MIT6036L01e 1

so if we're gonna try to do a good job on our data we said okay we have to say what's the space of possible solutions the next thing we have to do is talk about what makes one solution better when purported solution one hypothesis better than another one okay so before we talk about what makes a hypothesis better than another one we need to talk about what makes one prediction better than another one okay so let's let's start with that so there's this idea of a loss function and oh and I have to apologize not really but I'm gonna I there's some particular notation that we're using in the notes than that I'll write on the board if you read another textbook or another blog post every single textbook and blog post you read will do something mildly different so this is like a new field which is growing like crazy everybody does everything slightly differently and you just have to be aware of that so I will do things in some way but hopefully is internally consistent but beyond that I can't promise it's how every other book you read will do it okay so loss function is function it takes in elements of the Y set so in our case if we're doing classification these two arguments will be elements of the plus or minus 1 right so G is in plus 1 minus 1 a is in plus 1 minus 1 and G is I use the letter G for guests and a for actual actual and the idea is that this is how sad how sad are we how sad are we that we predicted gee when a was the true answer