The Trainer defines the model and training loop within the TrainingPipeline.

Standard trainers

A standard FeedForwardTrainer step is provided in the source code, which defines a simple feed-forward neural network in Tensorflow.

Create custom trainers

In the case of the Trainer, the built-in methods are just convenience to access popular model types. Most of the times, custom model code is required. This is how to create custom trainer steps:

Base Trainer

ZenML comes equipped with a BaseTrainer that all trainers should inherit from. This is how the interface looks like:

    def run_fn(self):

    def input_fn(self,
                 file_pattern: List[Text],
                 tf_transform_output: tft.TFTransformOutput):

    def model_fn(train_dataset:,

This doc section is incomplete. Please refer to the docstrings in source code while you wait to complete this section.

Tensorflow-based Trainers

PyTorch-based Trainers

Coming soon.

Other libraries

Coming soon.

If you need the above functionalities earlier, then ping us on our Slack or create an issue on GitHub so that we know about it!