How to do Classification?
1.Binary Class
2.Multi-Class
According to learning beforehand,we can use regression for classification.
This is non-differentiable function.
Now,we wanna judge a pokemon beyond data collected,but we can easily recognize the turtle belong to water type.
P(turtle|water)=0?
According to describe the points on the plot,we can assume the points are sampled form a Gaussian distribution.
Beacuse we assume the Gaussian distribution,machine can imagine some pokemon beyond data to fix the lack of data.
Gaussian Distribute
L:likelihood
In my opinion:
For judging what type is the pokemon,we use the regression.(Binary Class)
1.If f(x) is closer to 1,class 1
2.If f(x) is closer to -1,class 2
If we catch a pokemon beyond the data we collected,how to process on this situation?
We assume the points are sampled from a Gaussian distribution(actually,the intial model may be any distribution with different bias and variance)
How to find the ?
On training data,making the likelihood(L) maxmimum.
Further,we can modify the model to linear boundary.===>using same covariance.
Besides,we consider spliting high dimension to many 1-dimenson.
We have to assume all the dimensions are independent as premise.