Quick start

Learn how to use OrcaNet in just 4 simple steps! This assumes you have already installed orcanet.

Step 1

Create a directory for your training with an adequate name. E.g.:

mkdir my_first_orcanet_training

This directory will contain all the logfiles, plots, checkpoints etc. of this training.

Step 2

Create your toml configuration files and move them into your directory. These toml files describe:

• The data that will be used (list.toml)

• The configuration of the training (config.toml)

• The architecture of the model (model.toml)

See Toml files for more details on these files. Make sure to give them these exact names (list.toml, config.toml, model.toml). This allows OrcaNet to automatically discover them in your directory.

Step 3

Start the training like this:

orcanet train my_first_orcanet_training

You can monitor how the training goes by looking at the summary plot in my_first_orcanet_training/plots/summary_plot.pdf. Or, if you prefer an interactive plot, with this command:

orcanet summarize my_first_orcanet_training

To resume a training, just run the train command again. Once the model has fully converged, you can continue with the next step.

Step 4

Save the output of the model to hdf5 like this:

orcanet predict my_first_orcanet_training

or like this:

orcanet inference my_first_orcanet_training

Depending on wether you want to use the validation files or the inference files as an input. The result gets saved to my_first_orcanet_training/predictions.

Further Info

If you want to learn how to use the python interface of OrcaNet, you can check out OrcaNet python overview.