[Train] Hard deprecate BatchPredictor (ray-project#38209)

Upgrades the deprecation of BatchPredictor from warning to now raising a DeprecationWarning exception. Closes ray-project#37035 --------- Signed-off-by: amogkam <[email protected]>
charleslai2000 · Aug 9, 2023 · 0416259 · 0416259
1 parent b2dd418
commit 0416259
Show file tree

Hide file tree

Showing 36 changed files with 81 additions and 2,386 deletions.
diff --git a/doc/BUILD b/doc/BUILD
@@ -224,7 +224,6 @@ py_test_run_all_subdirectory(
  size = "large",
  include = ["source/train/doc_code/*.py"],
  exclude = [
- "source/train/doc_code/predictors.py", # Too large
  "source/train/doc_code/hf_trainer.py", # Too large
  ],
  extra_srcs = [],
@@ -261,14 +260,6 @@ py_test_run_all_subdirectory(
 # Run GPU tests
 # --------------
 
-py_test_run_all_subdirectory(
- size = "large",
- include = ["source/train/doc_code/predictors.py"],
- exclude = [],
- extra_srcs = [],
- tags = ["exclusive", "team:ml", "ray_air", "gpu"],
-)
-
 py_test(
  name = "pytorch_resnet_finetune",
  size = "large",

diff --git a/doc/source/data/batch_inference.rst b/doc/source/data/batch_inference.rst
@@ -506,76 +506,4 @@ The rest of the logic looks the same as in the `Quickstart <#quickstart>`_.
 .. testoutput::
  :options: +MOCK
 
- {'predictions': 0.9969483017921448}
-
-Benchmarks
-----------
-
-Below we document key performance benchmarks for common batch prediction workloads.
-
-
-XGBoost Batch Prediction
-~~~~~~~~~~~~~~~~~~~~~~~~
-
-This task uses the BatchPredictor module to process different amounts of data
-using an XGBoost model.
-
-We test out the performance across different cluster sizes and data sizes.
-
-- `XGBoost Prediction Script`_
-- `XGBoost Cluster Configuration`_
-
-.. TODO: Add script for generating data and running the benchmark.
-
-.. list-table::
-
- * - **Cluster Setup**
- - **Data Size**
- - **Performance**
- - **Command**
- * - 1 m5.4xlarge node (1 actor)
- - 10 GB (26M rows)
- - 275 s (94.5k rows/s)
- - `python xgboost_benchmark.py --size 10GB`
- * - 10 m5.4xlarge nodes (10 actors)
- - 100 GB (260M rows)
- - 331 s (786k rows/s)
- - `python xgboost_benchmark.py --size 100GB`
-
-
-GPU image batch prediction
-~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-This task uses the BatchPredictor module to process different amounts of data
-using a Pytorch pre-trained ResNet model.
-
-We test out the performance across different cluster sizes and data sizes.
-
-- `GPU image batch prediction script`_
-- `GPU prediction small cluster configuration`_
-- `GPU prediction large cluster configuration`_
-
-.. list-table::
-
- * - **Cluster Setup**
- - **Data Size**
- - **Performance**
- - **Command**
- * - 1 g4dn.8xlarge node
- - 1 GB (1623 images)
- - 46.12 s (35.19 images/sec)
- - `python gpu_batch_inference.py --data-directory=1G-image-data-synthetic-raw --data-format=raw`
- * - 1 g4dn.8xlarge node
- - 20 GB (32460 images)
- - 285.2 s (113.81 images/sec)
- - `python gpu_batch_inference.py --data-directory=20G-image-data-synthetic-raw --data-format=raw`
- * - 4 g4dn.12xlarge nodes
- - 100 GB (162300 images)
- - 304.01 s (533.86 images/sec)
- - `python gpu_batch_inference.py --data-directory=100G-image-data-synthetic-raw --data-format=raw`
-
-.. _`XGBoost Prediction Script`: https://github.com/ray-project/ray/blob/a241e6a0f5a630d6ed5b84cce30c51963834d15b/release/air_tests/air_benchmarks/workloads/xgboost_benchmark.py#L63-L71
-.. _`XGBoost Cluster Configuration`: https://github.com/ray-project/ray/blob/a241e6a0f5a630d6ed5b84cce30c51963834d15b/release/air_tests/air_benchmarks/xgboost_compute_tpl.yaml#L6-L24
-.. _`GPU image batch prediction script`: https://github.com/ray-project/ray/blob/master/release/air_tests/air_benchmarks/workloads/gpu_batch_inference.py#L18-L49
-.. _`GPU prediction small cluster configuration`: https://github.com/ray-project/ray/blob/master/release/air_tests/air_benchmarks/compute_gpu_1_cpu_16_aws.yaml#L6-L15
-.. _`GPU prediction large cluster configuration`: https://github.com/ray-project/ray/blob/master/release/air_tests/air_benchmarks/compute_gpu_4x4_aws.yaml#L6-L15
+ {'predictions': 0.9969483017921448}
diff --git a/doc/source/data/doc_code/preprocessors.py b/doc/source/data/doc_code/preprocessors.py
@@ -84,13 +84,25 @@
 
