aaltay commented on a change in pull request #11075: [BEAM-9421] Website section that describes getting predictions using AI Platform Prediciton URL: https://github.com/apache/beam/pull/11075#discussion_r393365169
########## File path: website/src/documentation/patterns/ai-platform.md ########## @@ -0,0 +1,87 @@ +--- +layout: section +title: "AI Platform integration patterns" +section_menu: section-menu/documentation.html +permalink: /documentation/patterns/ai-platform/ +--- +<!-- +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. +--> + +# AI Platform integration patterns + +This page describes common patterns in pipelines with Google AI Platform transforms. + +<nav class="language-switcher"> + <strong>Adapt for:</strong> + <ul> + <li data-type="language-java">Java SDK</li> + <li data-type="language-py" class="active">Python SDK</li> + </ul> +</nav> + +## Getting predictions + +This section shows how to use a cloud-hosted machine learning model to make predictions about new data using Google Cloud AI Platform Prediction within Beam's pipeline. + +[tfx_bsl](https://github.com/tensorflow/tfx-bsl) is a library that provides `RunInference` Beam's PTransform. `RunInference` is a PTransform able to perform two types of inference. One of them can use a service endpoint. When using a service endpoint, the transform takes a PCollection of type `tf.train.Example` and, for each element, sends a request to Google Cloud AI Platform Prediction service. The transform produces a PCollection of type `PredictLog` which contains predictions. + +Before getting started, deploy a machine learning model to the cloud. The cloud service manages the infrastructure needed to handle prediction requests in both efficient and scalable way. Only Tensorflow models are supported. For more information, see [Exporting a SavedModel for prediction](https://cloud.google.com/ai-platform/prediction/docs/exporting-savedmodel-for-prediction). Review comment: Beam does not have hooks for job start and job finish. One way to emulate those would be using an airflow wrapper to run some setup and cleanup tasks before and after a task. I do not believe we have a recommendation here but an example pattern might be useful (@rezarokni -- in case this will be a good pattern to add to the backlog.) For this version, it is probably fine to assume that the life cycle of the model will be managed by the user separately. For a next phase, we can try to build a graph similar to IOs (like BQ IO) to execute a certain setup and clean task in a single worker. (Although we cannot guarantee that it will not be retried) ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
