[dees]

Here's something I think ought to be more or less straightforward but don't yet know any of the relevant machine learning theory for and so I've been putting it off. I could find a variety of handy uses for it if implemented the way I'm imagining. If it exists already, I don't know how to search for it.

It should ideally be a simple/minimal-dependencies CLI program that can take a command and arguments for that command, go ahead and run the command with given arguments, and record those arguments in relation to each other. When called with any of the command arguments missing, it will run the command filling in the blanks with a prediction for the best values for the missing arguments, based on the historical usage patterns.

My example use case is that I currently manually run a terminal command to adjust my external monitor brightness to a comfortable level at various times of day, and I would love to have a simple cron task scheduled to have it auto adjust throughout the day based on time of day and possibly other variables I can retrieve such as ambient brightness or whatever else. The dumb version would just be a predefined lighting schedule hardcoded in cron, but I specifically want to continue making adjustments as needed based on comfort and have the automation trend towards optimum over time, minimizing the frequency that I need to adjust without replacing my option to do so temporarily. The ML part I don't have the education for yet and I've been putting it off, but I can imagine a lot of other use cases for something like this implemented in a very generic way. The way I first imagined this working, I'm not entirely sure any sophisticated algorithm is even necessary, it might be doable with some high school level math run against the stored data. I presume that with such a generic approach it won't perform impressively well right out of the gate but with an accumulation of training data can be coerced into remembering what the user wants/expects gradually. As I said, I'm not educated in machine learning but I suspect this should be straightforward to do at least a crude version of given some of the libraries available.