moover helps animal scientists turn movement data into behaviour outputs.
It is designed for people who want to work locally on their own computer, without needing a high-performance computing (HPC) environment. Larger runs can still take time, but the goal is to make behaviour modelling practical and approachable for beginners.
In moover, the usual pattern is:
- raw accelerometer files can live wherever they already live, including a network drive or portable drive
- the workspace keeps the local derived outputs for each run, such as previews, epoch features, optimisation results, model bundles, and test vectors
- epoch-level data is expected to be held locally so runs stay reproducible and easier to revisit
Who moover is for
moover is for animal scientists, including nutritionists and other researchers who may be new to accelerometer pipelines, machine learning, or R package workflows.
Installation
Install the development version from GitHub with:
install.packages("devtools")
devtools::install_github("wobblytwilliams/moover")Start here
If you are new to moover, start with the first tutorial:
Start Here: Your First Successful Run
That walkthrough uses packaged example data so you can get one complete success before working with your own files.
Typical beginner journey
- Set up a workspace on your computer.
- Try the packaged example data.
- Prepare your own accelerometer files.
- Record or tidy your observations for training.
- Build a model.
- Predict behaviour on new data.
Advanced users
If you already know the workflow and want reproducible reruns, saved instructions, or command-line execution, see the advanced tutorial on scripted reruns and saved specs:
Documentation
The full tutorial and reference site is here:
https://wobblytwilliams.github.io/moover/
The beginner path is organised as a 10-chapter walkthrough and walks through:
- your first successful run
- setting up folders and understanding the workspace
- preparing accelerometer files
- recording observations for training
- building a first model
- predicting with an existing model
- understanding the results and export folder
Included example data
The package ships with a small example workspace under:
system.file("extdata", "example_workspace", package = "moover")What moover does
- ingests raw accelerometer data
- converts it into a standard 5-column movement format
- generates features from fixed time blocks
- trains and optimises Random Forest models
- evaluates models with beginner-friendly summaries and metrics
- exports reusable model bundles
- predicts behaviour on new datasets
Current focus for v1
- beginner-friendly local workflows
- Random Forest models
- reusable and shareable model bundles
- support for CQU-style and generic delimited input files
- support for external raw-data folders with local run outputs
- fixed-size chunked reading for larger raw files when needed
Storage note
Version 1 keeps the storage choices simple on purpose:
- raw files can be read directly from outside the workspace
- local run outputs are written as CSV.gz and RDS files
- Parquet is not required for the beginner workflow
We may add optional Parquet or Arrow-based local caching later if intermediate-file read and write time becomes a real bottleneck, but it is not needed to get started.