Skip to content

Command Line Utilities

The myco package includes a wide range of command line tools for drawing samples, extracting and merging datasets & training and applying models.

Each component was designed to put as much control in the hands of users to fine-tune how train/test data are generated and how models are trained, while also providing intuitive default values that require minimal user input.

This documentation section will be fleshed out further in the coming weeks. In the meantime, here's the general order of operations required to begin training deep learning models.

Model training workflow

  1. nets draw-samples - create a set of point-location samples based on the spatial extent of a response dataset.
  2. nets create-archive - extract (n_samples, height, width, n_bands) feature/response samples from raster data at each point-location sample.
  3. nets merge-archives - combine a series of archives from multiple feature/response archive datasets into a single training dataset.
  4. nets fit-scaler - fit feature/response scalers to normalize/standardize training data archives.
  5. nets get-config - create a template model training configuration file and edit it to define your desired model architecture.
  6. nets train - train a 2d deep learning model based on feature/response archives and a model configuration file.
  7. nets apply-model - apply the trained model to a stack of feature data to map predictions as a raster.

Other utilities

  • nets apply-scaler - apply feature/response transformations using previously-fit scalers
  • nets evaluate - compute model performance on independent test data
  • nets tunnel - opens ports to allow loading tensorboard or jupyter notebooks running on remote compute instances