ML-15 Add pre/post processing ECG2AF image

[kilonzi]

This pull request introduces a new processing pipeline for the ECG2AF model deployment. The changes include the creation of a Docker image, implementation of data preparation and finalization scripts, and the addition of necessary dependencies. These updates enable the preprocessing of ECG data and the generation of risk scores from model predictions.

Docker Image Setup:

• Added a Dockerfile to define the environment for the processing pipeline, using a Python 3.9 slim base image, installing dependencies, and setting up the entry point for Python scripts. (model_zoo/ECG2AF/deployment/v1/processing_image/Dockerfile, model_zoo/ECG2AF/deployment/v1/processing_image/DockerfileR1-R7)

Data Preparation:

• Implemented prepare.py to process raw ECG data into normalized HDF5 tensor format, supporting various storage backends (e.g., GCS, Azure, local). (model_zoo/ECG2AF/deployment/v1/processing_image/prepare.py, model_zoo/ECG2AF/deployment/v1/processing_image/prepare.pyR1-R53)

Data Finalization:

• Added finalize.py to convert model predictions into a structured CSV format, including risk score calculations and validation of input-output consistency. (model_zoo/ECG2AF/deployment/v1/processing_image/finalize.py, model_zoo/ECG2AF/deployment/v1/processing_image/finalize.pyR1-R44)

Dependency Management:

• Introduced a requirements.txt file specifying dependencies such as pandas, numpy, h5py, and smart-open[gcs] for the pipeline. (model_zoo/ECG2AF/deployment/v1/processing_image/requirements.txt, model_zoo/ECG2AF/deployment/v1/processing_image/requirements.txtR1-R4)

[lucidtronix]

Thanks JK! Glad to get these into the repo! overtime I think we can integrate this code more fully with the rest of the ml4h library...

[lucidtronix]

This is ok for now, but since these constants are already defined in ml4h/defines.py it would be better to include ml4h and import from there. But that would require making the docker image much bigger.