sgoldenlab/simba

SimBA (Simple Behavioral Analysis), a pipeline and GUI for developing supervised behavioral classifiers

What it solves

SimBA (Simple Behavioral Analysis) helps researchers automatically identify and measure animal behaviors—like social interactions, movement, and time spent in certain areas—from video recordings. Instead of watching hours of footage and manually scoring behavior by hand, researchers can train a machine-learning classifier that labels each video frame with the behavior happening at that moment. It is built for behavioral neuroscience and works with pose-estimation data (the tracked positions of body parts) produced by tools like DeepLabCut, SLEAP, and others.

How it works

SimBA takes pose-estimation data—the x/y coordinates of an animal's body parts over time—as its input. Users annotate a subset of frames to label which behavior is occurring (e.g., "attack," "grooming," "freezing"). SimBA then computes a large set of features from the pose data (distances, angles, velocities, etc.) and trains a supervised machine-learning classifier, typically a random forest, to predict behavior from those features. The toolkit includes a graphical user interface (GUI) so no programming is required. It also provides evaluation metrics (precision/recall, learning curves, permutation importance), SHAP-based explainability to understand why the model makes decisions, and a wide range of downstream analyses: bout counts, durations, velocity, distance traveled, time-in-zone, directionality (which animal is oriented toward which), behavior-sequence coupling, and unsupervised clustering. Results can be exported to CSV and visualized as plots or overlays burned onto the original videos.

Who it's for

Behavioral neuroscientists and animal researchers who want to quantify animal behavior from video without writing code. It is designed for users with no programming background, works on Windows, macOS, and Linux, and is validated in mice and rats. It is also useful for researchers who already use pose-estimation tools like DeepLabCut or SLEAP and want to turn that tracking data into behavioral classifications.

Highlights

  • No programming required — a full GUI workflow from importing pose data to training and evaluating classifiers.
  • Broad input compatibility — accepts pose-estimation from DeepLabCut, SLEAP, DeepPoseKit, DANNCE (3D), MARS, FaceMap, APT, SuperAnimal-TopView, YOLO, and blob tracking, plus user-defined pose schemes.
  • Explainable machine learning — SHAP-based tools show which features drive each classification, plus standard evaluation metrics like precision/recall and learning curves.
  • Rich behavioral analytics — bout counts, durations, latency, velocity, distance, time-in-zone, directionality, behavior-sequence coupling, and unsupervised clustering.
  • Turn-key assays — built-in support for spontaneous alternation, pup retrieval, light/dark box, freezing, circling, and more.
  • Visualization suite — tracking overlays, path plots, heatmaps, Gantt charts, classifier probability plots, and multi-panel videos.
  • Validated and open — classifier libraries validated in mice and rats, with all data, models, and annotations available on OSF, and a Nature Neuroscience publication describing the platform.

Related

  • Project
    DeepLabCut/DeepLabCut

    DeepLabCut is a markerless pose estimation toolbox that allows researchers to track animals and objects in video without physical markers using deep learning.

  • Project
    talmolab/sleap

    A deep-learning framework for multi-animal pose tracking that uses a human-in-the-loop GUI to rapidly label and quantify animal behavior.

  • Project
    SwanHubX/SwanLab

    An AI training analysis and metric observation platform that provides training visualization, automatic logging, and experiment comparison for model training teams.

  • Dispatch
    SetFitABSA: Few-Shot Aspect Based Sentiment Analysis

    Hugging Face and Intel Labs introduced SetFitABSA, a prompt-less, few-shot framework for Aspect-Based Sentiment Analysis that outperforms larger generative models like Llama 2 and T5 in low-data scenarios.

  • Dispatch
    Hugging Face Guide: Training a LLaMA 2 Chatbot Without Code

    Hugging Face provides a no-code workflow using Spaces, AutoTrain, and ChatUI to allow non-engineers to fine-tune LLaMA 2 and deploy it as a functional chat application.