Is there a reasonable way to export a generative design study as a .csv data set?
The hall of game results?
I want an exported CSV file such that there is a column for a every input and every output and each each outcome as a row. So I did this example by hand, using just 3 outcomes - but it explains what I’m looking for, turn this:
!
I can imagine that if I can output the data set in this format I can run a regression analysis and help clients understand what specific inputs are having the largest impact on the outputs. Providing this hierarchy, weighting the impact of each variable, would help clients and design teams to focus on which variable to question / investigate / refine (first, second third, etc.) to get the highest and best use of their resources. The Refinery table does a good job at describing the tradeoffs, but that kind of assumes that the significance of each variable is 1:1. I want to go further and quantify which variable(s) are best to rethink. So it’s not just optimizing a given scenario but understanding how to adjust the scenario to get closer to goal(s). This is what you get when you combine a M.Arch degree with a MBA degree.
You can actually weight the variables by using multipliers - that is if evaluation A is more important than evaluation B you an hit it with A multiplier and there will be a larger range, so slight adjustments in inputs that would otherwise impact A small amount will now be exaggerated (say 0.1 better on the metric, but multiplied by 100 becomes 10, which could be the upper limit of B). This can cause issues with the larger optimization algorithm, but worth playing with so you understand it better.
For your actual question ‘can I export a CSV’, there isn’t a need. Navigate to this path: %appdata%\GenerativeDesign
The ho-hist-results files show the way the hall of fame (what is shown in the studies) evolved over time (basically what was pushed out when). The hof-results files are the current hall of fame. The sln_set_hist-results files show the variables and outputs for each test which was run in a study, grouping them by generation. The random strings which proceed are the study GUIDs. I find that usually it’s easiest to ID which study is which by use of timestamp.
2 Likes
Yes! this is exactly what I’m looking for, thank you!