 
 # __predictor_start__
-from ray.train.batch_predictor import BatchPredictor
 from ray.train.xgboost import XGBoostPredictor
 
 test_dataset = ray.data.from_items([{"x": x} for x in range(2, 32, 3)])
 
-batch_predictor = BatchPredictor.from_checkpoint(checkpoint, XGBoostPredictor)
-predicted_probabilities = batch_predictor.predict(test_dataset)
+
+class XGBoostPredictorWrapper:
+ def __init__(self, checkpoint):
+ self.predictor = XGBoostPredictor.from_checkpoint(checkpoint)
+
+ def __call__(self, batch):
+ return self.predictor.predict(batch)
+
+
+predicted_probabilities = test_dataset.map_batches(
+ XGBoostPredictorWrapper,
+ compute=ray.data.ActorPoolStrategy(size=2),
+ fn_constructor_kwargs={"checkpoint": checkpoint},
+ batch_format="pandas",
+)
 predicted_probabilities.show()
 # {'predictions': 0.09843720495700836}
 # {'predictions': 5.604666709899902}

diff --git a/doc/source/data/preprocessors.rst b/doc/source/data/preprocessors.rst
@@ -119,7 +119,7 @@ Predictor
 A ``Predictor`` can be constructed from a saved ``Checkpoint``. If the ``Checkpoint`` contains a ``Preprocessor``,
 then the ``Preprocessor`` calls ``transform_batch`` on input batches prior to performing inference.
 
-In the following example, we show the Batch Predictor flow.
+In the following example, we show the batch inference flow.
 
 .. literalinclude:: doc_code/preprocessors.py
  :language: python

diff --git a/doc/source/ray-air/api/predictor.rst b/doc/source/ray-air/api/predictor.rst
@@ -48,33 +48,6 @@ Supported Data Formats
  predictor.Predictor.preferred_batch_format
  ~predictor.DataBatchType
 
-
-Batch Predictor
----------------
-
-Constructor Options
-~~~~~~~~~~~~~~~~~~~
-
-.. autosummary::
- :toctree: doc/
-
- batch_predictor.BatchPredictor
-
-.. autosummary::
- :toctree: doc/
-
- batch_predictor.BatchPredictor.from_checkpoint
- batch_predictor.BatchPredictor.from_pandas_udf
-
-Batch Prediction API
-~~~~~~~~~~~~~~~~~~~~
-
-.. autosummary::
- :toctree: doc/
-
- batch_predictor.BatchPredictor.predict
- batch_predictor.BatchPredictor.predict_pipelined
-
 .. _air_framework_predictors:
 
