Model Definitions¶

The model definition is the interface between Determined and the user’s application framework (e.g., Keras, TensorFlow), in terms of loading training data, describing a model architecture, and specifying the underlying iterative optimization training algorithms.

File Structure and Entrypoint¶

A model definition consists of a directory of Python files and an accompanying entrypoint which specifies from where to load a trial class. The entrypoint specification is expected to take the form:

<module>:<object reference>

<module> specifies the module containing the trial class within the model definition, relative to the root. It may be an empty string if the model definition is a Python package and the trial class is exposed in the top-level __init__.py file.

<object reference> specifies the naming of the trial class within the module. It may be a nested object delimited by dots.

Examples:

1) :MNistTrial: expects an MNistTrial class that is exposed in a __init__.py file at the top-level of the model definition.

2) model_def:CIFAR10Trial: expects a CIFAR10Trial class that is defined in a file model_def.py at the top-level of the model definition.

3) determined_lib.trial:trial_classes.NestedTrial: expects a NestedTrial class that is an attribute of trial_classes, where trial_classes is defined in a file determined_lib/trial.py

Note that this follows the Entry points specification defined in the Python Packaging User Guide with a single difference: the directory name of the model definition is prefixed to <module>, or used as the module if <module> is empty.

Since project directories might include large artifacts that should not be packaged as part of the model definition (e.g., data sets or compiled binaries), users can optionally include a .detignore file at the top-level that specifies file paths to be omitted from the model definition. The .detignore file uses the same syntax as .gitignore. Note that byte-compiled Python files (e.g., .pyc files or __pycache__ directories) are always ignored.

For backward compatibility, Determined also supports model definitions that consist of a single Python file. In this case, the <module> should be an empty string and <object reference> should be the name of the trial class defined in the Python file.

Model Definition Interfaces¶

To create a model definition, we should implement the Trial interface provided by Determined. This interface returns information about the machine learning task the user wants to perform, like the model architecture to use or the validation metrics that should be computed.

Determined provides versions of the Trial interface for each of the application frameworks it supports. Specifically, Determined currently supports five types of Trial interfaces encompassing three application frameworks:

Best Practices¶

To learn more about some best practices when writing model definitions, see Best Practices for Model Definitions.