MIT6036L01d 1

okay and so then the next thing we're gonna do is think about a space of possible solutions so what is it that we're looking for ultimately we said we wanted to make predictions but what we're gonna do now is think about the learning algorithm so learning to kind of think about the learning algorithm generally speaking what we need to do is come up with a hypothesis class so a hypothesis is gonna be a guess an idea about the relationship between the X's and the Y's so a hypothesis will say hypothesis looks like some rule that would let us take an X and generate a lot I'll put my parentheses over there because I'll stick something in there in a minute so I bought this as class H it's some set of possible functions and we're interested in these little H's a little H in the set of possible ages is a function that would take an X and give me a lot right so for my particular problem my X's are gonna be an art of the D and my eyes are gonna be boolean plus 1 minus 1 in other cases they might be something else so hypothesis class generally speaking the way we think about hypothesis classes is by describing them in terms of some set of parameters and this is going to be a little too abstract but I'll make it concrete pretty soon so generally what we do is we say well here's a class of hypotheses a kind of a way of describing a relationship between X's and Y's and it has some parameters and these parameters will actually pick a particular hypothesis from the set so we'll talk about this form of a hypothesis but basically then what you want to think about it is that a hypothesis is some kind of a box and X comes in a Y comes out and we'd like to find a hypothesis that agrees in some way with our data right that does it does a good job somehow with our data and we think is going to perform well in the future