📖 Glossary

Common terms used throughout the SimBA documentation. Terms defined here can be cross-referenced from anywhere in the docs with the :term: role (e.g. :term:`ROI` renders as ROI).

No terms match your filter.

↑

aggregate statistics

Session- or video-level summaries of classifier output (total time, bout counts, mean bout duration, latency, etc.) saved to the project logs.

AMBER

A pose-estimation pipeline for maternal-pup interaction analysis whose tracking data SimBA can import.

anchored ROI

animal-anchored ROI

An ROI (bounding box or shape) attached to, and moving with, an animal or body-part across frames - as opposed to a fixed, frame-static ROI.

annotation

labelling

The process of marking, frame-by-frame, whether a behavior is present or absent in a video. These human labels are the ground truth used to train a classifier.

arena

The experimental enclosure (open field, home cage, box, etc.) in which an animal is recorded. SimBA maps arena pixels to real-world units via px/mm and can restrict analyses to it or to ROIs drawn within it.

background subtraction

A markerless segmentation technique that models the static scene and flags pixels differing from it as the moving animal; the basis of SimBA’s blob tracking.

behavior

target

The action a SimBA classifier is trained to detect (e.g. attack, grooming, freezing). Also referred to as the target.

BENTO

A MATLAB GUI (Caltech) for browsing, annotating and analysing synchronised behavioral, tracking and neural data; SimBA imports its annotations.

blob tracking

contour tracking

Markerless tracking that segments each animal as a single connected region (a blob) via background subtraction, instead of tracking individual body-part keypoints. Useful when full pose estimation is unnecessary or unavailable.

body-part

A single tracked point on an animal (e.g. nose, left_ear, tail_base), produced by pose estimation and stored as x, y (and probability) columns.

BORIS

Behavioral Observation Research Interactive Software — a free, open-source event-logging program for manual behavioral coding; SimBA imports its exported annotations.

bounding box

The smallest axis-aligned (or rotated) rectangle that encloses an animal or a set of body-parts; used for overlap, area and proximity computations.

bout

A continuous, uninterrupted episode of a behavior — i.e. a run of consecutive frames classified as the same behavior. Bout-level statistics summarise the count, duration and timing of these episodes.

centroid

The geometric centre of a set of body-parts (or a shape); often used as a single location for an animal.

circular statistics

Statistics for angular/directional data (degrees), where 359° and 1° are close. Used in SimBA for heading, turning and directional analyses.

CLAHE

Contrast Limited Adaptive Histogram Equalization - a local contrast-enhancement step in SimBA’s video tools, used to improve tracking on low-contrast footage.

class imbalance

The common situation where the behavior of interest is present in far fewer frames than it is absent, which can bias a classifier toward predicting “absent”. Addressed with re-sampling (see oversampling, undersampling, SMOTE).

classifier

A supervised machine-learning model (typically a random forest) trained on annotated features to predict the presence of a behavior on each frame.

clustering

embedding

Unsupervised grouping of behavioral data without labels — e.g. projecting features with UMAP/t-SNE and clustering the result to discover behavioral motifs.

confusion matrix

A table of predicted vs. true labels (true/false positives and negatives) used to evaluate a classifier.

convex hull

The smallest convex polygon enclosing a set of body-parts; a common basis for animal area, shape and overlap metrics.

cross-validation

Splitting annotated data into train/test folds to estimate how well a classifier generalises to unseen frames, guarding against over-fitting.

cue light

An experimentally controlled light stimulus; SimBA’s cue-light tools detect when each light is on or off and quantify behavior and movement relative to those states.

DeepEthogram

An open-source supervised deep-learning tool that classifies behaviors directly from raw video frames; SimBA can import its predictions as labels.

DeepLabCut

DLC

A popular open-source pose estimation toolbox. SimBA imports DLC tracking data (single- and multi-animal).

directing

An animal orienting toward a target — another animal, a body-part or an ROI. SimBA quantifies directing (see directionality) from the angle between an animal’s heading and the target.

directionality

Whether, and where, an animal is facing — e.g. toward another animal, a body-part or a ROI.

discrimination threshold

probability threshold

The probability cut-off above which a frame is scored as the behavior. Raising it makes detection stricter (higher precision), lowering it more permissive (higher recall).

egocentric alignment

Re-centering and rotating each frame so a chosen body-part is fixed in position and orientation, removing the animal’s global location/heading from the analysis.

ethogram

A catalogue of the distinct behaviors an animal performs, and (in a session) their occurrence over time.

Ethovision

Noldus EthoVision XT — commercial video-tracking and behavioral-analysis software; SimBA can import its exported annotations.

FaceMap

An open-source keypoint/behavioral-motion tracking tool whose output SimBA can import as a pose estimation source.

feature

feature extraction

A numeric quantity computed per frame from pose estimation data (distances, velocities, angles, areas, etc.). Feature extraction turns raw tracking into the inputs a classifier learns from.

feature importance

A ranking of how much each feature contributes to a classifier’s decisions (e.g. Gini importance, permutation importance, or SHAP).

FPS

Frames per second — the video frame rate. Required to convert frame counts to seconds and to compute time-based metrics.

FSTTC

Forward Spike Time Tiling Coefficient — a measure of the temporal association between two behaviors (how often one tends to follow another within a time window), adapted from spike-train analysis.

