Create Training Set
The training set for RFM consists of two components: 1) a training set pattern (call template) and 2) at least 100 presence and absence validations (50 presence and 50 absence). Creating a training set allows you to add a Region of Interest (ROI) which is used by the model to create a training set pattern (also called a template). This article covers how to create a training set pattern.
Tip: Creating a training set applies to the Random Forest Model only. If you are interested in running Pattern Matching over your data, please reference Creating a New PM Template instead.
Tip: Random Forest Models run optimally on both Firefox and Chrome.
1. To create a training set folder for one species click on the Visualizer tab in the top navigation bar then scroll down the left panel and click Training Sets. Click the + button to add a new one.
A new training set can alternatively be created under Data on the top navigation bar, Training Sets in the left panel, and then press the + Plus button.
2. Assign a name, select ROI set (default) and your species sound from the drop menu and then hit Create. If you haven't yet added a species, reference this article on how to Add a Species.
3. Choose your newly created training set, then go to the spectrogram and draw a box representing the species call. Crop as closely as possible - use the zoom buttons on the top right hand corner of the visualizer if necessary. We recommend using the most common call of the species (e.g. territorial song) to create the call template. Select a maximum of 3 ROIs of the species call within a training set.
Tip: An ROI is the best example of your focal species call (strong signal with no overlapping calls). Several ROIs can be added in the training set but we recommend using only 1(one) ROI per species. If using several ROIs, please make sure that the selected ROIs belong to the same species and call type and have similar acoustic properties (e.g. shape, bandwidth, duration, maximum and minimum frequency).
4. Click Add Data
Tip: To compare or remove training sets and ROIs, reference this article on the Training Set View
5. Validate the presence/absence of the focal species in a subset of recordings. Open the Visualizer page and find recordings with and without the focal species. Validate the species as present or absent accordingly.
Tip: Having enough validations (at least 100 total) of both presence and absence data are required for creating the random forest model. We recommend validating species in your recordings first, then once you find an ideal species call, add the ROI to your training set.
Next you'll want to read how to validate species in your recordings!