Training ML models

pykanto can be used to organise and export data to train any model, and the output (for example, similarity scores) can be imported back into the dataset, which provides a convenient way of keeping every step of the project together.

There is a complete example in the article presenting pykanto that walks you through the steps of training a ‘deep learning’ model to differentiate between different birds based on their songs. The code to run this example, which you can easily adapt and reuse with your own data, is available in its own repository: nilomr/pykanto-example. In particular, scripts 3 and 4 demonstrate how easy it is to generate training and testing sets from the data in a pykanto dataset, and how to fine-tune a fairly powerful model using PyTorch and Pytorch-Lightning.