What sort of models work for unsupervised reinforcement learning, or is deep learning the way?

I’m setting out on an adventure to automate the statuses of the lights around my home. The lights should have different brightness in the range [0, 100] depending on some factors, which I have boiled down to:

• Time till sunset
• Sun Intensity
• Overcastness
• Possibly whether it’s a weekend or not

Now, obviously, I can implement an algorithm that computes the answer as some equation involving these metrics, but I would like it if, if over time, me manually adjusting the lights in the house because they aren’t right should affect the way the algorithm computes. As far as I see it, this is clearly reinforcement learning: when I have to change the lights manually, this is a negative reinforcement for the predictor.

I know that Deep Learning is one way of performing reinforcement learning, but traditionally, from what I understand, it needs a lot of initial data, which I don’t have. Realistically, you could expect a data-point every 10 minutes but that’s still not a lot. Obviously, in this setting, accuracy isn’t so critical, because it’s just a light, but I’m unsure about whether this is a reason to avoid Deep Learning. The problem is, other than having done a course on Deep Learning during my degree, I haven’t had any training on any classical machine learning approaches or algorithms. I’m aware of some, but don’t know enough about them to know which ones would be applicable to me.

So my question is, for reinforcement learning problems like this one, is there a good model that can be chosen that is not Deep Learning, or is it the case that the lack of data doesn’t really have an impact?