Here, we cover the main concepts in AIR.
Ray Data <data>
is the standard way to load and exchange data in Ray AIR. It provides a Dataset <dataset_concept> concept which is used extensively for data loading, preprocessing, and batch inference.
Preprocessors are primitives that can be used to transform input data into features. Preprocessors operate on Datasets <dataset_concept>
, which makes them scalable and compatible with a variety of datasources and dataframe libraries.
A Preprocessor is fitted during Training, and applied at runtime in both Training and Serving on data batches in the same way. AIR comes with a collection of built-in preprocessors, and you can also define your own with simple templates.
See the documentation on Preprocessors <air-preprocessor-ref>
.
doc_code/air_key_concepts.py
Trainers are wrapper classes around third-party training frameworks such as XGBoost and Pytorch. They are built to help integrate with core Ray actors (for distribution), Ray Tune, and Ray Data.
See the documentation on Trainers <air-trainers>
.
doc_code/air_key_concepts.py
Trainer objects produce a Result <air-results-ref>
object after calling .fit()
. These objects contain training metrics as well as checkpoints to retrieve the best model.
doc_code/air_key_concepts.py
Tuners <air-tuner-ref>
offer scalable hyperparameter tuning as part of Ray Tune <tune-main>
.
Tuners can work seamlessly with any Trainer but also can support arbitrary training functions.
doc_code/air_key_concepts.py
The AIR trainers, tuners, and custom pretrained model generate a framework-specific Checkpoint<ray.air.Checkpoint>
object. Checkpoints are a common interface for models that are used across different AIR components and libraries.
There are two main ways to generate a checkpoint.
Checkpoint objects can be retrieved from the Result object returned by a Trainer or Tuner .fit()
call.
doc_code/air_key_concepts.py
You can also generate a checkpoint from a pretrained model. Each AIR supported machine learning (ML) framework has a Checkpoint
object that can be used to generate an AIR checkpoint:
doc_code/air_key_concepts.py
Checkpoints can be used to instantiate a Predictor
, BatchPredictor
, or PredictorDeployment
classes, as seen below.
You can take a checkpoint and do batch inference using the BatchPredictor object.
doc_code/air_key_concepts.py
Deploy the model as an inference service by using Ray Serve and the PredictorDeployment
class.
doc_code/air_key_concepts.py
After deploying the service, you can send requests to it.
doc_code/air_key_concepts.py