32, int8, int16, uint16 and int32 arrays go to the device as they are and are converted there; other dtypes and lists become float32 first.
runtime : str, default None
A runtime from runtime.available ("cuda", "hip", "wgpu", "cpu") for this call. By default: the one chosen with runtime.set, else DSPK_RUNTIME, else the first available (GPUs first).

Returns

np.ndarray — (channels,).
np.ndarray — (channels,), population (ddof=0).

Examples

>>> import numpy as np
>>> import dspk
>>> x = np.random.default_rng(0).standard_normal((4, 30_000)).astype(np.float32)
>>> mean, std = dspk.ops.math.channel_mean_std(x)

`channel_noise_std`

channel_noise_std(
    data,
    *,
    runtime: str | None = None,
) -> np.ndarray

Robust noise level of every channel: median(|x|) / MAD_TO_SIGMA.

The usual threshold for spike detection: unlike the standard deviation it is barely moved by the spikes themselves. The median here is the magnitude at index samples // 2 when sorted.

Parameters

data : array_like
Signal (channels, samples). float32, int8, int16, uint16 and int32 arrays go to the device as they are and are converted there; other dtypes and lists become float32 first.
runtime : str, default None
A runtime from runtime.available ("cuda", "hip", "wgpu", "cpu") for this call. By default: the one chosen with runtime.set, else DSPK_RUNTIME, else the first available (GPUs first).

Returns

np.ndarray — One value per channel (channels,).

Examples

>>> import numpy as np
>>> import dspk
>>> x = np.random.default_rng(0).standard_normal((4, 30_000)).astype(np.float32)
>>> noise = dspk.ops.math.channel_noise_std(x)

`window`

window(
    name: str,
    length: int,
    *,
    sigma: float | None = None,
    beta: float | None = None,
) -> np.ndarray

A symmetric window of length values.

Parameters

name : {"hann", "hamming", "gaussian", "kaiser"}
The window.
length : int
Number of values.
sigma : float, default None
Standard deviation, in samples; "gaussian" needs it.
beta : float, default None
Shape; "kaiser" needs it.

Returns

np.ndarray — (length,).

Raises

ValueError: For an unknown name, or sigma / beta missing from (or given to) a window.

Examples

>>> import dspk
>>> w = dspk.ops.math.window("kaiser", 51, beta=5.0)

Constants

Name	Type	Value	Description
`MAD_TO_SIGMA`	`float`	`0.67448974`