drawing models ✍️
My promoter models are continuous-time Markov chains. In the study, each state can either be active or inactive. As things tend to get complicated quickly, it's always a good idea to keep diagrams simple and easy to digest. For projects in the past, I've incoporated many of sukinapan's color palettes. For this occasion, their marumaru gum palette fits the bill.
Once the code was set up, drawing arbitrary models became just as easy as calling their visualise method. So, after letting a genetic algorithm search for good models, we can swiftly rummage through what it's found and marvel at their structure.
It's no surprise: if we up the edges, the legibility drops. For medium sized models, we can make do without the label background. For rather large models, it's good to avoid labels entirely. Occasionally, I also opt for a background coming from the palette to complete the immersion.
Drawing model silhouttes is as easy as removing labels and shrinking them down. It's a good idea to do this for presenting many models all at once and when their architecture, not rate parameters, is the primary concern.
...was pretty ugly - but a good start all things considered.
Look, if you're ever having a bad time - channeling the pain that is this model - know, at the very least, that it was loved.