Logging in Large Math Models

At Man AHL, they are using an interesting approach to logging. They are storing inputs and outputs from their maths functions, serialised in HDF5 files. The HDF5 files are stored in a shared filesystem so they are available to all developers.

In building algo trading models, we had to log all the decisions made by the algorithm. This meant logging all the Order Books the decision was made on, as well as other inputs to the decision making process. We built custom Java code to handle this process.