Skip to content

Training

Old documentation

This page is still under construction.

Simple SMP train and test

To train a simple SMP (Segmentation Model Pytorch) model, you first need to preprocess your S2 data:

[rye run] darts preprocess-s2-train-data --your-args-here ...

This will create three data splits:

  • cross-val, used for train and validation
  • val-test 5% random leave-out for testing the randomness distribution shift of the data
  • test leave-out region for testing the spatial distribution shift of the data

Now you can train a model:

[rye run] darts train-smp --your-args-here ...

Change defaults

Even though the defaults from the CLI are somewhat useful, it is recommended to create a config file and change the behavior of the training there.

This will train a model with the cross-val data and save the model to disk. The training relies on PyTorch Lightning, which is a high-level interface for PyTorch. You can now test the model on the other two splits:

[rye run] darts test-smp --your-args-here ...

The checkpoint stored is not usable for the pipeline yet, since it is stored in a different format. To convert the model to a format, you need to convert is first:

[rye run] darts convert-lightning-checkpoint --your-args-here ...