Like the individual who posted this question on the AWS forum, I am also looking to use create_monitoring_schedule. In a record_preprocessor_script file, I want to import various dependent Python scripts, also download and use .pkl files and CSVs from S3. Here's the link to the original question for reference: https://repost.aws/ja/questions/QUMXUCX9nPQWK0WdIg7e7nog/in-case-of-defaultmodelmonitor-create-monitoring-schedule-i-need-to-use-record-preprocessor-script-which-actually-imports-few-dependent-py-scripts.
In my situation, the inference endpoint uses a custom container image. The data processed by this endpoint undergoes complex preprocessing before the actual inference takes place. As a result, the data that's captured for monitoring also reflects the post-preprocessing state. This means that I need to implement similarly complex preprocessing for model monitoring.
I have explored the BYOC (Bring Your Own Container) sample codes. While I am keen on using the default model monitor code for handling data drift and model drift, apart from preprocessing, I attempted to extend the Dockerfile using the sagemaker-model-monitor-analyzer container image to see if implementation could be simplified. Unfortunately, it appears that this pre-built image is not publicly available, so I was unable to implement it. Starting from scratch to develop model and data drift monitoring in BYOC is quite challenging for me (as I have limited understanding in this area).
How might I be able to fulfill my requirements?
