Training

Run the following command-line train to train a model:

sleepless train -vv <dataset> <model> -o <path/to/output_folder>

Replace <dataset> and <model> by the appropriate configuration files.

Pytorch Model

To train a pytorch model, use the command-line interface (CLI) application sleepless train, available on your prompt. To use this CLI, you must define the input dataset that will be used to train, as well as the type of model that will be trained and its parameters (See examples in Preset Configurations). An output path must be also defined, where the model weights and the training data will be stored. You may issue sleepless train --help for a help message containing more detailed instructions.

Tip

We strongly advice training with a GPU (using “cuda:0”). Depending on the available GPU memory you might have to adjust your batch size.

Examples

To train Chambon CNN on the sc-edf dataset:

sleepless train -vv sc_filtered chambon -o <path/to/output_folder>

Scikit-learn Model or Grid-search

To train a scikit-learn model or grid-search use the command-line interface (CLI) application sleepless train, available on your prompt. To use this CLI, you must define the input dataset that will be used to train the model and its parameters, as well as the type of model that will be trained. An output path must be also defined, where the model and the training data will be stored.

You may issue sleepless train --help for a help message containing more detailed instructions.

Examples

To train a grid-search for a random forest model on the sc-edf dataset:

sleepless train -vv sc_filtered rf-gs-mne -o <path/to/output_folder>