clip

name	clip
description	Clamp signal amplitude to a maximum absolute value
layer	L3
group	channel
metadata	{"tags":["operator","clip","clamp","artifact","threshold"],"modalities":["eeg","seeg","ecog","meg"],"step_string":"clip","analysis_goal_allowed":["classification","source_localization","feature_extraction","clinical_screening","exploratory","generic","connectivity","phase_amplitude_coupling","online_inference"],"analysis_goal_forbidden":[]}

Clip

Function

Clamps all data values to the range [-max_abs, +max_abs]. Values exceeding the threshold are replaced with the threshold value. Simple but effective artifact suppression.

Parameter Format

clip:{max_abs}

Examples:

• clip:5 — Clamp to [-5, +5] (after standard scaling)
• clip:100 — Clamp to [-100, +100] (raw microvolts)
• clip:3 — Aggressive clipping at 3 standard deviations

Parameters

Parameter	Type	Default	Description
max_abs	float	required	Maximum absolute amplitude

When to Use

• After scaling, to suppress remaining outlier samples (> 3-5 SD)
• As a simple artifact rejection: large transients are truncated
• Before ML model input to prevent extreme values from dominating gradients
• When you want to keep all time points (vs. epoch rejection which discards entire segments)

When NOT to Use

• Before ICA (clipping destroys the linearity ICA relies on)
• When absolute amplitude is meaningful (e.g., analyzing ERD/ERS magnitude)
• For spike data (amplitude carries information)

Choosing the Threshold

After scaling method	Recommended clip	Rationale
scale:robust	clip:5	IQR-based; 5 IQRs captures 99.9% of clean data
scale:standard	clip:3 or clip:5	3 SD = 99.7% of Gaussian; 5 SD = very conservative
No scaling	clip:100 (uV)	Typical EEG range is ±100 uV

Ordering

• Apply AFTER: scale (so threshold is in normalized units)
• Apply as one of the last steps
• Apply BEFORE: final output / ML feature extraction

Reference Code

Implementation

import numpy as np

max_abs = 5.0  # threshold in scaled units
data = np.clip(data, -max_abs, max_abs)

Key API

• np.clip(data, -max_abs, max_abs) — in-place-safe clamping