Gantt plot

A timeline visualization showing when each behavior occurs across a session as horizontal bars.

geometry

Representing animals/arenas as shapes (points, lines, convex hulls, polygons, circles) via Shapely, enabling area, overlap, distance and containment computations.

heatmap

A spatial visualization of where an animal spends time (location heatmap) or where a behavior occurs, binned over the arena.

hyperparameters

The configurable settings of a classifier fixed before training — for a random forest: number of trees, max features, min samples per leaf, split criterion. Set per classifier in the SimBA training interface.

interpolation

Filling in missing body-part coordinates (e.g. dropped/occluded frames) by estimating values from neighbouring frames.

keypoint

Synonym for body-part — a tracked point produced by pose estimation.

Kleinberg smoothing

burst detection

A burst-detection algorithm (Kleinberg, 2003) applied to classifier output to merge fragmented detections into coherent bouts and remove noise.

latency

The time from the start of a session (or a trigger) to the first occurrence of a behavior; reported per video in SimBA’s aggregate statistics.

machine results

The per-video CSV files (in project_folder/csv/machine_results) holding the classifier predictions for each frame.

maDLC

Multi-animal DeepLabCut — the multi-animal variant of DeepLabCut.

minimum bout length

The shortest allowed bout duration; shorter detected episodes are removed as noise during post-classification smoothing.

multi-animal tracking

identity

Tracking several animals at once while maintaining each individual’s identity across frames (and recovering it after occlusion), e.g. via maDLC or SLEAP.

Observer

Noldus The Observer XT — commercial event-logging software for manual behavioral annotation; importable into SimBA.

occlusion

When a body-part is hidden (by another animal, an object or self) and so is poorly tracked or missing — often handled by interpolation.

outlier correction

Detecting and correcting implausible body-part coordinates (location- and movement-based) before feature extraction.

oversampling

Re-balancing training data by duplicating (or synthesising) minority-class (behavior-present) frames so the classifier sees them more often. See class imbalance.

p

pose confidence

The probability/likelihood score (0–1) that pose estimation assigns to each tracked body-part, indicating tracking reliability.

path plot

A visualization tracing an animal’s movement trajectory through the arena over time.

pose estimation

Tracking the 2D positions of animal body-parts across video frames, using tools such as DeepLabCut, SLEAP or YOLO.

precision

recall

F1

Standard classification metrics. Precision = fraction of predicted-positive frames that are correct; recall = fraction of true behavior frames detected; F1 = their harmonic mean.

project config

The project_config.ini file at the root of a SimBA project, storing all project settings (paths, body-parts, classifiers, thresholds).

px/mm

pixels per millimeter

The conversion factor between image pixels and real-world millimetres, used to report distances/speeds in physical units. Set per video via a known reference length.

random forest

The default supervised algorithm behind a SimBA classifier: an ensemble of decision trees whose votes give a per-frame behavior probability.

ROI

Region of Interest — a user-defined shape (rectangle, circle or polygon) drawn on the video frame, used to quantify time spent, entries, movement and directionality within specific areas.

sequential analysis

Analysing the order and timing of behaviors — which tend to precede or follow others (see FSTTC) — to uncover behavioral structure.

severity scoring

Grading the intensity of a detected behavior (e.g. attack severity) using movement/feature-based criteria.

SHAP

SHapley Additive exPlanations — a model-interpretability method giving each feature a contribution score, used in SimBA to explain why a classifier made a prediction.

SLEAP

An open-source multi-animal pose estimation framework whose output SimBA can import.

sliding window

rolling window

A fixed-length time window slid across the data to compute time-resolved features (e.g. mean velocity over the last 0.5 s).

smoothing

Reducing frame-to-frame jitter in tracking data (e.g. Savitzky–Golay or Gaussian) to stabilise body-part trajectories.

SMOTE

Synthetic Minority Over-sampling Technique — generates new synthetic minority-class training examples by interpolating between existing ones, rather than plain duplication. One of SimBA’s oversampling options.

Solomon

Solomon Coder — a free manual event-logging / ethogram coding tool; SimBA imports its coded annotations.

spontaneous alternation

A Y- or T-maze assay of spatial working memory, scored as the tendency to visit maze arms in non-repeating sequences. SimBA derives alternation metrics from pose estimation.

SuperAnimal-TopView

A zero-shot, pre-trained top-view mouse pose estimation model (from the DeepLabCut SuperAnimal family) that SimBA can use without user training.

third-party annotation tool

External behavior-annotation software whose frame-by-frame labels SimBA can import and append to extracted features as ground-truth annotation. Supported tools include BORIS, Ethovision, Observer, Solomon, DeepEthogram and BENTO.

time bins

Dividing a session into fixed-duration intervals (e.g. 60 s) to report how metrics change over the course of a recording.

UMAP

Uniform Manifold Approximation and Projection - a dimensionality-reduction method used in SimBA’s unsupervised workflows to embed high-dimensional features into 2-D for clustering and visualization.

undersampling

Re-balancing training data by randomly dropping majority-class (behavior-absent) frames to a target ratio of absent-to-present frames. See class imbalance.

validation

Checking a trained classifier on a held-out or new video — including the one-click “validation video” with the predicted probability overlaid frame-by-frame.

velocity

An animal’s speed of movement (distance per unit time), typically derived from the frame-to-frame displacement of a body-part or centroid, in px/mm-scaled units.

video info

The per-project table (video_info.csv) mapping each video to its FPS, resolution and px/mm.

YOLO

A fast real-time object/keypoint detection model family; SimBA supports YOLO-based detection and pose estimation workflows.