I am trying to understand how to implement a machine learning algorithm, where the preprocessing and postprocessing is an heavy taskm inside AWS Sagemaker. The main idea is to get data from S3, each time the data change in S3, Cloud watch triggers a lambda function to invoke a SageMaker endpoint. The problem is that, once the algorithm is trained, before predicting the new data, i need to preprocess the data (custom NLP preprocessing). Once the Algorithm have done the prediction, i need to take this prediction, do a post-process and then send the post-processed data to S3. The idea i have in mind is to create a docker:
├── text_classification/ - ml scripts
| ├── app.py
| ├── config.py
| ├── data.py
| ├── models.py
| ├── predict.py - pre-processing data and post-processing data
| ├── train.py
| ├── utils.py
So i will do the pre-processing and the post-processing inside "predict.py". When i will invoke the endpoint for prediction, that script will run. Is this correct?
Take a look at using Step Functions to orchestrate the entire workflow for you.
Have the CloudWatch event trigger a Step Function that would do the following:
You can also explore Amazon SageMaker Inference Pipelines.
An inference pipeline is an Amazon SageMaker model that is composed of a linear sequence of two to five containers that process requests for inferences on data. You use an inference pipeline to define and deploy any combination of pretrained Amazon SageMaker built-in algorithms and your own custom algorithms packaged in Docker containers. You can use an inference pipeline to combine preprocessing, predictions, and post-processing data science tasks. Inference pipelines are fully managed.