[GitHub] [beam] rszper commented on a diff in pull request #22250: Update RunInference documentation

[GitBox <gi...@apache.org>]

rszper commented on code in PR #22250:
URL: https://github.com/apache/beam/pull/22250#discussion_r922546806


##########
website/www/site/content/en/documentation/sdks/python-machine-learning.md:
##########
@@ -0,0 +1,201 @@
+---
+type: languages
+title: "Apache Beam Python Machine Learning"
+---
+<!--
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+-->
+
+# Machine Learning
+
+You can use Apache Beam with the RunInference API to use machine learning (ML) models to do local and remote inference with batch and streaming pipelines. Starting with Apache Beam 2.40.0, PyTorch and Scikit-learn frameworks are supported. You can create multiple types of transforms using the RunInference API: the API takes multiple types of setup parameters from model handlers, and the parameter type determines the model implementation.
+
+## Why use the RunInference API?
+
+RunInference takes advantage of existing Apache Beam concepts, such as the the `BatchElements` transform and the `Shared` class, to enable you to use models in your pipelines to create transforms optimized for machine learning inferences. The ability to create arbitrarily complex workflow graphs also allows you to build multi-model pipelines.
+
+### BatchElements PTransform
+
+To take advantage of the optimizations of vectorized inference that many models implement, we added the `BatchElements` transform as an intermediate step before making the prediction for the model. This transform batches elements together. The batched elements are then applied with a transformation for the particular framework of RunInference. For example, for numpy `ndarrays`, we call `numpy.stack()`,  and for torch `Tensor` elements, we call `torch.stack()`.
+
+To customize the settings for `beam.BatchElements`, in `ModelHandler`, override the `batch_elements_kwargs` function. For example, use `min_batch_size` to set the lowest number of elements per batch or `max_batch_size` to set the highest number of elements per batch.
+
+For more information, see the [`BatchElements` transform documentation](https://beam.apache.org/releases/pydoc/current/apache_beam.transforms.util.html#apache_beam.transforms.util.BatchElements).
+
+### Shared helper class
+
+Instead of loading a model for each thread in the process, we use the `Shared` class, which allows us to load one model that is shared across all threads of each worker in a DoFn. For more information, see the

Review Comment:
   Updated



##########
website/www/site/content/en/documentation/sdks/python-machine-learning.md:
##########
@@ -0,0 +1,201 @@
+---
+type: languages
+title: "Apache Beam Python Machine Learning"
+---
+<!--
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+-->
+
+# Machine Learning
+
+You can use Apache Beam with the RunInference API to use machine learning (ML) models to do local and remote inference with batch and streaming pipelines. Starting with Apache Beam 2.40.0, PyTorch and Scikit-learn frameworks are supported. You can create multiple types of transforms using the RunInference API: the API takes multiple types of setup parameters from model handlers, and the parameter type determines the model implementation.
+
+## Why use the RunInference API?
+
+RunInference takes advantage of existing Apache Beam concepts, such as the the `BatchElements` transform and the `Shared` class, to enable you to use models in your pipelines to create transforms optimized for machine learning inferences. The ability to create arbitrarily complex workflow graphs also allows you to build multi-model pipelines.
+
+### BatchElements PTransform
+
+To take advantage of the optimizations of vectorized inference that many models implement, we added the `BatchElements` transform as an intermediate step before making the prediction for the model. This transform batches elements together. The batched elements are then applied with a transformation for the particular framework of RunInference. For example, for numpy `ndarrays`, we call `numpy.stack()`,  and for torch `Tensor` elements, we call `torch.stack()`.
+
+To customize the settings for `beam.BatchElements`, in `ModelHandler`, override the `batch_elements_kwargs` function. For example, use `min_batch_size` to set the lowest number of elements per batch or `max_batch_size` to set the highest number of elements per batch.
+
+For more information, see the [`BatchElements` transform documentation](https://beam.apache.org/releases/pydoc/current/apache_beam.transforms.util.html#apache_beam.transforms.util.BatchElements).
+
+### Shared helper class
+
+Instead of loading a model for each thread in the process, we use the `Shared` class, which allows us to load one model that is shared across all threads of each worker in a DoFn. For more information, see the
+[`Shared` class documentation](https://github.com/apache/beam/blob/master/sdks/python/apache_beam/utils/shared.py#L20).
+
+### Multi-model pipelines
+
+The RunInference API can be composed into multi-model pipelines. Multi-model pipelines can be useful for A/B testing or for building out ensembles that are comprised of models that perform tokenization, sentence segmentation, part-of-speech tagging, named entity extraction, language detection, coreference resolution, and more.
+
+## Modify a pipeline to use an ML model
+
+To use the RunInference transform, add the following code to your pipeline:
+
+```
+from apache_beam.ml.inference.base import RunInference
+with pipeline as p:
+   predictions = ( p |  'Read' >> beam.ReadFromSource('a_source')
+                     | 'RunInference' >> RunInference(<model_handler>)
+```
+Where `model_handler` is the model handler setup code.
+
+To import models, you need to wrap them around a `ModelHandler` object. Which `ModelHandler` you import depends on the framework and type of data structure that contains the inputs. The following examples show some ModelHandlers that you might want to import.

Review Comment:
   Updated



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: github-unsubscribe@beam.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org