README.md

Constructing Models

This directory contains built-in and user-defined models.

Adding a new model

1. Create a new file (see, e.g., baseline.py).
2. Write your model class inheriting from one of the base Model classes:
   • TFModel: Use a TensorFlow backend.
3. Decorate it with registry.register_model.
4. Import it in __init__.py.

Accessing models

To access a registered (built-in or added using the steps above) model, in a main.py file, for example, use:

import carpedm as dm  # Important: registers models.
from carpedm.util import registry

...  # define hyperparameters, etc.

model = registry.model('registered_model_name')(**hparams)