 Built-in Predictors for Library Integrations

diff --git a/doc/source/ray-air/computer-vision.rst b/doc/source/ray-air/computer-vision.rst
@@ -10,8 +10,6 @@ This guide explains how to perform common computer vision tasks like:
 * `Reading image data`_
 * `Transforming images`_
 * `Training vision models`_
-* `Batch predicting images`_
-* `Serving vision models`_
 
 Reading image data
 ------------------
@@ -202,130 +200,4 @@ Training vision models
  :end-before: __tensorflow_trainer_stop__
  :dedent:
 
- For more information, check out :ref:`the Ray Train documentation <train-docs>`.
-
-Creating checkpoints
---------------------
-
-:class:`Checkpoints <ray.train.Checkpoint>` are required for batch inference and model
-serving. They contain model state and optionally a preprocessor.
-
-If you're going from training to prediction, don't create a new checkpoint.
-:meth:`Trainer.fit() <ray.train.trainer.BaseTrainer.fit>` returns a
-:class:`~ray.train.Result` object. Use
-:attr:`Result.checkpoint <ray.train.Result.checkpoint>` instead.
-
-.. tab-set::
-
- .. tab-item:: Torch
-
- To create a :class:`~ray.train.torch.TorchCheckpoint`, pass a Torch model and
- the :class:`~ray.data.preprocessor.Preprocessor` you created in `Transforming images`_
- to :meth:`TorchCheckpoint.from_model() <ray.train.torch.TorchCheckpoint.from_model>`.
-
- .. literalinclude:: ./doc_code/computer_vision.py
- :start-after: __torch_checkpoint_start__
- :end-before: __torch_checkpoint_stop__
- :dedent:
-
- .. tab-item:: TensorFlow
-
- To create a :class:`~ray.train.tensorflow.TensorflowCheckpoint`, pass a TensorFlow model and
- the :class:`~ray.data.preprocessor.Preprocessor` you created in `Transforming images`_
- to :meth:`TensorflowCheckpoint.from_model() <ray.train.tensorflow.TensorflowCheckpoint.from_model>`.
-
- .. literalinclude:: ./doc_code/computer_vision.py
- :start-after: __tensorflow_checkpoint_start__
- :end-before: __tensorflow_checkpoint_stop__
- :dedent:
-
-
-Batch predicting images
------------------------
-
-:class:`~ray.train.batch_predictor.BatchPredictor` lets you perform inference on large
-image datasets.
-
-.. tab-set::
-
- .. tab-item:: Torch
-
- To create a :class:`~ray.train.batch_predictor.BatchPredictor`, call
- :meth:`BatchPredictor.from_checkpoint <ray.train.batch_predictor.BatchPredictor.from_checkpoint>` and pass the checkpoint
- you created in `Creating checkpoints`_.
-
- .. literalinclude:: ./doc_code/computer_vision.py
- :start-after: __torch_batch_predictor_start__
- :end-before: __torch_batch_predictor_stop__
- :dedent:
-
- .. tab-item:: TensorFlow
-
- To create a :class:`~ray.train.batch_predictor.BatchPredictor`, call
- :meth:`BatchPredictor.from_checkpoint <ray.train.batch_predictor.BatchPredictor.from_checkpoint>` and pass the checkpoint
- you created in `Creating checkpoints`_.
-
- .. literalinclude:: ./doc_code/computer_vision.py
- :start-after: __tensorflow_batch_predictor_start__
- :end-before: __tensorflow_batch_predictor_stop__
- :dedent:
-
-Serving vision models
----------------------
-
-:class:`~ray.serve.Deployment` lets you
-deploy a model to an endpoint and make predictions over the Internet.
-
-Deployments use :ref:`HTTP adapters <serve-http>` to define how HTTP messages are converted to model
-inputs. For example, :func:`~ray.serve.http_adapters.json_to_ndarray` converts HTTP messages like this:
