Run a Pattern Matching Job

Overview

Video Tutorial

The new RFCx Arbimon Pattern Matching algorithm and visualization tool can be used for species-specific identification. It can replace the previous model/classification approach. Our goal is to continue supporting you in your research, so we are pleased to offer you Pattern Matching for free without any usage limitations. In order to ensure that everyone in the community can benefit from its use, we may need to reevaluate this at a later date if costs become prohibitively high. Therefore, please read and follow the best practices below so that everyone can benefit from this powerful feature, and jobs can run more quickly!

If you'd prefer to watch a more detailed video on how to create a template and run a Pattern Matching job, check out our WildLabs Tech Tutors talk here

Tip: Pattern Matching runs optimally on both Firefox and Chrome.

Important Best Practices

  • For testing templates and parameters (e.g. threshold), use playlists with less than 500 recordings.
  • Do not run a pattern matching job with a large playlist (>500 recordings) until you are confident in the template and parameters.  
    • Example: a project with 100,000 recordings, 20 species.  We expect that the project may include 3-4 pattern matching training jobs (<500 recordings) per species for testing purposes. Then run 1 final pattern matching job per species with a playlist of 100,000 recordings. 
  • We expect that this new feature will greatly assist your project. But, overly redundant analyses with this feature could lead to prohibitively high and unsustainable costs for RFCx 

Get Started

Before you get started please make sure to read the best practices above. This will allow everyone's jobs to run more quickly (including yours!)

To create a pattern matching job you’ll need to first create a template, then you can proceed with creating a pattern matching job. 

1) In the main menu, click Analysis.

2) Select Pattern Matching and then click + button to create a new Pattern Matching job and provide the following information:

  • Name: create a short name without special characters.
  • Template: select a template that has already been created.
  • Select Playlist: the pattern matching algorithm will be applied to all the recordings in the playlist you select. 
  • Matches per recordings: select the number of matches per recording to be displayed.
  • Matches per site: select the number of matches per site to be displayed.
  • Threshold: assign a similarity (i.e. correlation) threshold. Regions of interest (ROIs) with values above the threshold will be displayed as potential positive detections. Tip: Low threshold values (~0.1 – 0.15) will likely include many false positives, but may capture variations in the call type.  In contrast, a high threshold value (~<0.25) will reduce the number of false positives. 
  • Allow citizen scientist validations - to activate this function, please contact us at support@rfcx.org 

3) Click Create to initiate the job. 

4) In the main menu, click Jobs to inspect the job status. Pattern matching jobs will take ~5 minutes to classify ~ 80,000 recordings. Important Note: only begin post-classification validation once the job has completed. 

Post-Classification Validation

Important NoteYou can only begin this step once the pattern matching job has been completed.

1) In the main menu, click Analysis and refresh the page.

2) Search for your recently created Pattern Matching job by clicking on Timestamp to filter by most recent, and click Show Details.

3) To validate ROIs, use the control panel located in the upper right corner of the pattern matching page. Click on the ROI, select the status (present or not present) and then click Submit. Multiple ROIs can be selected. If you validate the pattern matching results as present, the validation of the species will also be saved in the visualizer page.

4) To assist in the validation process, you can listen to the ROI by clicking the play button (highlighted in the yellow square below). To open the original recording, click on the icon highlighted in the orange square below. You will be able to see the validation of the species as present if you validated the template as present. 

5) To accelerate the assessment of presence or absence of a species at a location click in the Search box and use one of the preset queries such as the best matches per site and/or day. 

6) Once the validation process is completed, download the data by clicking on the Download icon. 

Note: If you validate Pattern Matching results as present, it will show as present in the Visualizer. However, if you validate PM results as absent or clear, it won't show up in the visualizer. So technically, the absent and clear validations in the visualizer page are done manually and not using the PM results.

Great job!