UKH

Nearly all DOE training exercises seem to use baking / cookery examples. Over Christmas as I contemplated two family members attempts to modify further my homemade mincemeat, based originally on a Delia recipe, I started to wonder if any UKC scientists-cum-bakers have used DOE tools to optimise their own recipes.

I did try googling, but the afore mentioned common use of baking examples in DOE tutorials swamps the results. I found a Mini tab Christmas cookie exercise and an advert for DOE training based on bread.

I'm tempted to make it a new hobby for 2024 and try to optimise cheese scone, victoria sponge and mince pies beyond the traditional integer ratio measures. Given the complexity of mince pie recipes this might need to become a distributed problem - with climbing meets needed to exchange and sample pies.

How did you dispose of the bodies?

Made them into mincemeat.

single or multiobjective optimisation? I'm guessing multi? Flavour, consistency etc. Performance evaluation meetings to rate against the objective function. Sounds like fun, but keep the number and granularity of the variables low or you're going to be at this forever, which could also be fun. However, even a simple hill climbing generally out performs DOE. I was going to say 'theres no free lunch' in optimisation but in this case I think there is

I feel like Sheffe cubic might be required for mince pies due to the complex interplay of alcohols, dried fruits, and citrus fruits.

And as of the objective - that is where the complexity starts. I feel like there are some simple elements that can be ranked but then there is a degree of the flavour being more than the sum of it's parts that might be complex to quantify.

I think I'll start with cheese scones and go for some parameters like height, density, elasticity. 🤣

I'm not sure that subjective measures are applicable to experimental design.

Unless you're in the social 'sciences', in which case, it's more about the design of the subjective assessment methods, and the subsequent statistical analysis...

I think it depends what you optimise for.

I think with scones and cakes there are some objective outcomes. Mince pies are more complicated. I've been considering it more. I think my aim is to balance the citrus, alcohol, sweet fruit and spice so that all are present.

For example Waitrose mince pies with brandy have a lovely alcohol and vine fruit flavour but are too sweet and lack spice.

One was my mum swapping apple for gooseberry for to her brother's allergy to apple. That was acceptable.

The other is my [current] wife who has been gradually inserting herself into my domain of the kitchen. I don't know why, I never mess with her washing machine 🤣

I picked up a couple of packs of Tesco's basic mince pies, at 34p a pack. The difference between those and their premium ones was brandy & port. So I drizzled a bit of each onto the pies when warming them. Worked nicely...

Is she trying to become your currant wife...?

We considered how to optimise cheese v biscuits, for ‘best’ biscuit to use with each cheese. 6 cheeses, 6 types of biscuit, and seven C&B eaters. Full factorial with 7 replicates would give 252 data points. However, if we could put the cheeses on a scale with Norfolk Dapple on the left through to Binham Blue on the right, then maybe we could get a result sooner.

🤣

Unless your kitchen is temperature and humidity stabilised, I suspect the variance between different cooks in a domestic oven is going to seriously limit the finess of an optimisation approach.  

likewise, if you want to go for a genuinely scientific optimisation approach, you need to define a suitably high quality metric for evaluating outputs.  Given how what you taste depends on myriad human factors like how well you’ve slept, what you’ve eaten and drunk that day, your illness levels, how tired you are, how hungry you are and so on, I’d argue it’s impossible to do a time sequential optimisation.  You have to go parallel and bake all variants at once and then do a ranked comparison, or if there’s more than one orthography axis (eg taste and mouth feel) a full matrix comparison.

The absolute subjectivity of evaluation of quality of baked products seems to make them a really bad illustration for DOE.

But you will have eaten something during the single-sitting testing...

Time sequential testing has the same problem as comparing hifi items; some misguided 'super ear' audiophiles think they can remember how something sounded decades ago...

I'm glad that, as a mere engineer, my subjective/objective discriminator appears to be working correctly...

That used to be my anchor, but then I moved in to the life sciences and now the difficulty of achieving a repeatable standard haunts my days…

Master Bakers?

it’s difficult, but we’ll have to embrace a qualitative objective function 😳. Also, 10 ingredients/temperatures at 10 levels each(say) needs a constrained search space or we’ll be on it forever. Literally. Also what is the termination criteria? Pragmatic engineers would accept ‘great’ but perfectionists wouldn’t be convinced with anything but the brute force approach which would take a very long time. Also is ten levels a fine enough granularity?

Cease your murmmurations, this is the gospel*, and any deviation is heresy.

https://www.wob.com/en-gb/books/jane-grigson/english-food/9780091770433?cq_...

*Scots included due to mention of Haggis and its true origin.