Predictor

See also

See this user guide on performing model inference in AIR for usage examples.

Predictor Interface

Constructor Options

predictor.Predictor([preprocessor])

Predictors load models from checkpoints to perform inference.

predictor.Predictor.from_checkpoint(...)	Create a specific predictor from a checkpoint.
predictor.Predictor.from_pandas_udf(pandas_udf)	Create a Predictor from a Pandas UDF.

Predictor Properties

predictor.Predictor.get_preprocessor()	Get the preprocessor to use prior to executing predictions.
predictor.Predictor.set_preprocessor(...)	Set the preprocessor to use prior to executing predictions.

Prediction API

predictor.Predictor.predict(data, **kwargs)

Perform inference on a batch of data.

Supported Data Formats

predictor.Predictor.preferred_batch_format()	Batch format hint for upstream producers to try yielding best block format.
DataBatchType	The central part of internal API.

Batch Predictor

Constructor Options

batch_predictor.BatchPredictor(checkpoint, ...)

Batch predictor class.

batch_predictor.BatchPredictor.from_checkpoint(...)	Create a BatchPredictor from a Checkpoint.
batch_predictor.BatchPredictor.from_pandas_udf(...)	Create a Predictor from a Pandas UDF.

Batch Prediction API

batch_predictor.BatchPredictor.predict(data, *)	Run batch scoring on a Dataset.
batch_predictor.BatchPredictor.predict_pipelined(data, *)	Setup a prediction pipeline for batch scoring.

Built-in Predictors for Library Integrations

XGBoostPredictor(model[, preprocessor])	A predictor for XGBoost models.
LightGBMPredictor(model[, preprocessor])	A predictor for LightGBM models.
TensorflowPredictor(*[, model, ...])	A predictor for TensorFlow models.
TorchPredictor(model[, preprocessor, use_gpu])	A predictor for PyTorch models.
HuggingFacePredictor([pipeline, ...])	A predictor for HuggingFace Transformers PyTorch models.
SklearnPredictor(estimator[, preprocessor])	A predictor for scikit-learn compatible estimators.
RLPredictor(policy[, preprocessor])	A predictor for RLlib policies.