https://github.com/denajgibbon/torch-for-r-gibbons

https://github.com/denajgibbon/torch-for-r-gibbons

Repository

README.Rmd

---
output: github_document
---



```{r, include = FALSE}
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
```

# torch-for-R-gibbons





This repository provides a comprehensive pipeline for training, evaluating, and deploying convolutional neural networks (CNNs) for gibbon call detection using the `torch` and `gibbonNetR` packages in R. It includes data preprocessing, model training, evaluation, and deployment across a range of configurations and test conditions.

## Getting Started

### Prerequisites
#### If you don't have devtools installed
install.packages("devtools")

#### Install gibbonNetR
devtools::install_github("https://github.com/DenaJGibbon/gibbonNetR")

For detailed usage instructions and examples, refer to the gibbonNetR documentation (https://denajgibbon.github.io/gibbonNetR/).

#### Install BirdNET
Follow the detailed instructions here: https://github.com/birdnet-team/BirdNET-Analyzer. For this paper we used BirdNET v2.4.

Current citation: 
Clink, Dena J., et al. "Automated detection of gibbon calls from passive acoustic monitoring data using convolutional neural networks in the" torch for R" ecosystem." arXiv preprint arXiv:2407.09976 (2024). 
https://doi.org/10.48550/arXiv.2407.09976

### Data availability
You can download the data and results files from Zenodo:
https://zenodo.org/records/10948975.

To run the scripts you can open the folder in your exisiting R project directory, or use 'setwd'.

## Repository Structure

### R Scripts Overview

- **Part 1a. Variability Benchmarking Results.R**  
  Processes model performance metrics across different configurations to assess variability.

- **Part 1b. Evaluate Variability Benchmarking Results.R**  
  Analyzes and visualizes variability in model performance.

- **Part 2a. Train CNNs Over Multiple Epochs.R**  
  Trains CNN models over multiple epochs for performance benchmarking.

- **Part 2b. Train CNNs Over Multiple Epochs.R**  
  Continuation or alternative run of multi-epoch training using different parameters or models.

- **Part 3a. Data Augmentation.R**  
  Performs data augmentation on spectrograms to enrich the training dataset.

- **Part 3b. Evaluation Data Augmentation.R**  
  Evaluates how data augmentation impacts model performance.

- **Part 3c. Evaluation Data Augmentation on Different Test Set.R**  
  Tests the augmented models on a novel test set to assess generalization.

- **Part 4a. MultiSpecies Gibbon Comparison BirdNET CLI**  
  Runs BirdNET CLI to generate multispecies comparison results for Crested and Grey Gibbons.

- **Part 4b. Comparison with BirdNET.R**  
  Compares internal CNN model outputs with BirdNET predictions.

- **Part 5. Final Model Performance.R**  
  Extracts and summarizes final model performance statistics including F1, AUC, and threshold selection.

- **Part 6. Deploy Model Over PAM Data.R**  
  Applies the trained models to passive acoustic monitoring (PAM) datasets.

- **Part 7. Call Density Plots.R**  
  Generates visualizations of call densities across space.

### Plots and tables
[Click here to view the figures and plots](https://github.com/DenaJGibbon/torch-for-R-gibbons/blob/main/R/Recreate-plots-for-manuscript.md)

Owner

GitHub Events