-
-.. code-block::
-
- {"array": [[1, 2], [3, 4]]}
-
-To NumPy ndarrays like this:
-
-.. code-block::
-
- array([[1., 2.],
- [3., 4.]])
-
-.. tab-set::
-
- .. tab-item:: Torch
-
- To deploy a Torch model to an endpoint, create a predictor from the checkpoint you created in `Creating checkpoints`_
- and serve via a Ray Serve deployment.
-
- .. literalinclude:: ./doc_code/computer_vision.py
- :start-after: __torch_serve_start__
- :end-before: __torch_serve_stop__
- :dedent:
-
- Then, make a request to classify an image.
-
- .. literalinclude:: ./doc_code/computer_vision.py
- :start-after: __torch_online_predict_start__
- :end-before: __torch_online_predict_stop__
- :dedent:
-
- For more in-depth examples, read about :ref:`Ray Serve <serve-getting-started>`.
-
- .. tab-item:: TensorFlow
-
- To deploy a TensorFlow model to an endpoint, use the checkpoint you created in `Creating checkpoints`_
- to create a Ray Serve deployment serving the model.
-
- .. literalinclude:: ./doc_code/computer_vision.py
- :start-after: __tensorflow_serve_start__
- :end-before: __tensorflow_serve_stop__
- :dedent:
-
- Then, make a request to classify an image.
-
- .. literalinclude:: ./doc_code/computer_vision.py
- :start-after: __tensorflow_online_predict_start__
- :end-before: __tensorflow_online_predict_stop__
- :dedent:
-
- For more information, see :ref:`Ray Serve <serve-getting-started>`.
+ For more information, check out :ref:`the Ray Train documentation <train-docs>`.
diff --git a/doc/source/ray-air/doc_code/computer_vision.py b/doc/source/ray-air/doc_code/computer_vision.py
@@ -23,13 +23,11 @@ def test(*, framework: str, datasource: str):
  preprocessor, per_epoch_preprocessor = create_torch_preprocessors()
  train_torch_model(dataset, preprocessor, per_epoch_preprocessor)
  checkpoint = create_torch_checkpoint(preprocessor)
- batch_predict_torch(dataset, checkpoint)
  online_predict_torch(checkpoint)
  if framework == "tensorflow":
  preprocessor, per_epoch_preprocessor = create_tensorflow_preprocessors()
  train_tensorflow_model(dataset, preprocessor, per_epoch_preprocessor)
  checkpoint = create_tensorflow_checkpoint(preprocessor)
- batch_predict_tensorflow(dataset, checkpoint)
  online_predict_tensorflow(checkpoint)
 
 
@@ -327,32 +325,6 @@ def create_tensorflow_checkpoint(preprocessor):
  return checkpoint
 
 
-def batch_predict_torch(dataset, checkpoint):
- # __torch_batch_predictor_start__
- from ray.train.batch_predictor import BatchPredictor
- from ray.train.torch import TorchPredictor
-
- predictor = BatchPredictor.from_checkpoint(checkpoint, TorchPredictor)
- predictor.predict(dataset, feature_columns=["image"], keep_columns=["label"])
- # __torch_batch_predictor_stop__
-
-
-def batch_predict_tensorflow(dataset, checkpoint):
- # __tensorflow_batch_predictor_start__
- import tensorflow as tf
-
- from ray.train.batch_predictor import BatchPredictor
- from ray.train.tensorflow import TensorflowPredictor
-
- predictor = BatchPredictor.from_checkpoint(
- checkpoint,
- TensorflowPredictor,
- model_definition=tf.keras.applications.resnet50.ResNet50,
- )
- predictor.predict(dataset, feature_columns=["image"], keep_columns=["label"])
- # __tensorflow_batch_predictor_stop__
-
-
 def online_predict_torch(checkpoint):
  # __torch_serve_start__
  from io import BytesIO