MSML is a library for standardizing the creation of mathematical specifications as JSON objects as well as aiding in the automation of report and visualization creation from these standardized JSON.
It uses block diagram wirings and spaces to represent the actions in complex systems in line with current BlockScience research on Generalized Dynamical Systems. It also adds some enhancements to the primitive blocks to represent richer sets of behaviors.
One good example is the wiring report for the Root Finding Simulation canonical example.
To install the library, simply pip install by running "pip install math_spec_mapping"
Writing mathematical specifications can be a difficult process, especially when variable names are changed or new mechanisms are introduced. MSML seeks to streamline the process with automations as well as enhance the abilities of static math specs to deliver deeper insights. Because it is automated, one can write specifications at different levels of details or for different purposes.
- Automation: Automate writing of a specification
- Standardization: Ensure standardization across teams working to spec out a system
- Flexibility: Allow for creating views on the fly and in multiple ways depending on what stakeholders find important
- Trackability: Keep a repository of a JSON file to track changes to the spec with the same enhancements git provides for projects already
graph TD
A[JSON Object \n\n Each spec has a repo for tracking changes \n Must conform to the json specification \n Defines all aspects of the spec including blocks, spaces and actions] -->B[MSML Object \n\n JSON file is parsed, with validations and mappings along the way \n Can show different views on the fly]
B --> C[Report Outputs & Obsidian Directory \n\n Automatically build reports for the full spec or subviews \n Example: all blocks with an effect on variable XYZ\n Also builds an entire Obsidian directory of all components as linked notes]
D[Python Function Implementations \n\n Optional enhancement to actually execute code\n Done for each referenced policy option, mechanism, etc. \n Just needs a function definition for each] --> B
B --> E[Python Wirings & Simulations \n\n MSML can be used to run blocks \n Wirings automatically work to pass between domain/codomains \n Entire simulations can be built up as composed wirings]
The engineering lifecycle as defined and visualized in "Block by Block: Managing Complexity with Model-Based Systems Engineering" is depicted below.
MSML can aid in all five of these phases in different ways.
During ideation phases, users of MSML can leverage the markdown writing tool to begin organizing different thoughts into components. For example, if one were trying to model a system that has multiple currencies, i.e. USD and the Euro, those could be added in as MSML types as they are discovered. The markdown report writing supports wiki-links for use in Obsidian or a similar tool allowing users to iteratively catalog different components they find in their research and ideation.
When moving into requirements and design, MSML provides a suite of reports so that presentations to stakeholders can be insightful but tailored to the different audiences. Feedback can be iteratively incorporated into the spec with reports being re-run.
In its basic form, a spec from MSML can be used to guide implementations because blocks can be transformed into actual code/functions and spaces act as the parameterizations of those functions. There is also experimental work being done on meta-programming so that MSML could either template simulation models or even be used to hold and write code where applicable for things such as A/B testing.
Thanks to some of the more advanced features, MSML can be used as an aid for debugging and system validation. The functionality around seeing what parameters effect which blocks directly or downstream indirectly helps developers quickly identify root causes of issues. The linkages between mechanisms and what pieces of state they update allows for developers to quickly see all possible paths to variable changes there are in case something looks amiss.
The ability to fork the repository of an MSML spec as well as the ability to use it for A/B testing with the policy options makes it well suited for iterating on model evolution.
The documentation on the technical details of using the MSML can be found here
Dummy/Starter Repository Root Finding Simulation
| Feature | Dummy | Root Finding |
|---|---|---|
| Action Transmission Channels | X | X |
| Stack Block | X | X |
| Parallel Block | X | |
| Split Block | ||
| Boundary Actions | X | |
| Control Actions | X | X |
| Entities | X | |
| Mechanisms | X | X |
| Parameters | X | X |
| Policies | X | X |
| Spaces | X | X |
| State | X | X |
| Stateful Metrics | ||
| State Update Transmission Channels | X | X |
| Reports | X | X |