Hacker News Reader

Me too, and I was looking forward to it.

When I have time, I want to experiment with doing a sort of gradient descent on an array of different cookie recipes to find the best ratios of ingredients. That would be an interesting project.

View on HN

You would have to constrain your experiment to some definition of "cookie" to prevent it from converging on a "cookie" that is pure dark chocolate.

You'd have to get there through gradient descent, though. A pile of dark-chocolate chips is quite good, but a "cookie" that is a pile of chips baked with a small but nonzero amount of butter, flour, and baking soda... isn't. Gradient descent would have trouble with getting you through that, because cutting out the baking only makes it worse, and cutting out the other ingredients without cutting out the baking _also_ only makes it worse, and reducing the amounts of everything that isn't chocolate without bringing them to zero _still_ only makes it worse.

Instead of gradient descent, I should have said the shotgun method. I imagine giving out a few cookies of varying quantities of ingredients to neighbors and then choosing the best. Though, this method does lead to potential local maxima.

Reply on news.ycombinator.com