README.md

Software Engineering for Machine Learning Systems

Scaffolding

This directory contains code supplied to students alongside coursework specifications, to aid structuring and building submissions.

model.py is a trival implementation of an acute kidney injury detection model that picks a result randomly. It serves to document the interface between submissions and the continuous integtation pipeline used to test submissions - specifically, the flags used to specify the location of input and output data, and the format of the predictions.

simulator.py is a simple simulator for a hospital environment. It can replay a stream of pre-generated HL7 messages over the MLLP protocol, and respond to requests to page clinicians made over a HTTP-based protocol described in the lectures. It serves to give students scaffolding to build their inference system around. You can run it with:

./simulator.py --messages=messages.mllp --mllp=8440 --pager=8441

• --messages specifies the file containing the messages to replay. The messages should be be encapsulated within MLLP, which could be generated by the generator in this repo.
• --mllp specifies the port on which to listen for MLLP connections. Each new connection is sent all messages, starting from the beginging of the supplied file, with a new message sent as soon as the current message is acknowledged. The simulator always listens on 0.0.0.0.
• --pager specifies the the port on which to listen for pager request, over HTTP. The simulator always listens on 0.0.0.0.

Aside from scaffolding to work from, the code itself attempts to:

• Provide a simple example of socket programming, to aid students who've not seen it before (which is why we don't use supporting libraries).
• Highlight common difficulties that arise in socket programming, for example, fragmented messages: --short_messages encourages the simulator to deliberately split HL7 messages across two read() calls, for example.
• Provide an example of how to implement graceful shutdown via signals.
• Provide an example of integration-style testing, in simulator_test.py.

Note that this simulator is intended as scaffolding only. This repo contains a separate simulator used to assess coursework submissions.