Thank you @ih . I've had a go at building on this approach. I've never tried outputting a random table before.
The plot you showed has sweetness as the Y. I've replotted it with Sweetness desirability as y, which shows where best sweetness is, not just low to high. It makes sense, too much sugar is too sweet. Too little has no flavour.
I didn't give enough information in the first email. The other responses are colour, flavour and temperature as well, and I need to visualise the intersecting sweetspot of all these responses to result an "in-spec" tea. It's father's day in Australia tomorrow so I need to get this right.
By playing with the prediction profiler, it looks like if desirability >0 implies that that the response is within the lower and upper bounds that I set. The closer to the middle target, the closer it is to the target value in the range.
Based on this I applied a data filter to Sweetness Desirability/Flavour Desirability/Colour Desirability/Temperature Desirability and ommitted all values <0.1 for each response.
I then created a new column where I multiplied each of these desirabilities together to find the best results (assuming equal weighting). The output looked like this:
Do you see any risks with this approach?
@Mark_Bailey I will try this out as well. I have not used the simulate feature before.