[ https://issues.apache.org/jira/browse/BEAM-9421?focusedWorklogId=431701&page=com.atlassian.jira.plugin.system.issuetabpanels:worklog-tabpanel#worklog-431701 ]
ASF GitHub Bot logged work on BEAM-9421: ---------------------------------------- Author: ASF GitHub Bot Created on: 07/May/20 11:53 Start Date: 07/May/20 11:53 Worklog Time Spent: 10m Work Description: kamilwu commented on a change in pull request #11075: URL: https://github.com/apache/beam/pull/11075#discussion_r421445655 ########## File path: website/src/documentation/patterns/ai-platform.md ########## @@ -0,0 +1,90 @@ +--- +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 Cloud 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 [Google Cloud AI Platform Prediction](https://cloud.google.com/ai-platform/prediction/docs/overview) to make predictions about new data from a cloud-hosted machine learning model. + +[tfx_bsl](https://github.com/tensorflow/tfx-bsl) is a library with a Beam PTransform called `RunInference`. `RunInference` is able to perform an inference that can use an external service endpoint for receiving data. When using a service endpoint, the transform takes a PCollection of type `tf.train.Example` and, for every batch of elements, sends a request to AI Platform Prediction. The size of a batch may vary. For more details on how Beam finds the best batch size, refer to a docstring for [BatchElements](https://beam.apache.org/releases/pydoc/current/apache_beam.transforms.util.html?highlight=batchelements#apache_beam.transforms.util.BatchElements). + + The transform produces a PCollection of type `PredictLog`, which contains predictions. + +Before getting started, deploy a TensorFlow model to AI Platform Prediction. The cloud service manages the infrastructure needed to handle prediction requests in both efficient and scalable way. Do note that only TensorFlow models are supported by the transform. For more information, see [Exporting a SavedModel for prediction](https://cloud.google.com/ai-platform/prediction/docs/exporting-savedmodel-for-prediction). + +Once a machine learning model is deployed, prepare a list of instances to get predictions for. To send binary data, make sure that the name of an input ends in `_bytes`. This will base64-encode data before sending a request. + +### Example +Here is an example of a pipeline that reads input instances from the file, converts JSON objects to `tf.train.Example` objects and sends data to AI Platform Prediction. The content of a file can look like this: + +``` +{"input": "the quick brown"} +{"input": "la bruja le"} +``` + +The example creates `tf.train.BytesList` instances, thus it expects byte-like strings as input. However, other data types, like `tf.train.FloatList` and `tf.train.Int64List`, are also supported by the transform. + +Here is the code: + +{:.language-java} +```java +// Getting predictions is not yet available for Java. [BEAM-9501] +``` + +{:.language-py} +```py +import json + +import apache_beam as beam + +import tensorflow as tf +from tfx_bsl.beam.run_inference import RunInference +from tfx_bsl.proto import model_spec_pb2 + +def convert_json_to_tf_example(json_obj): + dict_ = json.loads(json_obj) + for name, text in dict_.items(): + value = tf.train.Feature(bytes_list=tf.train.BytesList( + value=[text.encode('utf-8')])) + feature = {name: value} + return tf.train.Example(features=tf.train.Features(feature=feature)) Review comment: No, I think it's fine. There's only one space, after the `return`. ---------------------------------------------------------------- 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 Issue Time Tracking ------------------- Worklog Id: (was: 431701) Time Spent: 9h (was: 8h 50m) > AI Platform pipeline patterns > ----------------------------- > > Key: BEAM-9421 > URL: https://issues.apache.org/jira/browse/BEAM-9421 > Project: Beam > Issue Type: Sub-task > Components: website > Reporter: Kamil Wasilewski > Assignee: Kamil Wasilewski > Priority: Major > Labels: pipeline-patterns > Time Spent: 9h > Remaining Estimate: 0h > > New pipeline patterns should be contributed to the Beam's website in order to > demonstrate how newly implemented Google Cloud AI PTransforms can be used in > pipelines. -- This message was sent by Atlassian Jira (v8.3.4#803005)